Feature Extraction And Recognition Of Pigmented Skin Image

Due to environmental degradation, various skin diseases are emerging, the incidence and mortality for skin disease are rising, which makes the World Health Organization began to pay attention to it. Now days, the traditional diagnosis for skin disease is highly dependent on subjective judgement, due to the common in diseases, there may be some misdiagnosis phenomenon. If the Computer Vision and Image Processing technology are used in computer-aided diagnosis system, the theoretical research and application value for early prediction and scientific treatment of skin diseases is very far-reaching.The main topic is about image segmentation and feature extraction in the computer aided diagnosis support system for skin diseases.(1) The LGBF-V model is proposed based on the LGBF(Local and Global Binary Fitting) model, and using the curve evolution theory for skin image segmentation, which could meet the requirements of image segmentation.(2) According to the characteristics of pigmented skin diseases, a method for extracting color histogram feature is proposed based on the Naive Bayesian Classifier, and it could reduce the dimension of color histogram feature and ensure the integrity of image color information.(3) Last, merging color feature, texture feature and sharp feature into one feature with CCA(Canonical Correlation Analysis) to enhance the recognition accuracy.The experiment results show that it is effective for skin disease image segmentation with LGBF-V model and reach the “full segmentation” requirements. In the process of recognition, this paper proposes a color feature extraction method based Native Bayesian Classifier, and fuses color, texture and sharp feature with CCA. Last, using SVM(Support Vector Machine) classifier for training and recognition. Eventually, a higher recognition accuracy is obtained.