Color structure recovering in strong specular text regions

Highlight component is unavoidable in natural text images, which is caused by the effects of degradations, environment conditions, illumination and material of objects during the capture process. There are some techniques which could reduce or remove the specular regions but almost of them are effective only in simple cases which contain small and slight highlight areas. Strong specular zones, especially in the cases of the destroyed color structure are still challenges for the current approaches. In this paper, we introduce an effective method to remove strong highlight components which ruin the color structure in the natural text images. By analyzing tensor saliency map based on saliency magnitude and saturation information, specular and diffuse candidates are determined. Then, diffuse nominees are used to remove highlight components from specular candidates based on the painting technique. Our proposed method is evaluated using the standard dataset of the ICDAR 2005. The promising results have confirmed the effectiveness of the proposed method.