Surface Texture Consistency in Object and Background Perception · 2015. 4. 17. · Surface Texture Consistency in Object and Background Perception Matthew X. Lowe Master of Arts

Surface Texture Consistency in Object and Background Perception

By

Matthew X. Lowe

A thesis submitted in conformity with the requirements

for the degree of Master of Arts

Graduate Department of Psychology

University of Toronto

© Copyright by Matthew X. Lowe 2014

II

Surface Texture Consistency in Object and Background Perception

Matthew X. Lowe

Master of Arts

Department of Psychology

University of Toronto

2014

Abstract

Seminal work on global processing has suggested that the precedence of global image features is

an inherent property of visual perception. Consequently, interference from global percepts may

influence the perception of local elements. Investigations of scene processing have been consistent

with this suggestion, demonstrating that the parahippocampal place area (PPA) represents scenes

by processing global spatial properties. Recent investigations have further revealed that PPA is

sensitive to processing non-spatial visual cues (such as surface texture) in both object and scene

perception. In the present study, we investigated potential global interference effects in object-

scene perception when attending to spatial and non-spatial visual features in both simple figure-

ground representations and more complex real-world scenes. Results revealed that non-spatial

surface properties such as texture can form a contextual link between the processing of object and

background information in scene perception, and this interactive processing proceeds from the

global to local scale of attention.

III

Acknowledgments

First and foremost, I would like to express my sincere gratitude to my advisors, Dr. Jonathan S.

Cant and Dr. Susanne Ferber, for providing encouragement, theoretical guidance, and for making

this thesis possible. I would also like to thank the members of the Ferber and Cant labs, including

but not limited to, Justin Ruppel, Kristin Wilson, and Sol Sun, for helpful discussion and support.

Additionally, I would like to express gratitude to the undergraduate research students in both the

Ferber and Cant labs. Finally, I extend a deep and heartfelt thank you to Lyndsay Jackson and

Geoffrey Hunnisett for their patience, advice, and unconditional support.

IV

Table of Contents List of Tables ................................................................................................................................................. V

List of Figures ............................................................................................................................................... VI

1. Introduction .............................................................................................................................................. 1

1.1 Global Precedence and Global Interference in Visual Perception ...................................................... 1

1.2 The Representation of Scene and Texture Perception ....................................................................... 2

1.3 Texture and Object Perception ........................................................................................................... 3

1.4 Current Study and Predictions ............................................................................................................ 4

2. Experiment 1 ............................................................................................................................................. 6

2.1 Methods .............................................................................................................................................. 7

2.2 Results ................................................................................................................................................. 9

2.3 Discussion .......................................................................................................................................... 12

3. Experiment 2 ........................................................................................................................................... 14

3.1 Methods ............................................................................................................................................ 16

3.2 Results ............................................................................................................................................... 17

3.3 Discussion .......................................................................................................................................... 19

4. Experiment 3 ........................................................................................................................................... 21

4.1 Methods ............................................................................................................................................ 22

4.2 Results ............................................................................................................................................... 24

4.3 Discussion .......................................................................................................................................... 26

5. General Discussion .................................................................................................................................. 28

5.1 Conclusions ....................................................................................................................................... 31

References .................................................................................................................................................. 32

Supplementary Materials............................................................................................................................ 36

Supplementary References ......................................................................................................................... 40

V

List of Tables

Table 1. Average accuracy and standard error for each condition across all experiments

VI

List of Figures

Figure 1. Examples of the stimuli used in Experiment 1.

Figure 2. Results of Experiment 1.

Figure 3. Exampled of the stimuli used in Experiment 2.




1

1. Introduction

A remarkable aspect of the human visual system is the ability to draw on a broad range of

cues to rapidly and efficiently identify and categorize objects embedded in a complex visual

scene. Knowledge about which objects and settings tend to co-occur facilitates the efficiency of

both the search for and recognition of objects (Hock et al., 1974; Palmer, 1975; Biederman et al.,

1982; De Graef et al., 1990; Boyce & Pollatsek, 1992; Henderson et al., 1999; Davenport &

Potter, 2004; Gordon, 2004; Joubert et al., 2007). In general, it has been found that objects which

appear within a consistent setting (e.g., a loaf of bread on a kitchen counter) are found to be

processed more quickly and accurately than those appearing within an inconsistent setting (e.g., a

mailbox on a kitchen counter; for a review, see Oliva & Torralba, 2007). While the majority of

this previous research has examined high-level semantic relations between an object and its

background, visual similarity (or dissimilarity) in object-scene contextual associations is also

important. Thus, the purpose of this paper is to examine the extent to which object-scene

consistency in spatial (i.e., form) and non-spatial (i.e., texture) visual features influences object

perception within a scene.

1.1 Global Precedence and Global Interference in Visual Perception

Scene perception may be governed by general mechanisms that apply across a range of

different types of visual processing. For example, seminal work on global processing has

suggested that the precedence of global image features is an inherent property of visual

perception, wherein the perception of global structure precedes the perception of local elements

or fine-grained analyses (Navon, 1977). Navon presented compound letters representing larger

figures (global configurations), which were spatially constructed from a suitable arrangement of

2

smaller figures (local elements), and observed an advantage in the processing of global

configurations over local elements (i.e., faster responses to global configurations compared with

local elements). These findings were interpreted as supporting the notion of global precedence in

the spatial processing of form, and were referred to as the ‘global precedence effect’. Moreover,

when global configurations and local elements were inconsistent, responses to the local elements

were subject to interference from the global configurations, but local features did not interference

with global perception. This result was subsequently referred to as the ‘global interference

effect’. In other words, involuntary attention to the global level was observed when attention was

directed to the local level, resulting in global inference in the perception of local elements.

The global precedence hypothesis has since been validated in numerous studies (for a

review, see Kimchi, 1992) and subsequent research on rapid scene identification has provided

support for the primacy of global features over local region and object information. Specifically,

is has been demonstrated that contextual information influencing object-scene interactivity is

guided by global image features which direct attention early in the visual processing stream

(Torralba et al., 2006). Moreover, it has been shown that human observers can accurately

identify natural scene categories based on global scene structure without first having to recognize

local region or object information (Greene & Oliva, 2009). Thus, Greene and Oliva suggested

that scene perception is based on an initial scene-centered visual representation containing global

percepts, which captures much of the variance in scene structure, constancy, and function in

natural scene categories.

