Multiscale Edge Detection Based On Nonsubsampled Contourlet Transform Related Technology Research

Edges are the prominent features in images and their detection are an essential task in computer vision and image processing. At present, although there are a large number of edges detection methods have been proposed, but the low accuracy and poor robustness in the presence of edge detection methods still not been solved. Therefore, how to find a "good" edge detection method has been the focus of many research scholars, and also we have to work towards it.In this paper, we propose a novel multiscale edge detection approach based on the nonsubsampled contourlet transform(NSCT): a fully shift-invariant, multiscale, and multidirection transform. Indeed, unlike traditional wavelets, contourlets have the ability to fully capture directional and other geometrical features for images with edges. The main contents of this paper are as follows:First,the research background, the significance and the development status of the image edge detection technique are introduced in detail. Besides, the main work and innovation of this paper also are summarized.Second, the basic concepts, the general procedures and the performance evaluation methods of the edge detection are expounded systematically. In addition, the basic principles and implementation method of contourlet transform and nonsubsampled contourlet transform have been careful analysis and research.Third, this paper proposes an image pixel classification algorithm based on k-means clustering. The algorithm firstly obtain the multiscale and multiple directions decomposition coefficients of the nonsubsampled contourlet transform.Second, distinguish noises from edges using nonsubsampled contourlet transform have the ability to fully capture the geometric information of images. Finally, the multiscale edge images are obtained.Fourth, for the non-single-pixel wide of the multiscale edge images problem, an edge thinning algorithm based on non-maximum suppression is proposed. According to the edge pixels in image has the largest energy coefficient at the edge directional subband of the nonsubsampled contourlet transform, the pixels gradient direction at all scales is obtained. Then we select the edge point candidates of the input image by identifying the NSCT modulus maximum at each scale. Experimental results show that this method can achieve edge thinning and improve positioning accuracy.Fifth, in order to solve the problem that how to effectively fusion edges from different scales in multiscale edge detection, this paper proposed a novel multiscale edge tracking algorithm. First, the proposed method obtained the input image’s multiscale edge images and gradient orientation maps. Second, based on the similar characteristics of the corresponding edge points of the adjacent scales, this algorithm improved the results of edge detection by performing coarse-to-fine edge tracking. Experimental results indicate that the proposed algorithm has the advantages of edge integrity and positioning accuracy, and has fewer false edge points.Finally, the multiscale edge detection approach based on nonsubsampled contourlet transform is obtained by effectively combining these three algorithms above mentioned. The simulation results show that, compared with other edge detection algorithms, the proposed algorithm can obtain better results with the measurement of human visual system and objective evaluation.