The Partial Theory Of Image Segmentation And Key Technology Research

As the basis of image analysis and image understanding,image segmentation plays an important role in computer vision,thus it has become one of hot topics in the area of image processing.Recently,great progress has been made in the theory of image segmentation,and a series of excellent image segmentation schemes have been proposed.But so far,there has not been a method appropriate for all types of image segmentation and achieves optimal segmentation performance,it also becomes a problem in the image segmentation field.On a basis of superpixel,support vector machine(SVM),dual tree complex wavelet transform theory and etc,we try to propose three image segmentation algorithms in this paper,and good segmentation results are obtained in many kinds of images by utilizing these algorithms,more detailed information according our work as follows:1.A color image segmentation algorithm based on TWSVM superpixel classification is proposed,which is based on the theory of entropy rate superpixel and twin support vector machine(TWSVM).Firstly,the original color image is divided into the superpixel region by using the entropy rate superpixel generation algorithm;Then,the color feature of the superpixel is extracted by histogram and the texture features of the superpixel are extracted by the average phase and the average amplitude of the double tree complex wavelet transform(DTCWT);Next,the training sample of TWSVM is selected by the maximum between class variance threshold method;Finally,the trained TWSVM model is used to classify the superpixel and get the segmentation result.In this method,the pixels with high compactness can be clustered by using the entropy rate superpixel method,and the classification of the pixels can be achieved by the TWSVM model with higher classification accuracy.The experimental results show that the proposed method has better segmentation accuracy and lower computation time than other traditional methods.2.The modified hybrid weight SVM has good classification performance,and a color image segmentation algorithm based on adaptive second-order statistics and hybrid weight SVM is proposed.Firstly,the adaptive second-order statistics is used as the pixel feature;Then,the training samples are selected by the linear two-dimensional Tsallis entropy;Finally,color image is classified by the modified hybrid weight SVM.This method uses the adaptive second-order statistics with strong anti-noise ability as the pixel feature,and improves the classification accuracy of the feature by the hybrid weight method.The experimental results show that the algorithm has desired segmentation results and good robustness to noise.3.Double density dual tree complex wavelet transform(DD-DTCWT)has strong ability to capture the important features of the image,hence a new image segmentation method based on DD-DTCWT-HMT is proposed.Firstly,the image is decomposed by DD-DTCWT,andthe relative phase local mean and amplitude are calculated;Then,the truncated generalized Cauchy mixture model(TGCMM)and the hidden Markov tree model(HMT)are used to model the relative phase local mean and amplitude.In the end,we use the context corresponding weight values fusion method to get the final segmentation results.In this scheme,we model by the TGCMM to replace the Gauss mixture model(GMM),which has strong capacity to describe peak heavy tailed,and the texture features of the image are better extracted by the relative phase local mean and amplitude of DD-DTCWT.The experimental results show that this algorithm has better segmentation accuracy compared with the partial existing segmentation method.