1.2 The Representation of Scene and Texture Perception

Consistent with these findings, investigations of scene processing using functional

magnetic resonance imaging (fMRI) have demonstrated that the scene-selective

3

parahippocampal place area (PPA), a region shown to respond selectively to scenes over

individual objects or faces (Epstein & Kanwisher, 1998), represents scenes by processing global

spatial form and structure (Epstein et al., 2003). While much of the early neuroimaging research

on scene perception focused on the role of spatial cues in the visual processing of scenes, recent

investigations into the neural representation of both scenes and objects have revealed that a

region of the collateral sulcus (CoS) overlapping with PPA is sensitive to processing non-spatial

visual cues such as surface texture (Cant & Goodale, 2007, 2011; Cant & Xu, 2012). Indeed,

behavioural research examining the contribution of global percepts to scene identification are

consistent with these findings, and draw attention to the role of both spatial (e.g., form) and non-

spatial (e.g., texture) features in capturing the diagnostic structure of an image in order to obtain

the overall gist of a scene (Oliva & Torralba, 2006). In other words, these visual cues enable

rapid scene identification and categorization necessary for the efficient processing of our

complex environment. In fact, evidence from computational modelling has demonstrated that

texture may be sufficient on its own in providing the means necessary for scene recognition

(Renninger & Malik, 2004). Thus, in addition to the known role of spatial cues in scene

perception, a growing body of work using behavioural psychophysics, computational modelling,

and functional neuroimaging has revealed the importance of non-spatial cues (i.e., surface

texture) in scene perception and recognition.

1.3 Texture and Object Perception

In addition to scene perception and identification, it has been argued that surface

properties (and the material properties that they signal) play a critical role in object recognition

and in how we interact with the world. Specifically, highly diagnostic visual cues such as surface

reflectance properties, surface texture, and surface structure can cue stored knowledge of object

4

material properties such as mass, compliance, and friction (Adelson, 2001; Motoyoshi et al.,

2007; Buckingham et al., 2009). These cues not only help us to recognize objects in our

environment, but also contribute to action planning (Gallivan et al., 2014), ultimately affecting

how we physically engage with objects of various tactile qualities (e.g., rough vs. smooth) and

how we adjust our gait when moving through an environment containing different surface

attributes (e.g., ice vs. grass). Moreover, these cues may be especially important in defining edge

and contour information used for finding partially occluded objects in complex and crowded

environments (Biederman, 1988). Finally, recent evidence has demonstrated independent

processing of form, colour, and texture in object perception (Cant et al., 2008), suggesting that,

in addition to the well-known role of shape in object processing (e.g., Biederman, 1987), surface

properties such as colour and texture also play an important role in object perception. In

summary, recent evidence has revealed that non-spatial surface cues such as texture play an

important role in both object and scene perception, but these studies have only looked at either

object or scene perception separately. While it has been argued that initial global scene image

statistics form a context in which objects can be rapidly and efficiently located and classified at a

later stage of visual processing (Schyns & Oliva, 1994), the influence of visual texture in the

interaction between an object and its background has yet to be explored, despite the importance

of texture as a cue in both object and scene processing.

1.4 Current Study and Predictions

Based on previous work revealing the precedence of global image features in visual

perception (i.e., Navon, 1977), in the present paper we aim to initially establish a global

interference effect of form and surface texture during the perception of local elements in figure-

ground representations (i.e., modified Navon stimuli, see Figures 1 and 3), before exploring the

5

role of these global visual cues in the interaction between an object and its background within a

scene (see Figure 5). Specifically, we first attempt to replicate the global interference effect in

form perception (i.e., slower judgments of local form when local and global form is visually

inconsistent, but not vice versa) using modified Navon stimuli, and then investigate, for the first

time, whether there is also a global interference effect in non-spatial processing (i.e., texture

perception). We then examine the processing of these visual features using more realistic scene

stimuli in order to determine if global interference effects (particularly for texture) generalize to

interactions between object and scene perception. If texture is indeed important as an identifying

cue in scene and object perception, and has been shown to activate regions of the brain

overlapping with those involved in scene processing (Cant & Goodale, 2007), we expect to

observe global precedence of texture in early-stage scene processing (i.e., global scene

properties) which would subsequently influence the perception of local object properties through

global interference. This finding would not only build on the existing body of literature

examining the role of global image information as a contextual cue in the perception and

recognition of objects within our environment, but would further our understanding of the

mechanism through which such contextual cuing occurs.

In Experiment 1, we measured the extent to which the perception of local form and

texture is influenced by global feature (i.e., form or texture) consistency when both features are

integrated within a single global contour (see Figure 1). In Experiment 2, we separated local and

global texture into foreground and background elements, respectively, in order to identify how

perceptual separation influences global- and local-directed attention (see Figure 3). To assess the

contextual influence of texture (and form) consistency in object perception within a scene, we

introduced a single object within a 3-D scene in Experiment 3 (see Figure 5). In Experiments 1

6

and 2, we predicted an initial replication of Navon’s (1977) seminal work (i.e., a global

interference effect in local form perception) using modified Navon stimuli, coupled with a

consistent interference of global surface texture in local texture perception. Finally, in

Experiment 3 we predicted the facilitation of object perception when object-scene visual cues

(both form and texture) were consistent. In all Experiments we also examined the global

precedence effect (i.e., faster overall responses when attending to global configurations

compared with local elements), but place more emphasis on the global interference effect since

we show that these interference effects are more informative with regard to the interaction

between global and local perception.

2. Experiment 1

In this experiment, modified Navon figures (Navon, 1977) were adapted to integrate both

form and texture within a single global contour (see Figure 1). Form classifications were

included in this paradigm as a control, enabling the suitability of these adapted Navon figures to

be examined, with the prediction of a global interference effect in the processing of form (slower

judgments of local form when local and global form is inconsistent, but not vice versa). These

findings would replicate data from Navon’s seminal work, and would thus validate the use of our

stimuli and experimental parameters in the investigation of the role of texture consistency in

global and local perception. Participants were cued to make speeded classifications of form or

texture at either a global or local scope of attention, in conditions where global and local features

(i.e., form or texture) were either consistent or inconsistent.

