| Summary: | 碩士 === 國立交通大學 === 資訊科學與工程研究所 === 100 === Color recognition is an important issue in automatic vehicle surveillance. The vehicle color is a critical feature to help identify cars. In this thesis, we propose a novel approach for vehicle color classification. Firstly, we use image segmentation algorithm to divide image into regions and extract the vehicle shell part from them. Since the light reflection will influence, the broken regions on car shell after image segmentation, we can’t find a complete car shell well in images. Thus, we use the specular-to-diffuse mechanism to separate specular component and diffuse component and reduce the light reflection influence. Finally, we extract the vehicle shell region and calculate the dominant color of the shell region to classify the vehicle color. We download the 7 kinds of different color vehicle images from Internet web sites. The experimental results demonstrate good performance and thus show the effectiveness of the proposed schemes.
|
|---|
