Home Articles Multi-model histogram linear contrast stretching for image enhancement

Multi-model histogram linear contrast stretching for image enhancement

ACRS 1998

Poster Session 3

Multi-model Histogram Linear Contrast Stretching For Image Enhancement

P. Komonvipaht, K. Wongsritong, F. Cheevasuvit, K. Dejhan, S. Chitwong, and S. Mitatha
Faculty of Engineering, King Mongkut’s Institute of Technology Ladkrabang
Bangkok 10520, Thailand.
E-mail: [email protected]

Abstract
Image enhancement, modifying the value of picture elements, is a classical method for improving the visual perception. Therefore, the processed image is more suitable than the original image for some applications. The most well-known method is the histogram equalization because of its automatic procedure. However, the effect of brightness saturation will be appeared in some quasi-homogeneous region. This causes from the merging or adjacent gray levels process for flattening the global histogram. Not only some lower brightness will be grouped together but also some higher brightness in order to uniformize the histogram distribution. Unlike the method of histogram linear contrast stretching the new wide histogram dynamic range can be assigned directly from the original narrow histogram dynamic range. However, the histogram of an image can be composed of many models of histogram. Different modals can be represented with different objects. To enhance each modals of histogram independently, the obtained result image will get better visual perceptibility than the global enhancement. The standard deviation of each modals will be calculated and summed for providing the proportional of stretching range. The resultant image can be provided the higher efficiency in image segmentation or image classification. Therefore, this paper tries to carry out a method of the multi-modal histogram linear contrast stretching. As each modal of histogram can be detected by using the change of eight consecutive signs, each sign is obtained from the difference between two probabilities of adjacent gray level.

Introduction
A digital image is encoded the picture element with L bits, the gray level of brightness will be varied from 0 to 2L-1. The probability of each gray level is defined by the following equation;

P(rk) = nk/N (1)

Where rk is the kth gray level, nk is the number of pixel for the kth gray level and n is the total number of pixels in the image. The distribution of all gray level probabilities is called histogram. If a digital image has a narrow dynamic range of histogram, the image will give low contrast. So the different objects in the scene will have almost the same brightness. This will cause the difficulty in some applications such as an object identification, an object classification, and etc. To improve the contrast for the image enhancement, the methods of histogram modification are always employed to spread out the dynamic range. The most well-known method is histogram equalization. However, this method will be spread out the histogram and always reach the whitest and the blacked brightness. By consequently, the saturation of brightness will be appeared not only in quasi-homogeneous of low gray level but also in quasi-homogeneous of high-gray level. Unlike the method of histogram linear contrast stretching, the problem of brightness saturation can be avoided if the narrow range of original histogram is prosperously assigned to spread out.

Model detection
Since a histogram of image will be discontinued and fluctuated, therefore linear interpolation and smoothing method are applied to conquer these problems. For the smoothing procedure, nine consecutives of gray level probabilities are used to average. This average value is replaced by the central probability. The average value can be calculated as the following equation.

Pn(rk) = 1/9 9Si=1 p(rk-s+1) for 5