Segregation of composite document images into textual and non-textual content

Abstract

Segregation of image into textual and graphical region is essential prior to feeding the document in optical character recognition. Text embedded in colored images, such as book covers, video frames, natural scene images and web images suffers variation in style, color, orientation, alignment, low-image contrast, degraded characters as well as multiplexed background. The proposed paper focuses on segregation of textual component in two junctures. Initially, textual blocks characterization is done followed by restoration of both textual and non-textual region into separate image sets. Textual component extraction is preceded by noise removal and color reduction which is followed by edge detection, detection of closely connected component which can be grouped together and identified as a separate blob. Basic morphological operation are used to restore the broken characters. Later on the textual and non-textual lobs are segregated into separate image sets, which can be further used for annotation, indexing, character recognition. Image in painting is also performed in order to restore the area in both the image sets from where the textual and non-textual blobs have been extracted in order to make the image sets appear like real-image sets. The proposed approach works irrespective of the font and language of the textual content in the image.