一個新的鞋印識別及分類之方法

Abstract

本篇論文中我們提出了一個新的自動處理鞋印影像的方法。鞋印是犯罪現場中十分常見之線索，其最主要的特徵便是其紋理的方向性。論文中，我們利用影像之灰度伴隨矩陣(co-occurrence matrices)、方向矩陣(directional matrix)及傅立葉轉換(Fourier transform)來截取出鞋印紋理之方向性特徵以做為比對的依據。由於在前處理上使用主軸轉換(principal component transform)，本方法並不會因為影像之旋轉或位移等失真造成錯誤的結果。實驗結果顯示，本方法能夠有效的識別各式的鞋印。

In this thesis, we present a method for automatically classifying/recognizing the shoeprint images based on the outsole pattern. Shoeprints are distinctive patterns often found at crime scenes that can provide valuable forensic evidence. Directionality is the most obvious feature in these shoeprints. For extracting features corresponding to the directionality, co-occurrence matrices, Fourier transform, and a directional matrix are applied to the shoeprint image. And with the stage of principal component transform, the method is invariant to rotation and translation variance. Experiments of matching shoeprints are conducted on the database of 315 shoeprint images to demonstrate the performance of the method.