7

2.1 Methods

Participants

Eleven participants (seven females) aged between 20 and 27 years of age (M = 21.81)

were recruited from the University of Toronto undergraduate community and received course

credit for their participation. All participants had normal or corrected-to-normal visual acuity,

were right-handed, and gave informed consent in accordance with the University of Toronto

Ethics Review Board.

Figure 1. Examples of the stimuli used in Experiment 1. The

stimuli could vary along two features (form and texture),

two levels of consistency (consistent and inconsistent) and

two scopes of attention (global and local).

8

Stimuli and Apparatus

Sixty-four stimuli were generated using Adobe Photoshop CS3 software (Adobe

Systems, San Jose, CA) and presented electronically using the E-Prime 2.0 software (Psychology

Software Tools, Pittsburgh, PA) on a ViewSonic 21-inch CRT monitor (1,280 x 1,024

resolution; 85-Hz refresh rate). Stimuli subtended 18.4° x 17.3° of visual angle and were

presented centrally against a white background following a black central fixation cross

(subtending 1° x 1°) at a viewing distance of 52 cm. The stimuli were constructed so that visual

features (form: heart versus star; texture: paint versus rock) could vary at both global and local

levels of attention, and importantly, variations in each feature were manipulated across levels

(consistent: similar global and local shape, similar global and local texture; inconsistent:

different global and local shape, different global and local texture). Variations in each visual

feature were matched at both global and local levels (i.e., 32 instances of global star, 32 instances

of global heart, 32 instances of global paint, 32 instances of global rock, etc.). In order to avoid

responses based on any one local element, local texture elements from the same category (i.e.,

paint or rock) were heterogeneous in nature and the locations and orientations of both local form

and texture elements were jittered across stimuli. To ensure that observer classification across

texture categories was independent of visually distinctive colour cues, the chromaticity of each

stimulus’ texture was calculated using Matlab software (MathWorks, Natick, MA) and modified

through level adjustment using Adobe Photoshop CS3 (see Supplementary Materials and

Supplementary Figure 1 for more details). Mean luminance for all stimuli averaged 239.36 (on a

0-255 luminance scale) with a standard deviation of 2.49.

9

Design and Procedure

The experiment contained four blocks of trials representing conditions of form and

texture perception at both a global and local scope of attention (i.e., global form; global texture;

local form; local texture). Before beginning the experiment, participants were given five practice

trials per condition to become familiar with the task. Each block began with an instruction to

attend to either “shape” or “texture” at either global- or local-levels of attention (instructed as

either “large” or “small”, respectively). Following an initial key press, each trial began with a

central fixation cross displayed for 2000 ms, after which the stimulus was presented and

remained on screen until response. Participants were instructed to make a speeded classification

(shape: heart or star; texture: paint or rock) after the onset of the stimulus using either the “1” or

“2” keys on the number pad of a computer keyboard, which would then terminate the trial. For

the experiment proper, each block contained sixty-four randomly presented stimuli with equal

numbers of consistent and inconsistent trials. The order of presentation of the four blocks was

counterbalanced across participants. Each block was separated by an instruction screen

informing participants that they may take a short break, and reminding them to respond as

quickly and accurately as possible in the following block. Accuracy and response latency was

recorded for each trial.

2.2 Results

As participants performed well across all conditions (see Table 1 for the average

accuracies in each condition), we focused our analysis on response latencies (for correct trials

only). An initial outlier analysis was performed separately on each participant, and response

latencies 2.5 standard deviations above or below the mean reaction time for each condition were

excluded from the analysis. An outlier analysis was not conducted on accuracy measures, but

10

participants who performed below chance (50 %) in any condition were excluded from analysis.

Response latencies were analyzed using a three way repeated measures analysis of variance

(alpha = 0.05), with scope of attention (global vs. local), feature (form vs. texture), and

consistency (consistent vs. inconsistent) as factors. Pairwise post-hoc comparisons were

performed using the Bonferroni-Holm procedure to correct for multiple comparisons (alpha =

0.05). This analysis procedure was performed in all three experiments.

Table 1. Average accuracy and standard error for each condition across all experiments

Condition Experiment 1 Experiment 2 Experiment 3

Global Form 95.31 (SE = 1.09) 96.48 (SE = 0.75) 96.93 (SE = 1.05)

Local Form 96.31 (SE = 1.12) 95.44 (SE = 1.10) 97.33 (SE = 1.09)

Global Texture 92.61 (SE = 1.43) 94.01 (SE = 0.81) 96.04 (SE = 1.74)

Local Texture 92.61 (SE = 1.63) 93.75 (SE = 1.00) 96.04 (SE = 1.28)

Mean response latencies can be seen in Figure 2 as a function of scope of attention,

feature, and consistency. Significant effects were found for the main effects of feature [form M =

556.7 ms; texture M = 700.9 ms; F(1,10) = 98.7, p < .001], consistency [consistent M = 606.5

ms; inconsistent M = 651.1 ms; F(1,10) = 42.7, p < .001], the feature-by-consistency interaction

[F(1,10) = 36.6, p < .001], and the three-way interaction between feature, consistency, and scope

of attention [F(1,10) = 8.57, p < .05]. To evaluate these significant effects in greater detail,

planned pairwise comparisons of consistent versus inconsistent trials were made for each of the

four conditions of interest (i.e., global form, local form, global texture, and local texture).

Significant differences between the mean latency of consistent and inconsistent trials were found

for local form [consistent M = 576.0 ms; inconsistent M = 628.5 ms; t(10) = 5.73, p < .001], local

11

texture [consistent M = 633.2 ms; inconsistent M = 670.2 ms; t(10) = 3.02, p < .05], and global

texture [consistent M = 712.1 ms; inconsistent M = 787.9 ms; t(10) = 4.14, p < .005), but not for

global form [consistent M = 504.6 ms; inconsistent M = 517.7 ms; t(10) = 1.7, p = .06] .

Importantly, when the consistency of the unattended feature was held constant (e.g., examining

differences between consistent and inconsistent trials when attending to global form, and holding

changes in texture constant), these effects were maintained across all three significant conditions

[local form: t(10) = 2.71, p < .05; local texture: t(10) = 2.68, p < .05; and global texture: t(10) =

2.97, p < .05], indicating that significant differences in mean latency between consistent and

inconsistent trials were driven by changes of consistency in the attended feature, independent of

changes in the unattended feature. Finally, significant differences between the mean latency of

global and local scope of attention were found for both form [global form M = 511.2 ms; local

form M = 602.3 ms; t(10) = 6.67, p < .001] and texture [global texture M = 750.0 ms; local

texture M = 651.7 ms; t(10) = 3.88, p < .002].

