Current systems of colorful trademark images retrieval mostly rely upon single feature which leads to lower retrieval accuracy. Therefore colorful trademark images retrieval based on multi-feature combination and user feedback is studied in the paper and an experimental retrieval system is built. Color moments and shape-region descriptors can be extracted as features of colorful trademark images. Gaussian normalization is used to normalize and combine different features. Absolute Euclidean distance similarity algorithm is applied to retrieve colorful trademark images initially. In addition, the experimental system adopts user feedback module through which users can estimate initial results, then adjust the weights needed and retrieve again. Experimental results display that the retrieval results obtained by multi-feature combination are much better than the results obtained by single feature, and weights adjusting by user feedback can retrieve better results and achieve higher accuracy.
Bitte loggen Sie sich ein, um Zugang zu diesem Inhalt zu erhalten
