| Sonar image processing, as the key technology in marine research and development, has important research value and application value in economic and military aspects, and some of the reliable optical image processing algorithms can not simply be effectively applied in sonar image processing, therefore it is urgent and valuable for the research of sonar image processing. Based on the basic characteristics of sonar image different from optical image, this paper did the research focusing on the following several major issues: problemâ… :to improve the performance of sonar image denoising without destroying the edge; problemâ…¡:to obtain clear, reliable, comprehensive sonar image; problemâ…¢:to get accurate sonar image segmentation; problemâ…£:to complete sonar image post-processing such as image retrieval, identification, tracking and so on. Then the corresponding solutions for these four questions were given one by one in the four chapters of this paper, sonar image denoising, fusion, segmentation and retrieval were done respectively, and the feasibility, validity and reliability of the proposed methods were verified through a lot of simulation comparing experiments.For sonar image denoising, this paper discussed the denosing methods in wavelet domain which are more popular in optical image denoising. Firstly, this paper introduced two relatively new wavelet forms-beyond wavelet and morphological wavelet, and made the application of contourlet and surfacelet in the former, morphological Haar wavelet and morphology median wavelet in the latter in sonar image denoising in order to discuss if its performance is as effective as in optical image denoising; secondly, the midpoint filter of nonlinear filter and morphology wavelet were combined together to construct the morphological midpoint wavelet under the perfect reconstruction condition, and then it was improved by multiplying, lifting, and enhancing to get more excellent denoising performance; finally, the simulation comparative experiments were taken and the results showed that the proposed algorithm has not only more excellent denoising performance but also more outstanding edge preserving capacity.For sonar image fusion, this paper discussed the problem that if getting a clear, accurate and reliable sonar image through fusion technique is feasible, and still in wavelet domain which is more popular in optical image fusion. Firstly, the application of ridgelet and curvelet of beyond wavelet and wavelet packet were maken in sonar image fusion, although these algorithms are relatively mature in optical image processing, they are rarely applied in sonar image fusion, so this paper discussed their feasibility; secondly, this paper proposed the concept of morphological wavelet packet from wavelet packet thoughts, and established morphological midpoint wavelet packet based on morphological midpoint wavelet constructed in denosing part, then applied in sonar image fusion, the simulation comparative experiments results showed that the proposed method is feasible and validity; finally, this paper made the fusion technology applied in image denoising and gave sonar image fusion denoising method based on multiple morphological wavelet packets, the experimental results showed that the proposed method has a better denoising effect and edge preserving ability.For sonar image segmentation, this paper tried to find a more meticulous and comprehensive segmentation method, and at the same time can overcome the negative effects of the shadow part in sonar image. First of all, several kinds of optical image segmentation methods were introduced:level set method and spectral clustering method, and the shadow effect in sonar image segmentation was overcome through the gray threshold transformation; secondly, by considering the relationship and difference between image segmentation and digital matting:the existing image segmentation is essentially based on pixel intensity or gradient operator, while digital matting considers transparency or the color percentage, this paper combined the two algorithms together to establish a new image segmentation method based on digital matting, and used the forefront spectral matting method in digital matting, on the other hand, aiming at the disadvantage that usually take shadow as target in sonar image segmentation, this paper gave morphological hat and bottom transformation preprocessing plan to set up a sonar image segmentation system based on spectral matting; thirdly, the simulation comparative experiments of various segmentation methods were taken, and the experimental results showed that the proposed method can get a better sonar image forground target segmentation.For sonar image retrieval, this paper focused on how image retrieval successfully achieved for the special sonar image. Firstly, the existing outstanding image retrieval method based on mutual information and improved region growing segmentation MI in optical image retrieval were introduced, and their shortcomings were explored though the application in sonar image retrieval; secondly, in view of the specific requirement that sonar image retrieval is primarily forground target retrieval, this paper introduced the idea of region mutual information and weighted the forground mutual information, and built the sonar image retrieval system based on maximum weighted region mutual information by means of making the weighted region mutual information instead of the original global mutual information; thirdly, this paper compared the improved mutual information retrieval method with other mutual information retrieval methods in sonar image retrieval system, and the results showed that the proposed method is more suitable for sonar image retrieval.
|
PDF Full Text Request