12

2.3 Discussion

As predicted, we found a global precedence effect coupled with a global interference

effect for form perception using our modified Navon stimuli. Specifically, participants were

faster to categorize global form compared with local form (global precedence effect), and were

Figure 2. Results of Experiment 1 for each condition (global form;

local form; global texture; local texture). Light bars represent

consistent global and local features, and dark bars represent

inconsistent global and local features. All statistical comparisons

are between the consistent and inconsistent conditions of each

attended feature. Results are based on data from 11

participants, in a repeated-measures design. Error bars indicate

95% confidence intervals derived using the mean square error

term. *p < 0.05, **p < 0.01, ***p < 0.001

13

slower to categorize local form when local and global form were inconsistent, but not vice versa

(global interference effect). This replicates previous research using classic Navon stimuli

(Navon, 1977), and adds to a large body of literature suggesting global primacy in the perception

of form (for a review, see Kimchi, 1992). But importantly, we also demonstrated, for the first

time, a global interference effect in texture perception, as judgments of local texture were slower

when local and global textures were inconsistent. Together, the global interference effects for

form and texture found here strengthen the notion that both spatial and non-spatial visual cues

play important roles in different types of visual perception (e.g., figure/ground segregation, or

the interactions between object and scene perception).

Interestingly, participants were also slower to classify global texture when global and

local textures were inconsistent (i.e., a local interference effect), indicating a bidirectional

influence of texture perception across global and local levels of attention. While this finding was

unexpected, previous research has demonstrated that the interaction of global and local

configurations is influenced by the overall quality and relative visibility of information at each

level (Hoffman, 1980). Specifically, by manipulating the quality and visibility of form at either

the global or local level, global or local precedence effects in visual perception can be obtained

depending on which level was distorted. With this in mind, it is possible that visual ambiguity in

global texture resulting from unclear boundaries between global and local textures may have

caused degradation of the global percept, and consequently resulted in the local interference

effect. In other words, participants may have found it more challenging to classify global texture

when local texture was inconsistent, because global and local texture were fairly integrated in

perceptual space (i.e., the distinction between global and local texture was not readily apparent;

see Figure 1). We did not find the same result when participants attended to global form, likely

14

because global and local forms were more separated in perceptual space (i.e., the distinction

between the global outline contour and the local outline contours was readily apparent).

Consistent with these suggestions, mean response latencies were considerably longer in

the global texture condition compared with the local texture condition (collapsed across levels of

consistency), indicating a local precedence effect in texture perception. As such, it is perhaps not

surprising that we observed local interference of texture in this Experiment. We did not observe

this same relationship for the comparison of global and local form (again, collapsed across levels

of consistency), as response latencies in the local form condition were longer than those in the

global form condition, which replicates Navon’s (1977) previous findings. Slower judgments of

global compared with local texture would not be predicted based on Navon’s previous work (nor

would the local interference effect described previously), and, as we reasoned above, these

findings might have arisen because keeping global texture contained within the boundaries of the

global outline contour made it more difficult to perceptually separate global and local texture.

We investigated this possibility in Experiment 2.

3. Experiment 2

The results of Experiment 1 replicated Navon (1977) when attending to form: response

latencies for judgments of local form were slower when global and local forms were

inconsistent, but not vice versa (i.e., a global but not a local interference effect in form

perception), and we observed longer overall responses for local compared with global form

perception (i.e., a global precedence effect). We found a global interference effect in texture

perception, but we also observed an influence of local texture when attending to global texture,

15

which we did not predict. In order to ascertain whether these findings in texture perception were

being driven by visual ambiguity induced by integration of global and local texture cues within a

global contour, the current experiment used figure-ground stimuli that separated local and global

texture into foreground and background elements, respectively (see Figure 3). This enabled a

clear perceptual division between local and global texture, consistent with the perceptual

separation between local and global form. We predicted that removing visual ambiguity between

local and global texture would eliminate the local interference effect in texture perception while

maintaining the global interference effect. Moreover, we predicted that removing this visual

ambiguity would significantly reduce the difference in response latency between global and local

texture perception observed in Experiment 1, thus eliminating the local precedence effect. If

these predictions hold, then they would reveal that global interference effects in visual perception

are not dependent on the global precedence effect, which would suggest that the dominance of

global features in visual perception is not a result of differences in the difficulty of

discriminating global versus local features.

16

3.1 Methods

Participants

Thirteen new participants (all female) aged between 20 and 32 years of age (M = 21.39)






The stimuli could vary along two features (form and

texture), two levels of consistency (consistent and

inconsistent) and two scopes of attention (global and

local).

17

Apparatus, Stimuli, and Procedure

The apparatus used in this experiment was identical to the apparatus used in Experiment

1. Sixty-four new stimuli matching the dimensions and conditions of Experiment 1 were

generated. Local and global form maintained separate boundaries, and local and global texture

was separated into foreground and background elements, respectively (see Figure 3). Mean

luminance for all stimuli averaged 202.10 (on a 0-255 luminance scale) with a standard deviation

of 9.49. The experimental design and procedure were identical to Experiment 1.

3.2 Results

One participant was excluded from analysis due to incomplete data collection because of

time constraints. Thus, the analysis was performed on the remaining twelve participants.

As with Experiment 1, the overall accuracy was high (see Table 1 for the average

accuracies in each condition). Mean response latencies can be seen in Figure 4 as a function of

scope of attention, feature, and consistency. Significant main effects were found for feature

[form M = 541.4 ms; texture M = 571.6 ms; F(1,11) = 12.40, p < .01) and consistency [consistent

M = 549.0 ms; inconsistent M = 564.0 ms; F(1,11) = 14.65, p < .005), and these effects were

qualified by a significant feature-by-consistency interaction [F(1,11) = 5.06, p < .05]. Further

planned pairwise comparisons between consistent and inconsistent trials for each of the four

conditions of interest were performed using the Bonferonni-Holm correction for multiple

comparisons. These comparisons revealed significant differences in response latency for the

local form [consistent M = 527.6 ms; inconsistent M = 557.4 ms; t(11) = 2.82, p < .05] and local

texture conditions [consistent M = 565.1 ms; inconsistent M = 585.9 ms; t(11) = 2.53, p < .05],

