Study On Recognition And Segmentation Technology Of Cervical Cell Image

Cervical cancer is a kind of malignant tumor with high incidence.The quantitative analysis and identification of cervical cells have significant practical value in precancerous lesions and cervical cancer of screening and diagnosis. Using image analysis and pattern recognition technology, we make a study on the techniques of cervical cells segmentation, characteristic parameters calculation and selection and identification combined with the basic knowledge of cervical pathology. The main contents of this thesis can be summarized as follows:On study of segmentation technology for Cervical Smear image,this paper takes an example of Cervical Smear image. Two methods are proposed, which are based on OTSU and improved CV model and double layer watershed segmentation algorithm respectively. In the first method, being OTSU algorithm has the unsatisfactory results in Cervical Smear image and CV model is sensitive to the initial level set function and has inefficient curve evolution respectively. OTSU is used for coarse segmentation two times, providing local region and initial situation for CV model. Therefore depresses the sense complexity. In traditional CV model, the edges combined with global information and local information would be seriously suppressed by the Dirac function in order to improve the accuracy. The second method, taking different color characteristics of nuclear edge into account, two times watershed segmentation of different parameters are carried out to the image, then two segmentation results are fused to one image to comprise final result.On study of recognition technology for Cervical Smear images, this paper takes an example of Pap Cervical Smear images. First, the image of cervical cancer was converted into gray scale, and characteristics of connected area in the image are analyzed. Then effective morphology parameters are extracted as input of neural network to the survey. The images of cells samples are recognized and classified by the improved neural network. The experimental result shows that the classification using the nerve network to the cell works well.Experiments of above schemes are conducted on MATLAB platform and all of them get results recognized by physician. Feasibility of these algorithms is proved.