Research On Moment Preserving Principle And Its Application In Image Segmentation

Image thresholding segmentation is an typical and basic issue in the image processing area, and it isalso a hotspot and hard part for the international research. The target and the background in one image donot have the same intensity and greatly affected by such factors as uneven light in the practical system.That is the reason why thresholding is significant. It is critical to image segmentation and image analysis.MomentPreserving Principle (MPP), on the principal of unchanged intensity moment of imagesegmentation, acquires thresholding without iteration and search. It is fast, but problems such asowe-segmentation, over-segmentation and ignorance of image details arise. According to related arithmeticresearch on the basis of MPP, and analysis of advantages and disadvantages of current thresholdingsegmentation, the thesis presents improved proposal. The results of this thesis are as follows:Firstly, after the detailed introduction of image thresholding segmentation and MPP, to solve theproblem of owe-segmentation and over-segmentation, the author addresses Fourier transform. After theinitial thresholding is acquired through MPP, the author ascertains the amplitude of threshold by Fourierspectrum of image histogram and the direction of threshold by image gray-mean. At the same time, theauthor will not adjust the initial thresholding which is effective. So that the problem of owe-segmentationand over-segmentation can be well resolved. The segmentation results of MPP are greatly improved.Compared with traditional MPP, the method introduced by the thesis is more effective.Secondly, traditional MPP may cause owe-segmentation, over-segmentation and ignorance of imagedetails, which could be negative for subsequent image analysis and image recognition. The thesis presentsan Moment-Preserving Segmentation Algorithm based on gradient fixed. The algorithm defines an image sharping function so that the original image can be sharped, and the details in the image are highlighted.Then the sharped image is segmented by MPP. The experimental results not only improve the segmentationeffect of MPP but also preserve more details of image edge, which entitle the segmented images moresemantic.Last but not least, it is difficult for traditional MPP, which is on the base of1-D gray-level histogram,to stretch to2-D. However, the unchanged intensity moment of image segmentation is one of the traits ofimages, in other words, moment-preserving trait of images. On some occasions, if using only onesegmentation method, the result could be not sastifactory or leads to failure. The thesis presents ansegmentation algorithm adopting MPP to choose1-D Otsu and1-D maximum entropy. Segmentationthresholdings acquired through MPP,1-D Otsu and1-D maximum entropy, and the best will be the onemost similar to the threshold acquired by MPP. Therefor, the segmentation results by this algorithm satisfysome certain principal on one hand and satisfy the trait of moment-preserving on the other hand. Theexperiment indicates that the algorithm resolves the problem of image complexity and diversity. Therefore,the MPP can be widely adopted in the field of image segmentation.