A Reliable Scene Matching Approach Based on the Fusion Feature of Correlation Peak

Abstract

In the scene matching system, difference between realtime image and reference image often makes the true matched point deviate from the maximum peak of correlation plane, which leads the conventional decision making method based on maximum peak("MPDM" for short)" to mismatch. Aiming at this problem, a new matching approach based on fusion feature of correlation peak is proposed in this paper. First, by analyzing the peak properties corresponding to the true matched point, fusion feature based on the relative height (RH) of peak, peak value ratio within 8-neighborhood of peak (NBR) and sharpness of peak (SP) is taken as matching criteria, then, by setting a proper threshold of indicating the permissible minimum difference among those chosen fusion features, the real-time images whose matching results are sensible to image alterations are discarded, which increase the credibility of matching result.