Image matching based on geometric feature of edges and the correlation in frequency domain

The new method of image matching based on feature extraction and transform domain correlation is proposed in this paper. By extracting long edge information from image and analyzing its geometric feature, the edge directionangle curve can be constucted as an important parameter for image matching. By calculating difference and comparing with relative chain codes of two images’ edge direction curves, the possible rotation angle between the two images will be obtained. After rotating images and transforming them with 2D Fourier, the phase correlation will be gained in frequency domain and the maximum of the correlation coefficient will be found in that position which indicates the displacement of two images. So the two images can be matched accurately. Experimental results prove that this matching method is accurate, fast and automatic, it decreases the disturbance caused by artificial factors and could achieve a satisfactory matching between the two images with low grey difference and noise.