Zhengya Xu, Hong Ren Wu, Xinghuo Yu, Fellow, IEEE, Bin Qiu, Senior Member, IEEE Colour Image Enhancement by Virtual Histogram Approach IEEE Transactions.

Zhengya Xu, Hong Ren Wu, Xinghuo Yu, Fellow, IEEE, Bin Qiu, Senior Member, IEEE Colour Image Enhancement by Virtual Histogram Approach IEEE Transactions on Consumer Electronics, Vol. 56, No. 2, May 2010

Outline Introduction Previous researches Principle of the proposed method Experiment

Introduction Image enhancement Computer vision Biomedical image analysis Remote sensing Fault detection Forensic video/image analysis

Surveillance videos Varied lighting conditions Few levels of brightness Histograms have one end is unused while the other end of the intensity scale is crowded with high frequency peaks.

Previous researches Point-operation-based image enhancement Contrast stretching Non-linear point transformation improve visual contrast in some cases, but impairing visual contrast in other cases Histogram modeling DRSHE, BPDHE, GC-CHE Grey-level to gray-level Global transform [6] G. Park, H. Cho, and M. Choi, A Contrast Enhancement Method using Dynamic Range Separate Histogram Equalization, [7] N. S. P. Kong and H. Ibrahim, Color Image Enhancement Using Brightness Preserving Dynamic Histogram Equalization, [8] T. Kim and J. Paik, Adaptive Contrast Enhancement Using Gain- Controllable Clipped Histogram Equalization,

Previous researches Retinex theory Spatial operations May enhance the noise or smooth the areas that need to preserve sharp details Pseudo-colouring Map the grey-scale image to a colour image Require extensive interactive trials Adaptive histogram equalization Local enhancement The enhancement kernel is quite computationally expensive May yield unsatisfactory outputs

Our goals A fast adjustable hybrid approach Controlled by a set of parameters Take the advantages of point operations and local information driven enhancement techniques Can enhance simultaneously the overall contrast and the sharpness of an image For both colour and luminance components of a colour image

Principle of the proposed method Histogram Used to depict image statistics in an image interpreted visual format Luminance histogram and component histogram Provide useful information about the lighting, contrast, dynamic range, and saturation effects relative to the individual colour components Must not have very large spikes in the histogram of the enhanced image Intention of the proposed method: Find a monotonic pixel brightness transformation q=T(p) for a colour image Meet specific requirements Be as uniform as possible over the whole output brightness scale

Definitions and notations Pixel coordinates of a colour image M:height N: width of the image The pixel in RGB (Red, Green, Blue) colour space The pixel in colour space

Definitions and notations For each RGB colour channel, each individual histogram entry cardinality function,, :scale of component, usually256 The luminance channel histogram of an image The cumulative histogram from grey-scale image

Local geometric information An enhanced image with good contrast will have a higher intensity of the edges Laplacian operator : Applied to each of the RGB channels The sum of absolute value of the pixel processed with a Laplace operator

Principle of the proposed method Define w default :2, v default :1 is designed to suit special enhancement requirements for the image interpretation Using a normalization coefficient

The output histogram can be approximated with (16) by its corresponding continuous probability density : M and N are the height and the width, and the output brightness range is The desired pixel brightness histogram transformation T is defined as : Principle of the proposed method

The quantisation step-size is obtained as follows : The second term is used to enhance contrast for a specified range, The third term is dependent on the image structure. Parameter v can be adjusted In most cases, v is fixed as 1, since the enhanced result is not very sensitive to the change of the v

Principle of the proposed method The quantisation step-size is obtained as follows : Human vision is very sensitive to the interval value The default values of these parameters are:

Principle of the proposed method The number of reconstruction levels of the enhanced image must be less than or equal to the number of levels of original image. When the contrast of a dark area whose histogram spans a broad range of the display scale is enhanced, the bright areas may be out of the display range. Therefore, a hard-limit is needed

A hard-limit Using parameter t 222 smooth the enhancement contrast over the full brightness scale 111 44 1 is actually similar to t for human vision The ratio (Weber fraction), is nearly constant at a value of about 0.02 The default value of t is set to 3 If an image with its histogram basically concentrated in a very bright region, the image can be inversed [2] William K. Pratt, Digital Image Processing, John Wiley & Sons, 2008.

Transformed back to the RGB colour space After the contrast enhancement in the luminance channel, the output image is transformed back to the RGB space A histogram of RGB channels may be saturated at one or both ends of the dynamic range The linear mapping of video signal from the RGB colour space to the YCC colour space where the luminosity (Y) is a function of R, G and B which are normalized to 1, and denoted as Y(R,G,B)

Transformed back to the RGB colour space We only need to find the corresponding Y values to the upper and the lower bounds of the RGB channels The conversion from the RGB space to the YC B C R space is the number of the saturated pixels in the image

Transformed back to the RGB colour space Compared with four classical enhancement methods linear contrast stretching contrast reverse gamma correction histogram equalization Recent developed histogram equalization based methods DRSHE BPDHE GC-CHE The test images include well-known typical test images including Mountain,Scene, Meat etc Image size: 500x362 or 721X481 or 768X768or 731X487

Result a)original image b)output of the proposed approach-1, c) output of proposed approach-2 d) output of proposed approach-3 e) output of modified linear stretching f) output of histogram equalization.

Result a) original image, b) output of the proposed approach, c) output of histogram equalization, d) output of linear stretching, e) output of contrast reverser, f) output of modified linear stretching.

Result a) original image, b) output of the proposed approach, c) output of histogram equalization, d) output of linear stretching, e) output of gamma correction, f) output of GC-CHE

Result a)original image, b) output of the proposed approach, c) output of histogram equalization, d) output of gamma correction, e) output of PBDHE, f) output of DRSHE

Conclusion Based on modification of a virtual histogram distribution Information extracted from salient local features A new way to integrate colour and brightness information extracted from salient local features, for global contrast enhancement. Output value scaling bounds control

Zhengya Xu, Hong Ren Wu, Xinghuo Yu, Fellow, IEEE, Bin Qiu, Senior Member, IEEE Colour Image Enhancement by Virtual Histogram Approach IEEE Transactions.

Documents

colour image slide

ieee colour image enhancement

image statistics

color image enhancement

greyscale image slide

colour space slide

component histogram

cumulative histogram