AN EFFICIENT IMAGE ENHANCEMENT USING DOMINANT BRIGHTNESS LEVEL ANALYSIS AND MORPHOLOGICAL

EDGE FILTERING

Abstract• The project presents satellite image contrast enhancement based on

Lifting wavelet based multi scale decomposition and morphological edge preserving filtering approach.

• Here the system processes the input true color image in separation of individual planes to adjust contrast.

• This combination will be implemented to increase the visual perception of satellite color images.

• The low frequency will be enhanced with dominant brightness level analysis and High frequency band coefficients are enhanced with top hat filtering model.

• All enhanced frequency subbands are reconstructed with inverse wavelet decomposition. Finally the reconstructed images will be post processed with non flat ball shaped structuring element and morphological erosion process.

Objective

• To improve the quality of low contrast satellite image using Lifting wavelet based multi scale decomposition dominant brightness level analysis, soft thresholding and morphological edge preserving top-hat filtering approach.

Existing method

• Contrast limited adaptive histogram equalization

• Singular Value Decomposition

• Retinex method

Drawbacks

• Low accuracy in image quality

• It degrades sharpening details due to image border

artifacts

• Less preservation of image content during enhancement

Proposed Method

Satellite image contrast enhancement for vision system based on,

• Lifting wavelet based multi resolution analysis, adaptive

intensity transformation method and Morphological Edge

preserving top- hat filtering

Methodologies

• Multi scale Decomposition with Lifting wavelet filter

• Dominant Brightness level analysis

• Intensity transfer function

• Wavelet Soft thresholding

• Edge Preserving Top hat Filtering

Block Diagram

Low Contrast image

Plane Separation

HF bands

Histogram Equalization

Multi Scale Decomposition

Adaptive intensity transfer function

Soft thresholding

LF band

Plane Concatenation

Quality Evaluation

Image Reconstruction

Top hat Filtering

Input Image with Separated Planes

R Channel G Channel B Channel

Image After Equalization

• LWT has been employed in order to preserve the high-frequency components of the image. LWT separates the image into different subband images, namely, LL, LH, HL, and HH.

• Low frequency subband contains overall brightness of an input and high-frequency subband contains the edge information of input image.

• To avoid problems with floating point precision of the wavelet filters, Lifting scheme based DWT will be used.

• Lifting scheme is used to map the wavelet coefficients to integer coefficients and here Daubechies type wavelet filter is used.

Lifting Wavelet Transformation

Block diagram

Horizontal(Rows)Vertical(Columns)

L even+(H/2)

H Even-odd

L

L

H

H

Image with resolution Level R

L: Low pass filter

H: High pass filter

N x M

N x M/2

N/2 x M/2

LL

LH

HL

HH

Image corresponding to resolution Level R-1

Detail Image corresponding to information visible at the resolution Level R

even+(H/2)

Even-odd

even+(H/2)

Even-odd

Forward Lifting in IWTStep1: Column wise processing to get H and L H = (Co-Ce) and L = (Ce+ [H/2]) Where Co and Ce is the odd column and even column wise pixel valuesStep 2: Row wise processing to get LL,LH,HL and HH,Separate odd and even rows of H and L,Namely, Hodd – odd row of H, Lodd- odd row of L Heven- even row of H, Leven- even row of LLH = Lodd-Leven ,LL = Leven + [LH / 2] HH = Hodd – Heven ,HL = Heven + [HH / 2]

Reverse Lifting scheme in IWT Inverse Integer wavelet transform is formed by Reverse lifting scheme. Procedure is similar to the forward lifting scheme.

Continues…

LWT Sub-band Structure

LL: Horizontal Low pass& Vertical Low pass

LH: Horizontal Low pass& Vertical High pass

HL: Horizontal High pass& Vertical Low pass

HH: Horizontal High pass& Vertical High pass

Snapshot

• To reduce image distortion and saturation artifacts and preserving the image quality, dominant brightness levels are analyzed from image luminance subband.

• This will be performed for low frequency component image and it is decomposed into three sub layers.

• These are low, middle and high intensity layer based on brightness region in the luminance image.

• The dominant brightness level for this subband will be determined by the following expression,

D(x, y) = {log L(x, y) + ε}Where, L(x, y) - Pixel intensity at (x, y) and ε - Small constant factor that prevents the log function from diverging to infinity.

Brightness level analysis

LL Enhancement

Low frequency subband coefficients

Find Dominant brightness

Decompose low, middle, high

intensity layers

Find adaptive transfer function

Enhanced LL band

Snapshot

• These transfer function will be determined by using knee transfer function and gamma adjustment function for enhancing the three intensity layers.

