Radon-Fourier Similar Invariant Feature Extraction

Image recognition technology is an important field of computer vision. It have a wide range application in video surveillance, human-computer interaction, traffic monitoring, behavior recognition, automatic navigation and many other aspects. In image target recognition technology, edge can be counted low-level image features and for human vision, that is the most intuitive. The edges preserving some important characteristics, so it is generally possible to detect the contents contained from the image edge recognition result. For the calculation of the visual, it have to use some other bottom level features, which is very important is that shape. Intuitively, the shape may be seen as a strong edge point, it can characterized the shape of its contour curve. For example:the corner point is those rapid changes in direction of the curve contour point and the point line is always tends to be the slow changes in the contour corresponding curve. So these special points play a very big role for shape description and shape matching, use them to compressed data with which to express the shape of important information. Currently, there have many important applications of target feature extraction and recognition method. According to target recognition feature extraction and recognition method, can roughly divided the method into the method based on contour, the method based on region. Based on the shape of the contour recognition method, simple implementation and calculation is limited by the length of the outline border, but this shape recognition method to extract only the boundary characteristics and to realize the important premise of the method is able to complete to extract the contour. And it only can use in a single contour or for simple outline, but in many cases, to identify the target shape is often with a lot of empty(like the Chinese characters’‘轮”,”廓”,”单”etc.). Because of mixed conditions can’t get the single, simple outline, so the shape recognition rate and application scope is limited. The shape recognition method based on contour area is contained much image information, so the extracted invariant features can better reflect all features of the object shape. For mixed cases that can’t get the single, also have very good recognition. In this paper, the target shape feature extraction and recognition method conducted a in-depth research, in the further study of classic recognition algorithm, Radon-Fourier shape feature extraction method is proposed with good performance and wide application scope. The main work is as follows:first, this paper systematically studied the image recognition algorithm, which based on contour feature and shape feature, shape feature extraction and recognition algorithm. Provides detailed information on chain code, Poisson equation shape feature extraction method, and respectively analyses the advantages and disadvantages. Mainly study the shape feature extraction method based on Poisson equation, and analyzed the advantages and disadvantages. Then, the research is mainly focused on the Radon transform and Fourier descriptor, based on the research of these two methods with considering the advantages of them, present a regional characteristics from multiple perspectives Radon-Fourier shape feature extraction method. Experimental results show this method have simple calculation, that means there have no need to contour extraction for the target object. And the experimental present that this method have very impressive recognition to MPEG-7image data base.