18

but not for the global form [consistent M = 534.9 ms; inconsistent M = 545.8 ms; t(11) = 1.53, p

= .16] and global texture conditions [consistent M = 568.5 ms; inconsistent M = 567.1 ms; t(11)

= .15, p = .44] . As with Experiment 1, when the consistency of the unattended feature was held

constant, significant effects were maintained across both local form and local texture conditions

[t(11) = 2.49, p < .05, and t(11) = 2.39, p < .05, respectively], again indicating that differences in

mean latency between consistent and inconsistent trials were driven by changes of consistency in

the attended feature, independent of changes in the unattended feature. Finally, consistent with

our predictions, the main effect of scope of attention was not significant [F(1,11) = .55, p = .47],

and no significant differences between the mean latency of global and local scope of attention

were found for either feature [global form M = 540.3 ms; local form M = 542.5 ms; t(11) = .114,

p = .46; global texture M = 567.8 ms; local texture M = 575.5 ms; t(11) = .675, p = .51].

19

3.3 Discussion

The results of Experiment 2 replicate those from Experiment 1, providing clear evidence

for the primacy of global percepts of form and texture (compared with local form and texture) as

Figure 4. Results of Experiment 2 for each condition (global form;

local form; global texture; local texture). Light bars represent

consistent global and local features, and dark bars represent

inconsistent global and local features. All statistical comparisons

are between the consistent and inconsistent conditions of each

attended feature. Results are based on data from 12

participants, in a repeated-measures design. Error bars indicate

95% confidence intervals derived using the mean square error

term. *p < 0.05

20

measured through global interference: we found slower response latencies for classifications of

local features when local and global features were inconsistent, but not vice-versa. As predicted,

separating global and local texture into visually distinctive elements removed visual ambiguity,

and consequently eliminated local texture interference. Interestingly, separating global and local

texture significantly reduced the difference in response latency between global and local

conditions in both form and texture processing observed in Experiment 1, yet a global

interference effect was maintained for both features. That is to say, global interference for both

form and texture was observed despite no significant overall differences in response latencies

between global and local attention conditions for both features. This indicates that interference

from the global percept is not dependent on slower response latencies for local compared with

global elements. Finally, these results contribute a novel and important finding regarding the

primacy of global over local surface-texture features in figure-ground perception, independent of

form. This last finding is consistent with the independence of form and texture in the perception

of single objects (Cant et al., 2008).

Since the stimuli in Experiment 2 were figure-ground images with a clear separation

between foreground and background elements, the finding of a global interference in form and

texture perception may also apply to the interaction between the processing of an object and its

background in situations of more natural scene perception. We investigated this possibility in

Experiment 3.

21

4. Experiment 3

In both previous experiments, we replicated Navon’s (1977) findings of a global

interference effect in form perception, revealing significant effects of consistency when attending

to local but not global form. Moreover, the results of Experiments 1 and 2 revealed a novel

finding of a global interference effect in texture perception. Since the previous experiments

utilized simple figure-ground representations, we wondered if the dominance of global visual

features in form and texture perception would extend to more complex, real-world stimuli. To

investigate this possibility, in Experiment 3 we examined whether global form and texture

features could influence the processing of a single object placed within an indoor scene (see

Figure 5). If primacy of global scene features is observed when participants attend to local

features, we would expect participants to make faster judgments of object (foreground) form and

texture when scene (background) form and texture are consistent, compared with conditions

where scene and object features are inconsistent.

22

4.1 Methods

Participants

Ten new participants (six females) aged between 18 and 21 years of age (M = 19.80)





Figure 5. Examples of the stimuli used in Experiment 3. The stimuli could vary along two

features (form and texture), two levels of consistency (consistent and inconsistent) and two

scopes of attention (scene and object).

23

Apparatus, Stimuli and Procedure

The apparatus used in this experiment was identical to the apparatus used in Experiments

1 and 2. Blender 2.0 software (Stichting Blender foundation, Amsterdam) was used to render 3-

dimensional indoor environments and generate stimuli. One-hundred and twenty eight new

stimuli, each subtending 33.45° x 21.28° in visual angle, were created and during the experiment

each was presented centrally against a white background following a black central fixation cross

(subtending 1° x 1°) at a 52 cm viewing distance. To maintain consistency with the previous

experiments, object texture was counterbalanced to contain equal representations of homogenous

stimuli (object-scene textures were selected from the same source image) and heterogeneous

stimuli (object-scene textures were selected from different source images from the same texture

category; i.e., paint or rock). Stimuli were rendered using a constant view-point with consistent

particle system lighting across feature conditions in order to maintain overall consistency in

surface area, perspective, and reflectance. The stimuli were created to have variations in visual

features (form: square versus triangle; texture: paint versus rock) at both a global (scene) and

local (object) processing level, and contained either consistent or inconsistent features across

levels. Variations in each visual feature were matched at both global and local levels (i.e., 64

instances of global square, 64 instances of global triangle, 64 instances of global paint, 64

instances of global rock, etc.). Mean luminance for all stimuli averaged 140.49 (on a 0-255

luminance scale) with a standard deviation of 28.57. The experimental design and procedure

were identical to previous experiments, except for the fact that more trials were conducted in this

experiment.

24

4.2 Results

In order to eliminate any potential difference in response latencies driven by differences

in the luminance of the images across conditions, twenty-seven stimuli were removed prior to

data analysis following a luminance outlier analysis (see Supplementary Material and

Supplementary Figure 2 for details). Mean luminance for all remaining stimuli averaged 147.90

(on a 0-255 luminance scale) with a standard deviation of 20.85.

