This research addresses issues of image registration, texture image segmentation, and quantification of minor differences between texture patterns, for example in medical images. The premise behind the outlined algorithms is that 2-D image registration cannot counteraffect 3-D positioning differences. The problem of precise image registration is surmounted by comparing regions, in a manner similar to analysis done by medical experts. The criteria for region identification include location and intensity characteristics. The selection procedure is divided into: (i) selection of control points in two images (external markers and/or selected internal points that have unique characteristics) to determine locations of centers of regions to be compared, and (ii) determination of region extent by image segmentation, thus ensuring that the comparison is carried out between statistically homogeneous regions. Hierarchical region splitting is proposed as a means of achieving region size sensitive to subtle intensity changes. Both intensity statistics and texture measurements are included in region comparison procedure and decision making about a region status. The results of these basic investigations should be valuable in attacking the problem of early breast cancer detection based on comparison of different screenings of the same patient.
