A study of automatic pattern retrieval for striation toolmarks

• Main Author: 莊閔傑
• Other Authors: Che-Yen Wen
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/r9764d

碩士 === 中央警察大學 === 鑑識科學研究所 === 97 === A tool mark is one of the most important evidence found at crime scene. A tool mark is a mark on a soft material left by a harder material. Different types of tools, such as screwdrivers, pliers, bolt cutters and hammers, may create different types of tool marks. Among these tools, screwdrivers are the most popular hand tools at home and almost every family has more than one screwdriver in Taiwan. Therefore, tool marks left by screwdrivers (especially, slotted screwdrivers) are usually found at crime scene. Especially in burglary cases, striation tool marks made by slotted screwdrivers often can be found on doors and window frames. We can verify whether they were made by specific suspected tools, if we have any. If we don’t have specific suspected tools, we may keep the tool marks by photographs. With the photographs, we may find clues in related or similar cases. However, it is not easy to compare the tool marks by visual examination, especially in a case with a lot of photographs. In this paper, we propose a feature extraction and similarity measurement method for striation marks. This method includes four steps: image preprocessing, striation orientation detection, signature extraction and retrieval. Image preprocessing can normalize the image size and reduce illumination effect. Striation orientation detection is used for striation alignment. In the signature extraction step, we take intervals and widths of striations as features, and convert them into a 1-D sequence. Finally, we apply the Smith-Waterman algorithm to find the best feature code alignment, and use this alignment to measure the similarity between two striation marks. With the similarity measure, we can find a list of “similar” striation marks in the database. The listed marks will then be examined manually by visual inspection for final validation