Image Segmentation Using Multiscale Geometric Analysis And PCNN

The traditional wavelet theory concentrates on the multiresolution representation of signals, has developed both in theories and practice and is applied to many fields in image processing effectively. The new generational multiscale geometric analysis theories overcome the limits of wavelet, which is isotropy, poor direction selectivity and fails to represent image edges and textures sparsely. Image segmentation is always hotspot issue of image processing fields. Recent researches focus on segmentation methods in some particular domain using multiscale geometric analysis theories.At first, we analyze the advantages and disadvantages of wavelet shrinking threshold denoising. Then combining the characteristics of curvelet transform, we propose an improved and adaptive denoising method. The result of the experiment indicates that, comparing to the traditional wavelet shrinking threshold denoising method, the improved method achieves better visual quality and higher PSNR.This thesis analyzes the discrete contourlet transform, continuous contourlet transform and Nonsubsampled contourlet transform. We propose an image segmentation method based on the PCNN model and contourlet transform. This method takes advantages of both the contourlet transform sparely representing image edges and features of PCNN. The experimental results show that the method can achieve satisfactory in particular segmentation circumstances, as well as achieving accuracy and efficiency of segmentation.