Automatic Electronic Book Construction from Scanned Page Images by Image Analysis Techniques

Abstract

Based on the use of image analysis techniques, a system for automatic construction and display of electronic book contents from scanned page images is proposed. First, the Hough transform is employed to estimate the skew angles of scanned page images, and adjust their orientations. The resulting image is then decomposed into basic blocks using moment-preserving thresholding and region growing techniques. An automatic method is proposed next to estimate the threshold parameters for the size and point density features of the basic blocks. These parameters are used to classify basic blocks into three types: text, graphic, and line blocks. Furthermore, a systematic repetitive pattern detection method is proposed to find common parts among different page images, which improves book content compression. For page images with pictorial backgrounds, also proposed is a method for extracting texts on pictorial backgrounds to gain more text information. Finally, a user-friendly interface is designed for displaying the electronic book content according to a proposed book content organization structure.