Finding Optimum Similar Regions

In this paper we present a novel rotation- and affine-invariant image regions detector, coined OSR (Optimum Similar Regions). The two steps of OSR is: first, segment an image with thresholds based on graph theory, these regions of the segmentation results are so called similar regions. Second, the subset of similar regions is confirmed according to the regulation of the similar region’s cardinality. The paper presents experimental results on a standard evaluation set, as well as on application of wide baseline image matching. Both of them show OSR’s strong performance.