Design of a Food Recognition System by Symmetrical Windows

• Main Authors: Yun-Fu Liang, 梁耘輔
• Other Authors: Wen-Fong Wang
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/95464002150838699991

碩士 === 國立雲林科技大學 === 資訊工程系碩士班 === 101 === Recently, a food recognition system using a SPIN descriptor has been proposed. The SPIN descriptor is extracted by circular windows with color histograms, and it has demonstrated good performance in recognizing rotated food images. However, one descriptor is not enough for distinguishing between certain foods, such as foods of a similar color. Therefore, we propose a food recognition system with various descriptors:bag-of-SURF, Haralick’s, and HSV. These descriptors are extracted from a food region with a circular window.Then, these descriptors are classified using an SVM(Support Vector Machine). In the experiments, we estimated descriptors with a circular window and a square window in our system and found that thedescriptor with the circular window had the best performancein our proposed system. It achieved a 98.66% recognition rate with the HSV-Haralick’s descriptor. In addition, when the descriptor was reduced to half-dimension, it achieved a 98% recognition rate and a 48.24% increase in speed in the stage of feature extraction.