Research On Cartridge Case Image Registration And Fusion

In recent years, the analysis of firearm cartridge case (cartridge case in short hereafter) marks based on image processing and pattern recognition is developing very fast and plays an important role in the field of forensic science. Due to the limitations of optical system and the depth of firing pin impression, it's very difficult to capture one image with both informative and clear details at the very limited area of firing pin impression and the overall structure of cartridge case bottom. Therefore, two kinds of images can be first obtained and then the information can be supplemented by image registration and fusion method, which facilitates experts'analysis and the following computer processing.This paper begins with the concepts and popular methods of image registration and fusion, and then analyzes its current research situation. The study emphasis in this paper is on image registration and two registration methods are explored.Method 1: Cartridge case image registration based on integration mechanism and adaptive regulation is categorized as feature-based method. Because of the similar texture of cartridge case, the original feature-based registration has incorrect matches, and the traditional method of removing incorrect matches bears unsatisfactory results. Therefore, according to the character of cartridge case images, graph transformation matching, angle or scale constraint using adaptive K-means clustering are respectively applied to remove incorrect matches. In order to achieve the complementary advantages, voting mechanism is applied to integrate them; meanwhile, genetic algorithm is employed to combine parameters, making it possible to adaptively choose to integrate and registration results are optimized based on different images. Method 2: Cartridge case image registration based on Fourier-Mellin translation iscategorized as area-based method. It finds the location of optimal registration by log-polar coordinates and Fourier translation theory according to the value of phase correlation. After the above two registrations, the registration results are refined. Finally, the consistent images are obtained by pixel-level fusion.By experimenting the cartridge case image sets, the integration mechanism and adaptive regulation are proved effective in removing incorrect matching, and the performance of two registration methods are analyzed. In addition, the resultant image after fusion contains more information as compared with any original ones, which in turn facilitates the following analysis of cartridge case marks and impression.