Template matching based on deformation diversity similarity and scale filtering

To solve the problem that deformation diversity similarity (DDIS) has poor matching effects on rotation and multi-scale objects, a more robust template matching algorithm is proposed to deal with such conditions. The location distance of the nearest neighbor pairs is calculated by strengthening the constraint of the apparent features of the object on the location distance in polar coordinate system. The new confidence map composed of the DDIS response values is obtained and take the position with the highest values as the object center. The scale image pyramid is built according to the template to train the scale filter, and the scale filter is carried out on target image to determine the object scale. Experimental results show that the template matching effect of the proposed algorithm is better than DDIS, the new calculation method of location distance can effectively improve the robustness for rotating object and the scale filter can achieve the adaptive scale of object. Combining those two factors can further improve the matching accuracy.