Fast correlation matching based on fast Fourier transform and integral image

YIN Song-feng,WANG Yi-cheng,CAO Liang-cai,JIN Guo-fan,LING Yong-shun
DOI: https://doi.org/10.3788/gzxb20103912.2246
2010-01-01
Abstract:In order to overcome the computation complexity of traditional correlation matching algorithm, a fast correlation matching algorithm is proposed based on fast Fourier transform and integral image. By analyzing and simplifying the equation of correlation coefficient, fast Fourier transform is employed to calculate the correlation between the template image and the base image, and integral image is employed to compute the window integral of the gray value of the base image and its energy. The computation complexity of the proposed algorithm is approximately independent on the size of the template image and thus is greatly decreased, without changing the results of traditional correlation matching. Experimental results for image matching under different illumination conditions and object tracking in complex situations show that the proposed algorithm is robust to interferences, and its implementation efficiency is 1 to 2 order times faster than traditional correlation matching algorithm.
What problem does this paper attempt to address?