Fundamentals of digital image processing
Fundamentals of digital image processing
Pattern Classification (2nd Edition)
Pattern Classification (2nd Edition)
Biometric User Authentication for IT Security: From Fundamentals to Handwriting (Advances in Information Security)
Template Matching Techniques in Computer Vision: Theory and Practice
Template Matching Techniques in Computer Vision: Theory and Practice
New Algorithms for Complex Fiber Image Recognition
AICI '09 Proceedings of the International Conference on Artificial Intelligence and Computational Intelligence
Hi-index | 0.00 |
One main goal of today's forensic textile fiber analysis is the classification of fiber traces. This can be achieved by forensic experts in manual analysis with the help of microscopy. However, this examination process, including an optical matching, is due to its manual nature very time consuming and therefore cost intensive. Considering that, we want to support forensic experts during the complex process of trace examination by adding signal processing methods for analysis and decision-making. In this paper we propose the introduction of computer-aided methods to speed up the process and make it more objective in terms of comparability of results and overall transparency. In our approach a 3D laser scanning microscope for surface measurement is utilized for contactless acquisition of physical textile fiber traces. Distinctive optical features are derived from digital images. Based on these distinguishing characteristics a matching process is introduced for the assignment to associated classes or categories. In this paper we utilize template matching methods in order to associate samples to different fiber types. The suitability of these methods is evaluated in context of forensic textile fiber authentication based on a first test-set composed of 45 samples divided in 3 classes. Our first experimental results show that fibers can be correctly assigned to their corresponding class based on template matching. However, the overall matching accuracy achieved here is only about 44% in an equally distributed 3-class-problem. The achieved matching results based on our newly designed feature set are momentarily not satisfying and thus require improvements, mainly by the design of new, discriminatory features.