关键词: 特征提取, 自优化, 改进算法, 识别率

Abstract: The feature extraction algorithms such as Color moment, Hu moment, Zernike moment and Wavelet moment are analyzed and a size feature extraction algorithm is improved. According to the problems that a single feature extraction algorithm for picking up the feature information is not comprehensive, and its recognition rate for the samples is very low. An improved feature extraction algorithm which is composed of five algorithms through feature distance from the self-optimizing combination is proposed. The algorithm formula and flow of execution of this new algorithm are introduced, and the feature library of project is established. Selecting some easily confused fruits to do identification experiment, and the results show that using this improvement feature extraction algorithm is superior to single feature extraction algorithm in recognition rate. The average time of recognition is slightly extended, but it can meet the demands of real time identification very well.

Key words: feature extraction, self-optimizing, improved algorithm, recognition rate