In order to provide useful information for mobile robot, the target distinguish approach based on image matching needs to recognize image pattern. 2D Gabor filters were created by Gabor transform, which had excellent performance and didn′t need to segmente images catched by the single CCD camera. Gabor filters were robust to the various orientations and illumination of target images; They also satisfied the real time image processing. The multi-channel Gabor filters, with multi-frequencies and multi-angles, were used to convolute with images. The corresponding filter results included entirely information of images, from which the feature vectors of the images could be extracted. The method of classifying these feature vectors was SVM (support vector machine). SVM was a new method of machine learning developed by statistical learning theory. It resolved problems such as model select, over learning, non-linear, high dimension, etc. Experiment results indicate that the algorithm can reach up to high recognize rate. The algorithm can be applied in fields such as face recognition, robot vision localization and so on.
Ulrich I, Nourbakhsh I. Appearance-based place recognition for topological localization Proc IEEE Int Conf Rob Autom. Piscataway,NJ:IEEE,2000,2:1023-1029
Zhou Chao,Wei Yucheng, Tan Tieniu. Mobile robot self-localization based on global visual appearance feature Proc IEEE Int Conf Rob Autom. Piscataway,NJ:IEEE,2003,1:1271-1276
Jason J A, White M W,Chase T A,et al. Pattern analysis for autonomous vehicles with the region-and feature-based neural network:global self-localization and traffic sign recognition Proc IEEE Int Conf Rob Autom. Piscataway,NJ:IEEE,1996,4:3598-3604
Bovic A C, Clark M, Geisler W S. Multichannel texture analysis using localized spatial filters[J].IEEE Trans Pattern Anal Mach Intell.1990,12(1):55-73
Rui Y, Huang T S, Chang S F. Image retrieval:past, present, and future [J].Journal of Visual Communication and Image Representation.1999,10 (1):1-23