Multi-stage tattoo matching techniques

A system comprising:

• one or more processors;
• and
• one or more computer-readable media including instructions that, when executed by the one or more processors, cause performance of operations comprising:
• obtaining a search template associated with a search tattoo image, the search template comprising (i) the search tattoo image, (i) a plurality of search feature points, (iii) a search local descriptor for each search feature point within the plurality of search feature points;
• obtaining a reference template associated with a reference tattoo image that has previously been compared to the search template, the reference template comprising (i) a plurality of reference feature points, (ii) a reference local descriptor for each reference feature point, and (iii) a group of mated feature point pairs that each includes a reference feature point from among the plurality of reference feature points and corresponding matching search feature point;
• determining a mated region within the reference template based at least on respective locations of the reference feature points of the group of mated feature point pairs within the reference template;
• computing a plurality of transformation parameters based on (i) the search feature points of the group of mated feature point pairs that are located within the mated region of the search template and (ii) the reference feature points of the group of mated feature point pairs that are located within the mated region of the reference template;
• remapping, using the plurality of transformation parameters, the reference feature points within the mated region of the reference template onto the search tattoo image to generate a set of remapped search feature points;
• generating a respective remapped search local descriptor for each search feature point included within the set of remapped search feature points;
• comparing each of the remapped search local descriptors to reference local descriptors of reference feature points that (i) are not included in the group of mated feature point pairs and (ii) are located within the mated region of the reference template;
• and
• computing a match score between the search tattoo image and the reference tattoo image based at least on comparing each of the remapped search local descriptors to the reference local descriptors of the reference feature points.

One or more non-transitory computer-readable media having instructions stored thereon that, when executed by one or more processors, causing performance of operations comprising:

• obtaining a search template associated with a search tattoo image, the search template comprising (i) the search tattoo image, (i) a plurality of search feature points, (iii) a search local descriptor for each search feature point within the plurality of search feature points;
• obtaining a reference template associated with a reference tattoo image that has previously been compared to the search template, the reference template comprising (i) a plurality of reference feature points, (ii) a reference local descriptor for each reference feature point, and (iii) a group of mated feature point pairs that each includes a reference feature point from among the plurality of reference feature points and corresponding matching search feature point;
• determining a mated region within the reference template based at least on respective locations of the reference feature points of the group of mated feature point pairs within the reference template;
• computing a plurality of transformation parameters based on (i) the search feature points of the group of mated feature point pairs that are located within the mated region of the search template and (ii) the reference feature points of the group of mated feature point pairs that are located within the mated region of the reference template;
• remapping, using the plurality of transformation parameters, the reference feature points within the mated region of the reference template onto the search tattoo image to generate a set of remapped search feature points;
• generating a respective remapped search local descriptor for each search feature point included within the set of remapped search feature points;
• comparing each of the remapped search local descriptors to reference local descriptors of reference feature points that (i) are not included in the group of mated feature point pairs and (ii) are located within the mated region of the reference template;
• and
• computing a match score between the search tattoo image and the reference tattoo image based at least on comparing each of the remapped search local descriptors to the reference local descriptors of the reference feature points.