Feature point based text detection in signboard images
Chien-Cheng Lee, Shang-Fei Shen · 2016
This paper presents a method of using feature points to locate text area for signboards on street view images. The FAST corner detection was applied for the first step. FAST corner detection is fast and stable enough to retrieve potential text regions on street view images. The characteristics of each feature point color space were used to compute the color histogram and related information. For the second step, we used a gravity clustering method to find clusters of text area on signboard images and got the possible positions of the text area. For the third step, the distribution density was estimated and the average distance of feature points was calculated on the possible text area. The average distance was used to build text pattern regions. These regions were processed by the following steps: morphological closing, image binarization, and minimum bounding box finding to obtain a complete text region. Experimental results have shown the advantages and effectiveness of the proposed method in the text detection in the signboard images.