Research On Image Edge Detection Algorithm Based On Non-subsampled Shearlet Transform

Edge is one of the main structures of an image carrying important geometric structure information.Edge detection is the key technology of industrial detection,character recognition and other applications.The correct edge detection plays an important role in image analysis and understanding.Non-subsampled Shearlet transform(NSST)is an advanced multi-scale geometric analysis tool.It has many excellent characteristics,such as multi-scale,multi-direction and translation invariance.It can achieve sparse representation in the true sense of the image.In this thesis,NSST is used as the theoretical basis of the research to explore and study the image edge detection.The main research contents of the thesis are as follows:Firstly,we propose an improved adaptive Canny edge detection algorithm on the problems of low accuracy of edge location,susceptibility to noise and lack of self-adaptation in traditional Canny edge operator.This algorithm abandons the traditional way of setting threshold,and uses iteration method to update the calculation lag threshold.And it improves the adaptability of the algorithm.Meanwhile,it increases the gradient information of diagonal direction and can extract edge information from many directions.The experimental results show that the algorithm can overcome the influence of noise and improve the accuracy of edge location.Secondly,aiming at the problem that edge detection methods in traditional transform domain do not make full use of coefficient distribution,have direction limitation and pseudo-gibbs effect,an image edge detection algorithm based on improved Canny and Fuzzy C-mean clustering in NSST domain is proposed.In this algorithm,the image is decomposed into high and low frequency components by NSST and the distribution characteristics of the coefficients of every pixel in each high frequency direction subband are analysed deeply.It combines modulus maxima method and Fuzzy C-mean to detect the edge of high frequency subband.The improved Canny edge detection algorithm is used to process the low frequency subband.The simulation results show that the algorithm has the advantages of high positioning accuracy,fewer false edge pixels and continuous edge.Thirdly,aiming at the distribution characteristics of edge pixels in target image and the problem that traditional target edge detection does not make full use edge features of image in spatial domain,an image edge detection algorithm in NSST domain combining image block clustering with mathematical morphology,which is based on the improved transform domain edge detection framework.According to distribution characteristic of edge pixels in images,we add image segmentation and clustering steps to remove some image blocks without edge pixels before edge detection.Meanwhile,we use improved double structure anti-noise expansion corrosion operator to replace the Canny edge detection operator for low frequency component detection after NSST.Compared with other algorithms,this algorithm has better subjective edge detection effect,higher robustness and efficiency.