Study Of Cervical Cell Recognition Based On Snake Segmentation And SVM

Cervical cancer is the third malignant tumor of women after breast cancer and colorectal cancer in the world, which is the most common female reproductive tract malignant tumor. The existing medical screening methods need a large number of cell pathology experts to participate in, time-consuming, and it is unfavourable for a large number of screening for cervical cancer. With the development of medical image processing technology, combining with the cervical cells medical knowledge and the experience of cell pathology experts for cervical cancer screening, we can adopt this kind of technology to recognize cervical cancer cells and accomplish early screening, which is very valuable in the clinical diagnosis of cervical cancerThe cervical smear is very easily polluted in the process of production, thus making the cervical image existing the situation of image blurring, uneven dying, contrast of images are weak and so on. The main research content of the paper is how to segment the polluted cervical cell image with a degree of high precision and to extract the feature parameters to achieve the purpose of cervical recognition.To accomplish the segmentation with a degree of high precision, this paper is based on active contour model and gradient vector flow model. Then we ameliorate the gradient vector flow and use it as the segmentation algorithm. On the one hand, the modified GVF model behave like isotropic diffusion within homogeneous regions; On the other hand, it can keep the curve diffusion in the direction parallel to cervical cell or nucleus boundary, which can decrease the initial sensitivity and converge to the hollow area. The cervical cell change mainly reflected in the cervical cell shape, and this makes the modified gradient vector flow model is more suitable for high precision segmentation of cervical cell. Based on the segmentation result of cervical cell, and combining with the experience of cell pathology experts, we extracts the morphological features parameters and color features parameters, and then analyze the feature’s effect about recognition according to the one-dimension of dispersion degree. Meanwhile, this paper analyzes the spatial dispersion made by two uncorrelated feature parameters which come from selected feature parameters, and then obtains two-dimension feature vector that is suitable to cervical cell recognition. Based on the selected feature vector, we add the uncorrelated feature parameter to form the training set, and this article selects the support vector machine (SVM) as cervical cell recognition classifier. Afterwards, we adjust the corresponding feature vector in order to improve the recognition rate according to the recognition result.In this thesis, we focus on the research of cervical cell segmentation, and make a systemic description and recognition result analysis for feature extraction and recognition. Experimental results show that, the modified gradient vector flow model could segment the cervical cell and nucleus with high precision, and help SVM achieve the purpose of recognition base on the segmentation result.