Image Thresholding Segmentation Algorithm Based On Two Dimentional Histogram

Image segmentation is an very basic and important content in image processingtechnology,and it is a fundamental and pivota procedure in computer vision,the corecontent of image segmentation is dividing an image into a number of caregories of thesame nature according to certain criteria,to extract the part the people are interested in,and segmentation accuracy directly affect the work of subsequent image analysis andunderstanding, therefore, Image segmentation has extensive and important applicationvalue. Threshold segmentation method is the most commonly used imagesegmentation method, it is based on the gray histogram information of target image toacquire optimal threshold, meanwhile, it is simple, computationally efficient,segmentation results good.From the perspective of image thresholding, by reviewingand summarizing the existing research on the basis of a number of thresholdsegmentation method some corresponding studies, including the following aspects:The article first introduces the research background and significance, and theimage threshold segmentation research status quo at home and abroad are summarized,then introduces some basic theories of image segmentation, including the definition,the typical classification, several classical threshold segmentation algorithms, theimage segmentation quality evaluation, and the concept of one dimensional histogram,two dimensional histogram and three dimensional histogram.Thresholding based on minimum inner-cluster mean absolute deviation from themedian (MAD method) is a good promotion algorithm of the classical Otsu thresholdmethod. It can effectively overcome the problem that Otsu method generatesthreshold migration when the distribution of gray image in the gray level histogram isheavy-tailed or skewed. However, for some infrared images which are lowsignal-to-noise ratio, low contrast, blurred edges or of which the target and thebackground differ greatly, the obtained segmentation result by MAD method is oftenineffective as gray level histogram is out of certain regularity. In order to deal withthis problem, a two-dimensional thresholding based on MAD criterion is proposed.Taking advantage of two-dimensional histogram can present a heavy-taileddistribution or different-bimodal for those mentioned image, the proposed methodobtain the optimal threshold vector by means of2-D MAD threshold criterion, which is an extension of MAD threshold criterion, to achieve image segmentation. Thealgorithm of decomposition and dimensionality reduction is given in order to speed upthe calculation speed in the two-dimensional MAD method.Maximum Scatter Difference Image Thresholding is a very kind of improvedalgorithm for the classical Otsu thresholding,however,like other thresholdalgorithm,when the image is disturbed by noise or by illumination,the onedimensional maximum scatter difference threshold method is difficult to obtain goodsegmentation results,in order to resolve this problem we propose a two-dimensionaloblique MSD threshold algorithm,which is a extention of MSD threshold method,anda fast recursive algorithm is given to accelerate the speed of operation,it also analyzesand summarizes the variation of the two dimensional histogram’s region division bychanging the threshold criterion’s parameter C.Aiming at the shortage research of threshold output function which is used fortwo dimentional threshold segmentation,a new threshold output function and thecorresponding two dimentional histogram cross division are proposed, the histogram’sregion is divided by two lines which both via thresholding vector point.the anglebetween line one and gray level axis is α and the angle between line two and graylevel axis is β,different threshold output function and their correspngding histogram’sregion division will cause diffierent segmentation results,this will be reflected in theinternal uniformity of the target and background,accuracy of the edge and the noiseimmunity,in order to investigate this phenomenon,the article makes a detailedcomparison between the cross division method and several common histogram’sdivision method,including vertital division method,oblique division method andθ-division method,in the experimental results section the impact of the parameters αand β for the the internal uniformity of the target and background,accuracy of theedge and the noise immunity were summarized.Finally, the two-dimentional threshold image segmentation methods aresummarized and prospected.