Research On Multi-scale Image Segmentation Algorithm Based On Contourlet Transform

Image segmentation method has been a hot issue in the field of digital image processing. In the image segmentation, we often face the contradictions between segmentation accuracy and segmentation easy-divisibility, over-segmentation and under-segmentation, and these contradictions are difficult to be solved in regular scales. In engineering practice, multi-scale features or multi-scale effects are widespread, and it is often carried on in different scales while people observing and measuring processes. Therefore, only we utilize the multi-scale theory to describe and analyze these processes are we able to describe the essential characteristic of these processes more comprehensive and more accurately. We can obtain image information in different scales using multi-scale segmentation, so as long as we applying appropriate multi-scale data structures and selecting better extraction and segmentation strategy of multi-scale feature according to the data structure, it is possible to unify the contradictions written above. At present, using multi-scale analysis for image segmentation is a growing trend.In this paper, Contourlet transformation is set as a multi-scale decomposition tool of the image for multi-scale decomposition of the target image. The multi-scale segmentation of images is conducted by respectively applying segmentation strategies based on edge detection and Hidden Markov Tree Model. The main work of this paper is as follows:Firstly, multi-scale image segmentation is analyzed, two of the multi-scale analysis tools:wavelet transformation and Contourlet transformation are respectively introduced, and the characteristics and coefficients distribution are analyzed.Secondly, detecting the image edge is set as segmentation strategy, by analyzing the coefficients distribution features of Contourlet transformation, the image edge information is extracted using modulus maxima detection method and scale fusion of the edge information on different scales is conducted, then the edge detection segmentation method based on Contourlet transformation is proposed. It is confirmed by simulation experiments that the edge detected by this method is not only smooth and coherent but also richly detailed.Finally, by the study of image segmentation method based on wavelet domain HMT model combining with the characteristics of Contourlet transform coefficients, an image segmentation algorithm based on Contourlet domain HMT model is proposed, this method can capture high dimensional singular information of images better and the effect of the method is proved through experiments.