Abstract: With the rapid development of information science and technology, a large number of digital images are generated every day on the internet, leading to the demand for image retrieval tools in different fields. Because of the increasing amount of data, the traditional key word based image retrieval method is not able to meet the demand anymore. Therefore, it is becoming more and more urgent to develop a new image retrieval technology. In many image retrieval algorithms, the content-based image retrieval algorithm has been attracting the attention of the researchers in recent years. Different from the traditional way of retrieving by image name or other key words, the content-based image retrieval algorithm uses low-level image features, e.g., color feature, texture feature, and shape feature, for image retrieval. Generally, the development of content-based image retrieval has experienced a process from single feature image retrieval to multi features image retrieval. Compared with global features, local features have advantages in dealing with occlusion, clutter and adaptation to partial appearance changes. As a result, the trend of CBIR research has shifted from global features to local features, which has been proven to be a practical way to deal with the semantic gap. In this paper, a local multi-feature image retrieval algorithm based on discrete Tchebichef orthogonal polynomial and Fourier Mellin moment is proposed. By orthogonally transforming and multiresolution reordering, the texture, color and shape features of images are extracted from the transform domain, and the image feature with strong distinguishing ability is generated. By considering the invariance of Fourier Mellin moment to rotation transformation, this proposed method can perform well in dealing with the image where the rotation transformation and the translation are performed. Finally, the multiple data sets are utilized to carry out the retrieval experiment and the experimental results are compared and analyzed.