Two-dimensional affine-invariant hashing defined over any two-dimensional convex domain and producing uniformly-distributed hash keys

摘要

A uniform distribution of affine invariants is produced for a plurality of one or more two-dimensional objects. Each of the two-dimensional objects is defined by a set of object points selected from an object feature domain. By selecting one or more five-point tuples of the object points, three of the object points divide the object feature domain into a region arrangement of seven regions while the fourth point of the four-point tuple lies in one of the seven regions and each of the four-point tuples further defines each of the seven regions as one of four non convex quadrilateral arrangements or one of three convex quadrilateral arrangements. A four-point tuple is said to belong to class I if the fourth point of the tuple resides in the I-th of the 7 regions defined by the first three points. A tagger identifies each of the four-point tuples as having one of the arrangements with one of the regions containing the fourth point of the four-point tuple. During a knowledge accumulation mode, and using either synthetically generated or real data, an equalizer accumulates knowledge about occupancy patterns incurred by four-point tuples belonging to each of the 7 classes and then derives the necessary remappings that will result in an expected uniform distribution over the range of invariants for all produced invariants.