Trademark Retrieval Algorithm Based On SIFT Feature

Trademarks are the important character of products, which are used to distinguish the type of goods. In modern society, trademarks play an important role. They are the symbols of certain brands. Generally, a trademark is formed by graphics, text, color, or a combination of these, and it is unique and distinctive. With the dramatic increase in the trademarks, it is an urgent problem in the modern society to perform image retrieval accurately and quickly.Content-based retrieval means that the search will analyze the actual contents of the image rather than the metadata (keywords, tags). The term 'content' in this context might refer to the global features, such as colors, shapes, textures, or other local information like key points that can be derived from the image itself. Content-based image retrieval involves many study fields, such as image analysis, image matching and image recognition.SIFT (scale invarious feature transform) was used as the local image feature for trademark image retrieval in the thesis. SIFT is a method of image feature extraction and description. The multi-scale space of trademark images is established through feature selection of the image scale, in which the features points are identified, and the edge response points and unstable points are deleted. And finally, the feature descriptors are extracted. SIFT features are local features of images, which has rich information for the mass characteristics of the database for fast and accurate matching, and by selecting the appropriate threshold, the reliable and stable feature matching points can be found. Experimental results show that the SIFT based trademark retrieval algorithm is perfect and robust to image scale, viewing angle changes, occlusion, noise, etc. To reduce the computation and improve the matching speed, this thesis also improved and optimized the SIFT operator.