Image registration based on magnitude segmentation
MetadataShow full item record
The proposed idea introduces a new method for image registration based on magnitude segmentation. The template, or reference image, and the image are both partitioned into different segments based on intensity value ranges. The cross-correlations of corresponding segments with the same intensity ranges in the template and image are computed, and then added up to result in the proposed magnitude segmented cross-correlation, which is shown to improve localization sharpness with increased immunity to noise. Paired segmentation, in which the difference of each pair of judicially chosen segments is considered instead of the individual segments, is shown to further enhance and sharpen cross-correlation mainlobes while lowering peak sidelobes. Simulation results confirm the validity of the proposed algorithms. Keywords: cross-correlation (CC), magnitude segmentation (MS), paired magnitude segmentation (PS), mainlobe to sidelobe ratio (MSR).