To improve real-time performance of image registration and enhance robust performance of target tracking under the complex scene, a novel template-matching method based on image region texture is presented. Binary texture matrixes of horizontal and vertical direction were constructed according to gray direction of neighborhood pixels, and then the similarity measurement criterion was defined to measure correlation credibility between binary texture matrixes of each dimension. The correlation credibility of appointed object was obtained through combination of two-direction registration result at last. Experiments prove that the proposed algorithm is robust for image illumination variance and target partial occlusion.