Multi-threshold Segmentation Of Micro Cell Image Based On Maximum Information Entropy

Cytologic diagnosis is a basic diagnosis method of clinical medicine. It is significant for the early diagnosis and qualitative diagnosis of cancers. At the present time, clinical cytologic diagnosis lies on the pathologist. The experience and the work intensity of pathologists affect the diagnosis conclusion. If the manual experiential diagnosis can be superseded by the more scientific computer-aid diagnosis, the pathologists'work intensity can be reduced and the diagnosis speed and accuracy can be increased Automatic segmentation of the microscopic cell image, as a basic step of computer-aid diagnosis, directly affects the analysis and diagnosis of the cell image in the next approaches. The distinction between pathological changed cell and normal cell is that the nucleolus of pathological changed cell is large and freaky, the ratio of the nucleolus to cytoplasm is increased, multi nucleolus, bare nucleolus and the size of cytoplasm is deviant, etc.Therefore, calculating the sizes, the shapes and the ratio of the nucleolus to cytoplasm are important reference data for identifying pathological changed cells. In order to calculate the sizes and shapes of nucleolus or cytoplasm, and to calculate the ratio of nucleolus to cytoplasm, researching multi-threshold segmentation of microscopic cell image is very valued.In addition, the microscopic cell images are easy to be polluted by the noises at the stages of sampling, the producing and the transmitting in the clinical work, it makes microscopic cell pictures are not enough clear. There are many kinds of tumor cell and the shapes of tumor cell are different each other. All of these facts increase the difficulty of automatic segmentation of the microscopic cell image.This paper aims at realizing the multi-threshold segmenting and removing the plenty of noises of microscopic cell images. The paper adopts mainly the maximum information entropy algorithm which can segment well the microscopic cell images and possess favorable resisting noises power.The first, the single threshold segmenting algorithm is generalized to multi-threshold segmentation of microscopic cell images. The MATLAB program language is applied to realize this algorithm. Experimental results show the multi-threshold segmenting algorithm based on 1-D information entropy can realize the multi threshold segmentation, but the power of the resisting noises is weaken.The next, in the paper, a novel multi-threshold segmenting algorithm of a polluted cell image was proposed. Based on the maximum entropy of the mean value-gradient co-occurrence matrix of the image the algorithm performs the multi-threshold segmentation of microscopic cell images polluted by the noises. Because the mean value of an image can smooth noises well and the image gradient can preserve more information of the image edge, the combination of two approaches will improve the segmentation result.In the paper the traditional region division method of the calculating entropy was changed in order to perform multi-threshold segmentation. The traditional entropy algorithm was improved in order to more suitable to MATLAB language programming and to get the segmentation thresholds more quickly.Experimental results show that the algorithm can speed up the calculation, raise the power of resisting noises and improve the segmentation effects.