| Summary: | 碩士 === 國立成功大學 === 電腦與通信工程研究所 === 104 === The smart technologies realized in smart devices for human assistances and the demands of recognition technologies increase daily in recent years. The applications of recognition are wide, including robots recognition, face recognition, fingerprint recognition and commodity recognition. This thesis proposed a commodity recognition system which combines SIFT, Bag of Words and SVM. Because of traditional SIFT processes image in the gray scale, by this way, the color information is missing. In image processing, color information is an important data for an image. Thus, a color-SIFT descriptor which is based on the traditional SIFT is proposed. Color-SIFT descriptor combines original 128 dims descriptor of SIFT with color information around the feature points. Then, the descriptor is binarized for hardware and transmission friendly. Through the color-SIFT, we can discard the unreliable matching in the color domain, and increase the robustness of matching tasks. In experimental results, the proposed system is compared with two other systems, prove it still has high recognition rate in different situations.
|
|---|
