关键词: 特征选择, MI贪婪最优算法, 支持向量机（SVMs）

Abstract: This paper focuses on the MI-based image feature selection method[1]，briefly describes SVMs category principle and design[2].The paper proposes the MI greedy optimal algorithm，transforms high-dimensional data processing into a one-dimensional data processing，simplifies the difficulty of the operation and increases the speed and accuracy of classification.The results show that，the eight classification，more than 1，000 classified images are better dealt with by the MI greedy optimal algorithm than the traditional classification algorithm.

Key words: feature selection, MI greedy optimal algorithm, Support Vector Machines（SVMs）