Invariant Moment Based Image Recognition Of High Resolution Imaging Sonar

With the further exploration and utilization of ocean and the development of submarine acoustic imaging technique, the sonar image recognition has become an important subject in the field of digital image processing. To solve the recognition problem of under water target in short distance, this paper makes a deep study on the de-noising and feature extraction technique of high resolution imaging sonar. The main contents include:(1) The paper gives a review of the internal and external present situation of sonar image processing and introduces the composition of the sonar image recognition system and main technical means at present. And the technical characteristics and properties of the high resolution imaging sonar have also been introduced. Besides, the paper analysis the technical difficulties in the current sonar image processing.(2) The noise disturbance of sonar imaging is analyzed. The paper makes a deep study on the usage of PDE (partial differential equation) in the de-noising of Gaussian noise and speckle noise. The PDE has a better local self-adaptive de-noising and global optimization function. Comparing with the previous classical de-noising methods, this method not only can better inhibit noise, but also can preserve the edge and texture characteristics of image, improving the effect of repressing noise and image restoration.(3) To deal with the problem of image rotation, translation and scale variance caused by the vehicle motion, we make a deep research on vision invariant theories of image recognition and propose a invariant feature based on discrete Radon transform of image edges after scale normalization. Radon transform can project 2D edge to 1D space, then we construct moment invariant in the space and use SVD (singular value decomposition) of different order moment invariant matrix to detect target feature. This method is robust and have higher recognition rate, and because only use the edge information, so operating is faster.(4) Another invariant characteristic construction method—the improved histogram invariant moment is proposed. This method uses the histogram's advantage of insensitive to size, orientation and object movement, stands on the invariant characteristic of invariant moment, consideration to the shape factor of target, so it can overcomes the disadvantage of traditional histogram such as the similarity of different images' histogram due to the loss of spatial information and large space cost. This method has some robustness to the gray variation, rotation and translation of image and improves the recognition rate.