Abstract: For the problem of “semantic gap” in image retrieval, taking the trademark images as a research object, the natural, abstract and deriving meanings of the Latin letters are analyzed, and a set of “Latin letters descriptors” is constructed. At the same time, the origin, composition and features of the trademarks are analyzed. A method of using “Latin letters descriptors” to describe and encode the trademarks is proposed. A series of search results and the statistical data show that this method is of some practical significance.

Key words: computer application, image recognition, image retrieval, trademark retrieval, image feature description