• The knee points for low, middle and high intensity layers are,

Pl = bl + wl(bl – ml) , Ph = bh - wh(bh – mh) . Where, Pl - Low bound , Ph - high bound, wm - Tuning parameter

and mm , ml , mh – Mean brightness in three intensity layers.• Gamma adjustment function will be, Gk(L) = { (L/Mk)1/γ - ( 1 - L/Mk)1/γ + 1} Where, Mk = Size of each intensity range and Ml = bl ,

Mm = bh - bl, Mh = 1 – bh

Adaptive Intensity Transfer Function

Restoration for High Frequency band• Wavelet generated high frequency subbands are restored by soft

thresholding method. Here the threshold will be selected for shrinking high frequency subband coefficients to remove the noise.

• The soft threshold will be determined by level dependent method, Th= sqrt (2.*sigmahat.^2 * L)Where, L = Number of coefficients. sigma = median(C)./0.6745

Where, C - Coefficient Matrix,

Continues…

• The soft thresholding is defined by,

Coeff ’ = sign(Coeff) * (Coeff – T) if Coeff > T = 0 if Coeff < = T

• This Process is applied for all high frequency coefficients obtained from wavelet decomposition.

• These restored high frequency subbands are further post processed with morphological top hat filtering to smooth the detailed components.

Top-hat filtering

High frequency bands

Apply top-hat filter

Top-hat filter: Input – opening of Input

Edge Enhanced bands

Morphological opening: Erosion followed by

dilation

• It is used here to enhance the details present in the high frequency band and it sharpening the edges and textures

Continues…• Top-hat filtering is a type of morphological Edge

sharpening filter is used for High frequency subbands.• It processes the image based on shapes and here ‘line’

structuring elements are used for defining the shapes.• Top hat filtering requires an morphological opening

operation and opening is combination of dilation and Erosion.

• This filtering will be applied in all three directions such as horizontal, vertical and diagonal Edges Details.

• Dilation: It is the process of adding a pixel at object boundary based on structuring element.

• Erosion: It is to remove the pixel from the object boundary depends on structuring element.

Continues…

Dilation (D) HF (bit Xor) Se

Erosion (E) HF (bit Xnor) Se

Opening (O) E (bit Xor) Se

Top-hat Filter HF – O (HF)

I – Input Image, Se = Line Structuring Element, HF – High Frequency bands

Snapshot

Performance metrics

The performance of system will be evaluated with following metrics,

• Measure of Enhancement(EME):

EME = (1/(M*N))*[(Imax/(Imin + C))*log(Imax/(Imin + C))]

Where,• M,N represents the total number of elements in an image, • Imax represents the maximum Intensity Value,• Imin represents the minimum Intensity value • C represents a small constant to avoid dividing by zero.

Here C = 0.0001

Continues…Gaussian distribution Curve: The parameter μ in this definition is the mean or expectation of the distribution (and also its median and mode). The parameter σ is its standard deviation; its variance is therefore σ 2. A random variable with a Gaussian distribution is said to be normally distributed and is called a normal deviate.

It is used to illustrate how much changes occurred in the illumination from low contrast input to enhanced image due to this enhancement Process.

Performance Graph

Measure Of Enhancement for Input : 2.6981e-005

Measure Of Enhancement for Output : 2.3809

Advantages

• Better accuracy interms of edge preservation

• Flexible and highly compatible method

• It provide an optimal results for low contrast images from

satellite and digital camera

Applications

• Satellite imaging

• Digital Camera

Software Requirements

• MATLAB 7.5 and above versions

• Wavelet and Image Processing Toolboxes

ConclusionThe project presented the contrast enhancement approach based on dominant brightness level analysis and adaptive intensity transformation for remote sensing images. This algorithm computed brightness-adaptive intensity transfer functions using the low-frequency luminance component in the wavelet domain and transforms intensity values according to the transfer function gamma adjustment function based on the dominant brightness level of each layer. High frequency subbands are processed with shrinkage rule soft thresholding to reduce the impact of noises and sharpened with morphological top-hat filtering by preserving Edges. This method proved that an enhance the low quality images with less image distortion and preserves the edge details. The system performance will be measured through parameters such as measure of enhancement and Gaussian distribution function.

References[1] Bin Wang D, Rose M and Aly A Farag, “Local Estimation of

Gaussian-Based Edge Enhancement Filters Using Fourier Analysis”, IEEE Transactions on Acoustics, Speech, and Signal Processing, (1993), Vol. 5, pp. 13-16.

[2] Day-Fann Shen, Chui-Wen Chiu and Pon-Jay Huang, “Modified Laplacian Filter and Intensity Correction Technique for Image Resolution Enhancement”, IEEE International Conference on Multimedia and Expo, (2006), Vol. 7, Nos. 9-12, pp. 457-460.

[3] Cheevasuvit F, Dejhan K and Somboonkaew A “Edge Enhancement Using Transform of Subtracted Smoothing Image”, ACRS, (1992), Vol. 3, No. 12, pp. 23-28

[4] Jin Jesse S “An Adaptive Algorithm for Edge Detection”, MVA’SO IAPR Workshop on Machine Vision Applications, (1990), Vol. 9, November 28-30, pp. 14-17.