Binarization by Local K-means Clustering for Korean Text Extraction

Text information in a natural scene is very useful and important for understanding images. Detection and extraction of text information in such natural images have been used in many applications. Many conditions of natural scene make the problem of text segmentation quite intractable. In this paper, an effective method for the segmentation and binarization of Korean texts from signboard images is proposed, which is robust in blurred images, uneven illumination and strong boundary text images. The proposed approach is based on local K-means clustering on separate words in text region. Firstly, detected text region is divided into local areas with relatively uniform illumination, and then using 3-means clustering with Euclidean distance has been applied to segment text from the background. By dividing the region of interest into local areas, the effect of uneven lighting has been minimized. The comparison with Otsu´s method and 2-means clustering based on intensity will be representation in some metrics. Natural images from the test database, collected from mobile devices, are used in the experiment and the results show the performance of the proposed method.