Text localization based on the Discrete Shearlet Transform

This paper proposes a new method to locate the text regions from images with complex background based on the Discrete Shearlet Transform. Text localization is an important step of text extraction to obtain the useful information from images. It is now widely acknowledged that traditional wavelets are not very effective in dealing with multi-dimensional signals containing distributed discontinuities. Shearlets not only possess the main features of wavelets such as multi-scale, multi-direction and time-frequency localization, but also offers a high degree of directionality and anisotropy. Meanwhile, compared with Nonsubsampled Contourlet Transform (NSCT), Shearlets has obvious superiority in flexibility on directional selectivity and lower computing complexity. The proposed method applies the Discrete Shearlet Transform on an image to decompose it into set of directional subbands with texture details captured in different orientations and scales. Then, the binarization with dynamic thresholding is applied to these subbands so as to filter out the background and recognize the edges in multiple directions. Morphological operations are carried out on these binary images and connect the edges together in these images. At each scale, text regions are obtained with the help of the logical AND operator on all binary subband images; and the final text regions are achieved through voting-decision among different scales. This method of text localization is applied on different samples of images and the experiment results have a good performance.