The work aims to propose a rapid and effective trademark registration similarity inspection method to solve the problem of low retrieval accuracy induced by missing detection and wrong detection in current SIFT-based trademark retrieval system. Firstly, SIFT was used to create the scale space and detect the key points of trademark feature. Then, the main direction of the key points was determined to obtain the image shape characteristic descriptors with rotation, scaling, translation and view change invariance. Subsequently, according to the color and spatial location of pixels and their neighbors, an improved color correlation was defined to effectively avoid the possibility that different trademarks might have similar color features. A weight factor was applied to different colors, and thus a color feature reflecting the color space correlation and color arrangement density was obtained. Then, SIFT and color correlation feature vectors were weighted and combined, and the weights were changed according to the dominant features in the actual process. Finally, according to the weighted combination features, Mahalanobis distance was introduced to measure the similarity between query trademarks and database trademarks, and the retrieved trademarks were outputted. The experimental results showed that, compared with the current advanced trademark retrieval system, the proposed algorithm had higher retrieval accuracy and efficiency. The proposed algorithm has good retrieval accuracy and robustness which has certain practical value in trademark registration and other fields.