Currently, optical devices, such as microscopes and CCD cameras, are mainly utiliz ed f or identif ication of bullets and tool mark s in the f ield of f orensic science. While these optical methods are easily manageable and ef f ective, they are under great inf luence of illumination condition. Besides these appearance-based approaches, we can utiliz e 3D geometric data of tool mark s that are f ree f rom lighting condition. Nevertheless, a perf ect correspondence of two striation patterns is

more »

... rely encountered, even if the two bullets have been f ired f rom the same f irearm. Theref ore, more robust methods are required. I n this study, we propose a two-stage comparison method f ocused on 3D geometric. At f irst, we have aligned global shapes and evaluated a global shape similarity. Then small elevations are compared by neural network s; that is local matching. I n this stage, rendered images under a unif ied lighting condition are utiliz ed. By using 2-stage comparison, we have developed a robust method that searches f or similar striation patterns.