Research On Trademark Recognition Algorithm And Its Realization On Mobile Device

Logos are important marks that identify commodities and services, which are usually comprised of patterns and text. Logo recognition is a new computer vision technology which has significant value in the applications of logistics commodities surveillance, business information mining, logo protection and so on. Existing logo recognition algorithms mostly aimed to find the nearest point based on the distance between feature points, which ignore the spatial relationship of them. Based on existing works, this paper proposes a logo recognition algorithm based on the weighted distance retrieval and visual pattern mining, which can obtain preliminary result by distance weighting between the feature points. Then it constructs the spatial relationship of the matched points, retrieves the frequently occurring sub-graph structure, removes the mismatched points and obtains the final matching result.As the mobile devices are not restricted by the location, users can use them to recognize logos anytime anywhere. However, existing logo recognition methods are mostly designed for documents and advertisement videos. There are a few methods designed for mobile devices, some of which have high computing complexity and some of which have low recognition accuracy. Thus, we improve our proposed method for the effectiveness. Specifically, this paper also implements logo recognition on mobile devices based on motion estimation, SIFT features and the BOW model, and it introduces a new procedure framework for the logo recognition in the mobile internet by implementing the real-time feedback on the mobile devices. Experiments show that the presented algorithm and framework could implement logo recognition applications on mobile devices, which has satisfying effectiveness and accuracy.