Mathematical Morphology-Based Automatic Restoration and Segmentation for Degraded Machine -Printed Character Images

Abstract

This chapter presents a morphological approach (AutoS) for automatic segmentation with feature vector extraction of seriously degraded machine-printed character images. This approach consists of four modules. The first detects and segments natural pitch characters based on the vertical projection of their binary images. The second uses an algorithm based on the vertical projection and statistical analysis of coordinates to detect fragments in broken characters and merge them before the eventual segmentation. The third employs a morphological thickening algorithm on the binary image to locate the separating boundaries of overlapping characters. Finally, the fourth executes a morphological thinning algorithm and a segmentation cost calculation to determine the most appropriate coordinate at the image for dividing touching characters. By the automatic running of a suitable segmentation module for each problem, the AutoS approach has been robust, flexible and effective in obtaining useful and accurate feature vectors concerning degraded machine-printed character images.