As we observed in Experiments 1 and 2, participants made very few errors overall (see

Table 1 for the average accuracies in each condition). Mean response latencies can be seen in

Figure 6 as a function of scope of attention, feature, and consistency. A significant main effect of

feature was found [form M = 477.7 ms; texture M = 566.7 ms; F(1,9) = 24.47, p < 0.005], as well

as a significant three-way interaction between scope of attention, feature, and consistency [F(1,9)

= 19.57, p < 0.005]. Similar to previous experiments, planned pairwise comparisons for each of

the four conditions of interest were performed using the Bonferonni-Holm correction for

multiple comparisons. Significant differences in mean response latency between consistent and

inconsistent trials were found for local (object) texture [consistent M = 572.8 ms; inconsistent M

= 603.3 ms; t(9) = 3.69, p < .05] and global (scene) form conditions [consistent M = 465.2 ms;

inconsistent M = 484.9 ms; t(9) = 3.16, p < .05], but not for local form [consistent M = 480.6 ms;

inconsistent M = 480.2 ms; t(9) = .55, p = .48] and global texture conditions [consistent M =

549.4 ms; inconsistent M = 541.3 ms; t(9) = 1.08, p = .30]. As with previous experiments, when

the consistency of the unattended feature was held constant, the significant effects of both local

texture [t(9) = 2.48, p < .05] and global form [t(9) = 1.93, p < .05] were maintained. Lastly,

significant differences between the mean latency of global and local scope of attention (collapsed

across levels of consistency) were found for texture [global texture M = 545.4 ms; local texture

25

M = 588.1 ms; t(9) = 2.72, p < .05] but not form [global form M = 475.0 ms; local form M =

480.4 ms; t(10) = .376, p = .36].

Figure 6. Results of Experiment 3 for each condition (scene form;

object form; scene texture; object texture). Light bars represent

consistent scene and object features, and dark bars represent

inconsistent object and scene features. All statistical

comparisons are between the consistent and inconsistent

conditions of each attended feature. Results are based on data

from 10 participants, in a repeated-measures design. Error bars

indicate 95% confidence intervals derived using the mean

square error term. *p < 0.05

26

4.3 Discussion

Results of the present experiment demonstrate significantly lower mean response

latencies for judgements of object texture when the background scene texture was consistent,

compared with inconsistent. In addition, we observed longer response latencies when attending

to object texture compared to scene texture, consistent with Navon’s (1977) global precedence

hypothesis. These results replicate and extend the results of a global interference effect in texture

perception observed in Experiments 1 and 2, and indicate that global scene texture may form a

contextual cue in influencing object perception and recognition, through the mechanism of

primacy of global scene features. Moreover, global interference was established through the use

of both homogenous and heterogeneous local elements, indicating that it is not simply the same

low-level visual elements constituting texture which drive contextual associations and global

primacy (as evidenced through homogenous consistency), but also knowledge of their

corresponding properties and categorical identity through visual similarity (as evidenced through

heterogeneous consistency).

Surprisingly, although no difference in response latencies were observed between local

and global form (i.e., no global precedence effect), we observed interference of object form on

classifications of scene form, but not vice-versa (i.e., a local, or object interference effect). At

first, this evidence may seem difficult to reconcile with numerous studies advocating the primacy

of global information. However, these findings are not altogether unprecedented. In fact,

previous research has demonstrated a deleterious effect of salient objects on scene identification,

particularly when an inconsistent object is present (Joubert et al., 2007). Joubert and colleagues

theorized that such an effect could be explained by an exogenous capture of attention involving a

bottom-up processing bias, resulting in slowing the processing of background scene information.

27

These results suggest that object and scene context could be processed in parallel and interact

extensively. Consistent with these results, Gordon (2004) observed preferential attention to

inconsistent objects within approximately 150 ms of scene onset, suggesting that information

about the semantic relationship between objects and scenes is extracted rapidly. While it is

certainly possible that a global interference effect for form was not obtained in Experiment 3 due

to these salient object properties, it is also possible that this relationship does not exist in more

natural scene perception, where scene structure is unlikely to influence the perception of object

form. Future studies will need to explore these possibilities in greater detail.

Interestingly, in the present experiment no such influence of salient object properties was

found to affect scene texture classification, indicating both an independent and feature-specific

effect of object saliency on scene perception (i.e., affecting scene shape but not scene texture

perception), and an asymmetrical interference effect in texture and form processing. These

results extend an existing body of literature demonstrating both independent processing and

asymmetric interference of form and texture in object perception (Cant & Goodale, 2007, 2009;

Cant et al., 2008). Finally, these results demonstrate that non-spatial scene features (i.e., texture)

influence the processing of object texture, and spatial object features (i.e., form) influence the

processing of scene form. This suggests that there is an asymmetry in how spatial and non-spatial

visual features are utilized in object-scene interactions, but future studies should investigate the

validity of this possibility in greater detail.

28

5. General Discussion

The results of the present set of experiments clearly demonstrate a global interference

effect in texture perception, and provide novel evidence that the primacy of global scene texture

represents a contextual cue through which the perception of object properties is influenced.

Across all three experiments, local feature-based texture judgments were facilitated when their

global counterpart was consistent. Furthermore, the results of Experiment 3 provide the first

evidence that the perception of object texture is affected by the global scene percept, implicating

scene texture as an important contextual cue in object perception. These results are consistent

with our initial predictions based on evidence for the interference of global configurations on

local features in visual perception (e.g., Navon, 1977), the overlapping neural representations of

surface texture and scene processing in parahippocampal cortex (e.g., Cant & Goodale, 2007),

and the contextual influence of scene consistency in facilitating object perception (e.g.,

Davenport & Potter, 2004). Furthermore, the data provided in the current paper are consistent

with previous investigations, as the results of Experiments 1 and 2 replicate Navon’s seminal

work on the global interference effect in the perception of form, verifying that our stimuli and

experimental paradigm were valid in investigating a global interference effect in surface-texture

perception using modified Navon stimuli. We would argue that these global interference effects

(both form and texture) are not simply a result of local processing being more difficult than

global processing, as participants’ behavioural performance was equated across global and local

feature conditions for all three experiments (see Table 1). This argument is further supported by

the results of Experiment 2, which demonstrated global interference effects for both form and

texture in the absence of response latency differences between local and global conditions (i.e.,

in the absence of global precedence effects). Rather, we suggest the global interference effects

29

we observed reflect a true processing difference between global and local levels, where global

features are visually dominant and thus influence local features. Of course, there are

opportunities for local elements to influence global perception, and we discuss this in more detail

below. Finally, across all three experiments, we observed significant effects of consistency in the

attended feature independent of changes in the unattended feature, demonstrating independent

processing of form and surface texture in visual perception consistent with previous research

(Cant & Goodale, 2007, 2009; Cant et al., 2008). In addition, significant differences in response

latency were found between form and texture (i.e., form was processed faster than texture) in all

three experiments. These results are consistent with previous research demonstrating that

judgements of object form are generally processed faster than judgements of object texture (e.g.,

Cant & Goodale, 2008).

