Abstract: Since inherent noise of image sensors influences the precision and reliability of numerous existing image registration algorithms severely, a robust registration approach to low SNR images is presented. By using the technique of the regional multi-direction gray difference threshold to extract feature points, and then adopting least trimmed square Hausdorff distance(LTS-HD) measure based point set matching method to acquire parameters of the affine transform, the algorithm realizes the registration of images to be registered. Experimental results indicate that the method has the preferable adaptability for noise disturbance and scene change.

Key words: Hausdorff distance, LTS-HD measure, fixed pattern noise, image registration