Image registration for supper resolution reconstruction plays a paramount part in video-based criminal investigation and judicia1 forensics wherein the re1ated video materials taken from public surveillance system, mobile phones, and so forth are generally very poor in quality, and thus most of the popular super resolution image reconstruction methods fail to obtain satisfactory results

In image registration, we defined a new type of control point called dominant point to capture the local variation property of an image. The order relationship between the sorting dominant point sequences, respectively extracted from the reference image and the target image, matches the two images, and at the same time, their location relationship is adopted in projection kernel regression, a new representation for function approximation in KBHS, for coordinate mapping function approximation.