An edge-based method for aligning differently exposed images

This paper presents an edge-based method for aligning a sequence of differently exposed images. The edge focusing is employed in Canny detector to accurately extract structural boundaries without weak or spurious edges. Two criteria for selecting thresholds of Canny detector are proposed so that edge detection is independent of image intensities. A histogram-based matching is developed to fast estimate transform parameters in D² space instead of D³ space. Moreover, the coarse-to-fine technique by Gaussian Pyramid is employed to further reduce search space, and progressive image alignment is adopted to improve accuracy. Experimental results demonstrate that this method is accurate to 2 pixels.