The Research And Application Of Thresholding Algorithm Based On Class Uncertainty Theory

Segmentation is one of the most important techniques in image processing,visual analysis and understanding.As a popular image segmentation method,thresholding is widely used in a variety of fields and has been researched for many years.However,how to choose a optimum threshold adaptively is still a great challenge,especially for fuzzy and uneven lighting image with noise.As one of the most classic methods,class uncertainty theory based on image intensity can effectively measure the class uncertainty of intensity in the statistical sense.The statistical information and space information embedded in image were combined in this thesis,through class uncertainty theory and image texture information.Then energy function was constructed and optimized to search the optimum threshold.The main work of this thesis is as follows:(1)Classical MHUE method is extended to 2-dimension histogram which is constructed by image gray information and the mean of gray neighborhood.Then energy function is constructed by class uncertainty in 2-dimension of gray level and region homogeneity.Experiments demonstrated that 2-dimension MHUE has better performance and robustness than classical MHUE for fuzzy and noisy image.(2)A local 2-dimension MHUE algorithm was designed to solve the problem of thresholding selecting for the image which is slightly influenced by uneven illumination.The image was divided into several local neighborhoods,then the threshold is selected independently by 2-dimension MHUE in each sub-region.Experiments proved the effectiveness of proposed method in solving the problem of slightly uneven lighting image and the results of proposed method were better than some classical thresholding method,such as 2D Otsu,Sauvola's method,etc.(3)In order to solve the problem of image segmentation with fuzzy,noise and uneven lighting,the local gray mapping theory is proposed.The image is preprocessed by local gray mapping which can improve the contrast of the local area and can reduce the influence of uneven illumination.Then we put forward the theory of regional stability which reflects the spatial information of images.Combined with the theory of class uncertainty based on gray scale and regional stability theory,we build energy function.The threshold that minimized the energy function is regarded as the optimal threshold.Experiments compared the proposed methods with several classical thresholding methods,which demonstrated that our method has better performance and strong robustness.