Matching points in poor edge information images

Pattern matching and image patches correspondence have been described in several papers. However, in most cases the results are obtained using images with high level of detail, in other words, images with useful edge information. This paper describes a method to find correspondences between images with very poor edge information -for instance a painting with a cloudless sky- and its application to reflectographic images mosaicing. Unlike other pattern matching techniques, the proposed method solves the issue of low levels of detail or lack of good information, both needed for the determination of the correspondences used by common likeness measure (cross-norm correlation), features correspondence (Harris, SUSAN) and object recognition methods. Thus, intensity-augmented ordinal measure method is used: a window-based noise-robust method to calculate matching points. This method is improved with our Point Structure Selection (PSS) procedure to select good correspondences and reject false positives.