关键词: 模板匹配, 圆投影变换, 高斯混合模型, 聚类特征, 光照鲁棒

Abstract: To overcome the information loss in radical-projection-transform template matching algorithm, the proposed method introduces the clustering model parameters robust to linear illumination. The method eliminates the influence of illumination by linear contrast stretch, and takes the Gaussian mixture model clustering result as the template image feature. The matching errors are obtained by once iterative, and the iterative calculation can be implemented by using a simple look-up table. In searching stage, down sampling matching strategy is used to reduce computation cost and the average error of this stage is set to an adaptive threshold for precise matching. Then, the pixel-wise matching is operated in the neighborhood of the down sampling point whose error is less than the threshold, and finally the point of minimum error is set to be the target position. The test shows that the proposed method has better positioning accuracy and reliability than the normalized correlation based radical-projection-transform template matching algorithm.

Key words: template matching, Radical Projection Transform（RPT）, Gaussian mixture model, clustering feature, illumination robustness