View
248
Download
6
Tags:
Embed Size (px)
Citation preview
Region Filling and Object Removal by
Exemplar-Based Image Inpainting
IntroductionIntroduction
A new algorithm is proposed for removing large objects from digital images.
this problem has been addressed by two classes of algorithms:
1) “texture synthesis” algorithms for generating large image regions from sample textures
2) “inpainting” techniques for filling in small image gaps.
Key ObservationsKey Observations A. Exemplar-Based Synthesis Suffices
B. Filling Order Is Critical
AlgorithmAlgorithm Each pixel maintains a color value (or “empty,” if the p
ixel is unfilled) and a confidence value.
Algorithm iterates the following three steps until all pixels have been filled.
1) Computing Patch Priorities 2) Propagating Texture and Structure Information 3) Updating Confidence Values
1) Computing Patch Priorities the priority computation is biased toward
those patches which 1) are on the continuation of strong edges. 2) are surrounded by high-confidence pixels.
Given a patch centered at the point p for
some , we define its priority as the product of two terms
C(p) the confidence term that measure of the amount of C(p) the confidence term that measure of the amount of reliable information surrounding the pixel p.reliable information surrounding the pixel p.
D(p) the data term that is a function of the strength of isophotes hitting the front at each iteration. D(p) the data term that is a function of the strength of isophotes hitting the front at each iteration.
(1) np estimated as the unit vector orthogonal to the line through the preceding and the successive points in the list
(2) is computed as the maximum value of the image gradient in .Robust filtering techniques may also be employed here.
2) Propagating Texture and Structure Information propagate image texture by direct sampling of the sour
ce region.
the distance between two generic patches and is simply defined as the sum of squared differences
Synthesizing One PixelSynthesizing One Pixel
SAMPLE
Infinite sample image
Generated image
Instead of constructing a model, let’s directly search the
input image for all such neighbourhoods to
produce a histogram for p
Really Synthesizing One PixelReally Synthesizing One Pixel
SAMPLE
finite sample image Generated image
However, since our sample image is finite, an exact neighbourhood match might not be present
So we find the best match using SSD error (weighted by a Gaussian to emphasize local structure), and take all samples within some distance from that match
3) Updating Confidence Values: After the patch has been filled with new pixel values, the confidence is updated in the area delimited by
Results And ComparionsResults And Comparions
TimeTime
Shape of the selectShape of the select
Hand-drawHand-draw
Large objectLarge object
ENDEND
THANKS EVERYONETHANKS EVERYONE