Non-linear quantization and similarity matching methods for edge histogram bins

摘要

PURPOSE: A method for calculating non-linear quantization and a similarity of an edge histogram bin is provided to use a non-linear quantizer, and to encode a histogram bin value generated by 5 directional edge information, so as to reduce a bit amount necessary for storing and improve a similarity calculation speed for search. CONSTITUTION: Images are divided into 4x4 areas(S201). A directional edge histogram is generated(S203). A non-linear quantization table is calculated(S205). The calculated table is stored in a database(S207). An index is normalized(S209). A global and a semi-global histograms are generated(S211). And data are matched(S213). When generating the directional edges, features are extracted by the unit of block for the images divided into 4x4 areas, that is, 16 areas thereby generating 5 edges. The images are divided into sub-images that are not overlapped, and the edge histogram for each sub-image is generated as in the step of (S203).