Automatic date imprint extraction from natural images

Date imprint in a natural image is one of valuable information for indexing and retrieval. In this paper, we propose a method for extracting the date imprints from the complex backgrounds in color photographs. Firstly, a two-step color-based approach is applied, where the R color component contrast is enhanced to locate a candidate region in each of the four corner parts. Secondly, the located four regions are binarized with the histogram of the enhanced R component. The edge maps of these four areas drawn from R component images are also in consideration to clean up non-date pixels. This color-edge-based approach outperforms the ones only based on edge or texture which are usually used in video or image text detection. For the third step, projection and heuristics methods are deployed to fulfill a coarse-to-fine method to refine the location. Finally, the date characters can be obtained clearly by projection, morphologic and edge analyzing approaches. Experiments show that our approaches can effectively extract the date imprints for further recognition.