While our results are largely in support of an overwhelming amount of evidence

demonstrating the primacy of global scene information over local elements or object information

(for a discussion, see Introduction), the argument could be made that the interaction between

global and local levels of attention is dependent on the relative visibility or quality of

information at each level (Hoffman, 1980; Lagasse, 1993). Despite the elimination of local

interference on global perception through our stimulus manipulations in Experiment 2, it seems

plausible that in certain cases local object information may contextually influence the processing

of global scene features through the exogenous capturing of attention. Indeed, in addition to our

reported results of local object interference in feature-based attention to global scene form in

Experiment 3, previous research has demonstrated that identity-defining properties of salient

objects can influence the perception of their respective scenes (Davenport & Potter, 2004;

Joubert et al., 2007). Although these findings are certainly noteworthy in delineating the

30

limitations in which the precedence of the global percept occurs, they do not depreciate the

results of the vast majority of previous research or the current series of experiments. Here, we

provide strong evidence that is in-line with current theories surrounding the hierarchy of global

and local percepts in object and scene perception (Greene & Oliva, 2009). That is to say, the

results of the present study provide support for an initial scene-centered visual representation that

is formed from global image features (e.g., texture), which subsequently influences the

perception of later-stage local object properties.

Our main findings have shown that surface properties such as texture (and the material

properties it signals) can form a contextual link between the processing of objects and scenes,

and this interactive processing proceeds from the global to local scale of attention. Yet how

important is texture and material in context? It has been well established that spatial aspects (i.e.,

form) of object perception provide important cues in the search and recognition of objects (e.g.,

Biederman, 1987). However, the classification of objects in the natural world often requires

knowledge about non-spatial aspects such as the material properties of which an object is

composed (e.g., natural vs. manufactured, heavy vs. light, soft vs. hard, etc.), particularly when

form is degraded through occlusion or is uninformative. Texture is instrumental in providing the

visual cues necessary to infer such material properties, which subsequently aid in identification

and action planning necessary for interacting with objects in our environment (Adelson, 2001;

Buckingham et al., 2009; Gallivan et al., 2014). These cues may be highly influential in drawing

attention to contextually relevant objects within our immediate environment, and the nature of

the relationship between an object and its background. Consequently, contextual information

extracted from environmental texture cues can facilitate object search and recognition through

knowledge about real-world scene categories (e.g. natural vs. man-made). Interestingly, in

31

addition to its role in scene (e.g., Epstein & Kanwisher, 1998) and texture perception (e.g., Cant

& Goodale, 2007), parahippocampal cortex has also been implicated in processing contextual

associations (e.g., Bar et al., 2008). Thus, the results of the present study provide a bridge

between multiple types of visual perception, and provide a unique opportunity to further

understand how the processing of visual features and higher-level contextual associations work

together to influence interactions between object and scene perception.

5.1 Conclusions

In summary, the present research provides clear evidence for the importance of global

visual texture in object and background perception, and demonstrates how surface texture

consistency may form a contextual cue through which early scene information may facilitate

object perception and recognition. Our findings thus contribute to knowledge surrounding the

importance of texture in visual perception by providing a mechanism through which objects and

scenes are processed interactively. Finally, in order to fully understand the nature of object and

scene perception in real-world environments, we argue that it is crucial to not only explore the

influence of spatial processing, but to investigate the role of non-spatial processing as well. Here,

we explore the role of both spatial and non-spatial forms of processing in the relationship

between an object and the scene in which it is contained, highlighting the importance of non-

spatial visual cues (i.e., texture) in object-scene perception.

32

References

Adelson, E. H. (2001). On seeing stuff: the perception of materials by humans and machines.

In Photonics West 2001-Electronic Imaging (pp. 1-12). International Society for Optics

and Photonics.

Bar, M., Aminoff, E., & Schacter, D.L. (2008). Scenes unseen: The parahippocampal cortex

intrinsically subserves contextual associations, not scenes or places per se. The Journal of

Neuroscience, 28, 8539-8544

Biederman, I. (1987). Recognition-by-components: a theory of human image

understanding. Psychological Review, 94(2), 115.

Biederman, I., & Ju, G. (1988). Surface versus edge-based determinants of visual

recognition. Cognitive Psychology, 20(1), 38-64.

Biederman, I., Mezzanotte, R. J., & Rabinowitz, J. C. (1982). Scene perception: Detecting and

judging objects undergoing relational violations. Cognitive Psychology, 14(2), 143-177.

Boyce, S. J., & Pollatsek, A. (1992). Identification of objects in scenes: the role of scene

background in object naming. Journal of Experimental Psychology: Learning, Memory,

and Cognition, 18(3), 531.

Buckingham, G., Cant, J. S., & Goodale, M. A. (2009). Living in a material world: how visual

cues to material properties affect the way that we lift objects and perceive their

weight. Journal of Neurophysiology, 102(6), 3111-3118.

33

Cant, J. S., Arnott, S. R., & Goodale, M. A. (2009). fMR-adaptation reveals separate processing

regions for the perception of form and texture in the human ventral stream. Experimental

Brain Research, 192(3), 391-405.

Cant, J. S., & Goodale, M. A. (2007). Attention to form or surface properties modulates different

regions of human occipitotemporal cortex. Cerebral Cortex, 17(3), 713-731.

Cant, J. S., & Goodale, M. A. (2011). Scratching beneath the surface: new insights into the

functional properties of the lateral occipital area and parahippocampal place area. The

Journal of Neuroscience, 31(22), 8248-8258.

Cant, J. S., Large, M. E., McCall, L., & Goodale, M. A. (2008). Independent processing of form,

colour, and texture in object perception. Perception.

Cant, J.S., & Xu, Y. (2012). Object ensemble processing in human anterior-medial ventral visual

cortex. The Journal of Neuroscience, 32, 7685-7700.

Davenport, J. L., & Potter, M. C. (2004). Scene consistency in object and background

perception. Psychological Science, 15(8), 559-564.

De Graef, P., Christiaens, D., & d’Ydewalle, D. (1990). Perceptual effects of scene context on

object identification, Psychological Research, 52:317-329

Epstein, R., Graham, K. S., & Downing, P. E. (2003). Viewpoint-specific scene representations

in human parahippocampal cortex. Neuron, 37(5), 865-876.

