Search

Comparing to optical images, sonar images experience more severe conditions such as degradation, light scattering, low contrast, and blurring. To guarantee the performance of the post processing such as object detection and classification, the need of a proper denoising method for sonar images has claimed. Conventional wavelet transform(WT) domain denoising method may be good candidates but they do not properly work for some cases which have ambiguity in edge and noise separation. Since image blurring after denoising is generally caused by the loss of edge information with low signal-to-noise ratio, the denoising performance may be improved by the introduction of a proper edge decision rule. In this paper, we propose a stationary WT denoising method with edge map and non-local means(NLM) filter for edge decision. Experimental results show that the proposed method has better performance than the three conventional methods(Visushrink, Sureshrink, and Bayesshrink) in terms of the edge preservation and the peak signal-to-noise ratio. [ABSTRACT FROM AUTHOR]

In sonar images, the object shape is often distorted by the external conditions such as sedimentation, pulse repetition interval, ship speed, and redirection, which is generally modeled as the affine transform. Scale invariant feature transform(SIFT) and pyramid-SIFT(p-SIFT) are known as a good solution to the scale and rotation variance. However, these methods are vulnerable to the shearing problem. To solve the affine problem in sonar imaging systems, we propose a new keypoint descriptor extraction method. The proposed method generates multiple keypoint descriptors based on affine transformed images, and picks out effective keypoint descriptors. Test results show that the proposed method, in terms of recognition rate, has better accuracy than those of the conventional methods for scaling of x, y-axis, rotation, and shearing cases. The average improved accuracy rate of SIFT and p-SIFT is 7% and 4%, respectively. [ABSTRACT FROM AUTHOR]