Research Of Cervical Cell Image Segmentation Algorithm Based On GVF Snake And Classification And Recognition

In this paper, we took the cervical cell image as an example for the research of cell image segmentation, morphological characteristics, polar characteristics and identification application technology, mainly including cervical cell cytoplasm image, accurate contour extraction of the nucleus and the background, feature fusion cervical cells and cell classification method. The main researches include the following parts:(1). The GVF Snake model active contour based on adaptive threshold and rays gradient were proposed,which was used to locate the edge of the cervical single cell nucleus and cytoplasm of the image. Active contour model tracking algorithm was used widely in target edge GVF Snake active contour model, but for cervical cell images, especially when cytoplasm had relatively vague edge or the edge of the nucleus and cytoplasm adsorption was difficult to separate from each other, as well as blood cells and inflammatory cells influenced, dyeing degree distribution was not uniform, GVF Snake model cell edge was adsorbed to the wrong location. To solve the above problems, the following improvements were included in the paper:Firstly, we used adaptive threshold to remove cells background; Then, we used the rays gradient direction information calculation cell gray value; Finally, we used evolution GVF Snake model based on ray gray value on the stack in the evolution of the use of gray poor compensation algorithm. We also combined positive gradation difference suppress information to overcome the effects of noise, blood cells and inflammatory cells and other false edges.The Herlev database which verifies the effectiveness and feasibility of the proposed method in this article was used.(2). Based on the precise cervical cell image segmentation, we discussed the morphology parameters of the image of cervical cell, including nine kinds of geometric features and four kinds of textures. The 9 kinds of geometric features include:cytoplasmic perimeter, the perimeter of the nucleus, the length of the major axis of the vertical direction, the length of the horizontal width of the shaft, the ratio of nucleus and cytoplasm, the length of the central axis of the circumference, the average length of the shaft center to the circumference, the center of gravity to the maximum length of the perimeter, the average length of the circumference of the center of gravity. The 4 kinds of texture features include:co-occurrence matrix entropy, co-occurrence matrix contrast, contrast and roughness. Cervical single cell image includes the nucleus, cytoplasm and background.The three regions can be transformed into a polar coordinate system and can be extracted to the extreme value of the polar coordinates. The gradation value of 360 gray values make the formation of a very characteristic matrix. We will make the fusion of the characteristics of the polar coordinates and the previous morphological characteristics to study the identification of cervical cells.(3). Vector machines based on AdaBoost and SVM algorithm were used to improve the stability of the classification and diversity. Multiple features of extracted cervical cells will be integrated, identified and sorted in AdaBoost-SVM classifier in this paper. Combination of dual classifier can make up for their shortcomings and improve classification results. The results of experiment show that the combination of feature extraction methods and multi-feature fusion classifiers could improve the efficiency and accuracy of cervical smear screening and reduce the rate of misdiagnosis of cervical cancer.In this paper, we made a systematic research and improvement on cervical cell image segmentation, feature extraction and classification of cervical cells, etc.The results show that the method in this paper is a good way of complete quantitative analysis of cervical cells, and it is of good application value to automated screening analysis system for cervical cell image.