Gallivan, J.P., Cant, J.S., Goodale, M.A. & Flanagan, J.R. (2014) Representation of object

weight in human ventral visual cortex. Current Biology, 24(16): 1866-1873.

34

Greene, M. R., & Oliva, A. (2009). Recognition of natural scenes from global properties: Seeing

the forest without representing the trees. Cognitive Psychology, 58(2), 137-176.

Gordon, R. D. (2004). Attentional allocation during the perception of scenes. Journal of

Experimental Psychology: Human Perception and Performance, 30(4), 760.

Henderson, J. M., Weeks Jr, P. A., & Hollingworth, A. (1999). The effects of semantic

consistency on eye movements during complex scene viewing. Journal of Experimental

Psychology: Human Perception and Performance, 25(1), 210.

Hock, H. S., Gordon, G. P., & Whitehurst, R. (1974). Contextual relations: the influence of

familiarity, physical plausibility, and belongingness. Perception & Psychophysics, 16(1),

4-8.

Hoffman, J. E. (1980). Interaction between global and local levels of a form. Journal of

Experimental Psychology: Human Perception and Performance, 6(2), 222.

Joubert, O. R., Rousselet, G. A., Fize, D., & Fabre-Thorpe, M. (2007). Processing scene context:

Fast categorization and object interference. Vision Research, 47(26), 3286-3297.

Kimchi, R. (1992). Primacy of wholistic processing and global/local paradigm: a critical

review. Psychological Bulletin, 112(1), 24.

Lagasse, L. L. (1993). Effects of good form and spatial frequency on global

precedence. Perception & Psychophysics, 53(1), 89-105.

Motoyoshi, I., Nishida, S. Y., Sharan, L., & Adelson, E. H. (2007). Image statistics and the

perception of surface qualities. Nature, 447(7141), 206-209.

35

Navon, D. (1977). Forest before trees: The precedence of global features in visual

perception. Cognitive Psychology, 9(3), 353-383.

Oliva, A., & Torralba, A. (2006). Building the gist of a scene: The role of global image features

in recognition. Progress in Brain Research, 155, 23-36.

Oliva, A., & Torralba, A. (2007). The role of context in object recognition. Trends in Cognitive

Sciences, 11(12), 520-527.

Palmer, T. E. (1975). The effects of contextual scenes on the identification of objects. Memory &

Cognition, 3, 519-526. Renninger, L. W., & Malik, J. (2004). When is scene

identification just texture recognition? Vision Research, 44(19), 2301-2311.

Schyns, P. G., & Oliva, A. (1994). From blobs to boundary edges: Evidence for time-and spatial-

scale-dependent scene recognition. Psychological Science, 5(4), 195-200.

Torralba, A., Oliva, A., Castelhano, M. S., & Henderson, J. M. (2006). Contextual guidance of

eye movements and attention in real-world scenes: the role of global features in object

search. Psychological Review, 113(4), 766.

36

Supplementary Materials

Chromaticity

Chromaticity (hue and saturation) reflects the quality of colour information independent

of luminance. Previous research has suggested that chromaticity information may be diagnostic

in the recognition and identification of scenes, reflected behaviourally through optimal reaction

time and accuracy for appropriately coloured scenes (Oliva & Schyns, 2000), and

neurophysiologically through decreased frontal event-related potential (ERP) amplitudes and

delayed ERP onset for inappropriately coloured scenes (Goffaux et al., 2010). Therefore, in order

to obtain response latencies and accuracy representative of naturalistic scene and object

classification, the present paper aimed to maintain natural chromaticity within texture stimuli,

instead of presenting achromatic images misrepresentative of real-world environments. But to

prevent participants from simply using colour information in their judgements of texture, all

textures were initially selected based on similar and naturalistic variations in colour, limiting

diagnostic colour information across texture categories (see Supplementary Figure 1), and were

then adjusted through level adjustment (brightness, contrast and tonal range were adjusted in

order to approximate values across texture categories) using Adobe Photoshop CS3 software

(Adobe Systems, San Jose, CA). Finally, Matlab software (MathWorks, Natick, MA) was used

to calculate the chromaticity of each texture to ensure that all images were limited in distinctive

colour information (a visual representation of the chromaticity of each texture stimuli can be

seen in Supplementary Figure 1).

37

Luminance Outlier Analysis

As the stimuli used in Experiment 3 were rendered using dynamic lighting and contrast

representative of more natural scene environments, to prevent potential differences in response

latencies being driven by differences in the luminance of stimuli, a luminance outlier

investigation was performed using Matlab software (MathWorks, Natick, MA) prior to data

analysis. This investigation revealed a texture outlier with a significantly lower luminance (M =

79.65; see Supplementary Figure 2) than the mean luminance for all stimuli (M = 140.49). As a

result, all stimuli containing this texture (twenty-seven cases) were removed. It is important to

note, however, that the overall results reported in the main manuscript did not change markedly

after a post-hoc investigation with these outlier stimuli included in the data analysis, and all

significant effects were maintained.

38

Supplementary Figure 1. (A) Source images of each of the texture stimuli (4 paint, 4 rock) used in Experiments 1 and 2

prior to chromaticity and luminance adjustments (above), compared to a visual representation of the chromaticity

(calculated using Matlab software) for each image showing minimal distinctive colours (below). These images can be

compared to an example image (top left) depicting distinctive colours. (B) Each of the texture stimuli used in Experiments

1 and 2 after chromaticity and luminance adjustments, compared with an example image (top left) depicting distinctive

colours.

39

Supplementary Figure 2. (A) An example stimulus representing the average luminance prior to

outlier removal, and the histogram displaying its luminance (M = 140.49) on a 0-255 luminance

scale. (B) An example stimulus containing the outlier texture that was removed from the main

analysis and a histogram displaying its luminance (M = 79.65).

40

Supplementary References

Goffaux, V., Jacques, C., Mouraux, A., Oliva, A., Schyns, P., & Rossion, B. (2005). Diagnostic

colours contribute to the early stages of scene categorization: Behavioural and

neurophysiological evidence. Visual Cognition, 12(6), 878-892.

Oliva, A., & Schyns, P. G. (2000). Diagnostic colors mediate scene recognition. Cognitive

Psychology, 41(2), 176-210.

Documents

Surface Texture Consistency in Object and Background Perception · 2015. 4. 17. · Surface Texture Consistency in Object and Background Perception Matthew X. Lowe Master of Arts