Persian Text Detection in Images of Natural Scenes

Among the information found in the image, textual data have particular importance because they are easily understandable by humans and computers, and provide the description of the content of an image. A new algorithm is presented in this study to detect Persian texts. The proposed system is comprised of connected-component labeling of candidate areas that become a mask by blending the Maximally Stable Extremal Regions and extracted edges. Finally, text lines are formed and texts appear in images by functional morphology and analyzing connected components of pruned non-textual areas. PersianTextNSI dataset is used to investigate results, which is formed by images of natural scenes of Persian texts. This collection includes 150 images. In order to evaluate the performance of the proposed system, the recalling rate was increased from %56.81 to %65.77 which indicates good performance of the system.