A high-precision template localization algorithm using SIFT keypoints

High-precision localization is one of the important applications in the field of computer vision. In this paper a high-precision template localization algorithm based on SIFT (scale invariant feature transform) is presented. The proposed method is composed of three main steps. In the initial step the SIFT features are extracted. With these features the basic matching strategy and clustering method similar distance threshold (SDT) are investigated to match the keypoints between template and test images and eliminate the possibility of mismatch. Then iterative least square method (ILSM) is adopted to locate the template and improve the accuracy. Compared with the traditional template matching methods, the proposed method could enhance the robustness effectively, which ensures to give correct results, no matter the test image changes its scale, rotates itself or is covered partly. The localization accuracy reaches 0.1 pixels.