Seemingly similar 3D target recognition based on local wavelet-moment

When multi -view modeling method was used in the recognition of seemingly similar 3D target, local features extraction must be added to reinforce the recognition rates for the similar contours of targets. After analyzing the localization of wavelet moment, the result was gained that wavelet moment processed the localization in the radial interval only and was a global feature in the angle interval. Firstly, the prior information of target view were obtained by using the information sampling method, and then the target view were divided into several areas. Posterior information of these areas were estimated with Bayesian Posteriors Estimation, and the differences between the prior information and posterior information were compared and the most discriminative local region was acquired at last. Wavelet moment in this local region was calculated, which make it possible to get the local wavelet moment with the localization in the angle interval. Experimental results show that, compared with the traditional methods, this proposed method can get the better localization and distinguish seemingly similar target more effectively.