Based On Multiple Classifier Fusion Of Single Cervical Cell Image Segmentation,Feature Extraction And Classification Recognition Method Research

Cervical cancer is one of the most frequent malignant tumors in female and serious harmful to women’s health and life. The regular cervical screening can lower the death rate, while it produces an enormous number of cell smears which imposes great pressure on physician. On the basis of cervical cell pathological diagnosis technology, this paper uses computer to do a quantitative analysis and automatic interpretation on cervical smear, which has a good practical value in cervical cancer screening and early diagnosis.In this paper, the author takes the automated classification and recognition of cervical cell image as the research target. On the basis of previous studies, this paper uses the digital image processing and pattern recognition technology, proposing a cervical cell image classification method based on fuzzy integral multiple classifiers fusion, which mainly targeted on cervical cell segmentation, feature extraction and classification recognition, including a single cervical cell cytoplasm, cell nucleus contour image of accurate positioning, cell image texture feature extraction and multiple classifier fusion using fuzzy integral Predator-prey model optimization fuzzy measure to improve classification accuracy in cervical cells. The main researches are completed as follows:(1) This paper proposes a new model based on an improved CV, which used for the precise segmentation of nuclear and cytoplasmic contours in a single cell image. First of all, this paper combines with the gradient information of the active contour model fitting center weighted introduced adaptive weight w. Secondly, this paper constructs the negative exponential g|((?)I)|as velocity function and use |(?)φ(x,y)| instead of δε(φ(x,y)), to get clearer image edge. Finally, this paper introduces generalized fuzzy operator to strengthen the fuzzy boundary of curve evolution, improves the speed and accuracy of the segmentation.(2)This paper also provides quantitative analysis of a single cervical cell, generalize and summarize the classification characteristics, extract morphology, color, texture, light density characteristics from a single cervical cell. This paper proposes a fusion of LBP and GLCM method on texture feature extraction method, genetic algorithm is used to realize the feature dimension reduction, for optimum combination characteristic, used for subsequent classification recognition.(3) This paper constructs support vector machine (SVM) classifier, K-nearest neighbor (KNN) classifier, artificial neural network (ANN) classifier, using fuzzy integral fusion of the three single classifier for cervical cells image pattern recognition. To optimize classification and recognition, this paper proposes predator-prey model to optimize the fuzzy measure, then to make the classifier achieve its optimal performance.Finally, numerical simulation experiment on the algorithm of cervical cell image segmentation, feature extraction and selection and cervical cell recognition are completed in this paper. Experimental results show that the proposed method can complete cervical cells image pattern recognition well, which shows a good practical value.