Logo Recognition

Logo has been played a very important part in our daily life. Because different logos will have different information,there is a demand that automate logo recognition can be applied to various fields such as logistics transportation network, advertisement, E-commerce and so on. In this paper, a novel method for recognizing logos in natural images is presented.The difficulties lie in: (1) the logos in different images are highly varied in shape and location; (2) images will usually have several different kinds of logos; (3) the logo will be occluded by other objects or sometimes miss, which traditional methods usually fail;4) the condition when we take photos; 5)the complexity of the log. Though much effort has been paid, unfortunately currently there is no mature algorithm that can tackle with all the difficulties above.In this paper, a learning-based logo recognition method is proposed to detect and classify the logos in natural image which will confront all the above difficulties. First, the SIFT matching solution is applied to detect the interest region in images and extract the discriminate features, which is used as the signature of each logo. During this step we will also process with these features which will eliminate all the unrobust and unnecessary feature key points. Second, several different match algorithms are proposed. The first one is normal match algorithm; the second one is the approximate nearest neighbor searching strategy which is build up by formulating the separate data into tree-based structure, for the purpose of efficient matching, the last one is pyramid matching which will formulate all the date into vocabulary-guide structure and compute the similarity between two data sets. Finally, post process which utilize the spatial information is processed to improve the accurately of this algorithm.Promising results have been obtained in robustly classifying one thousand logos in the images captured and this algorithm has been used for commercial.