A Hybrid Adaptive Thresholding Method for Text with Halftone Pattern in Scanned Document Images

摘要

In this paper, a hybrid adaptive thresholding method for scanned document images containing text with halftone pattern is presented. The method is based on the topological feature and gray level statistics of those text with halftone pattern. Global histogram based thresholding methods often miss some halftone text after binarization, especially those close to background gray level. The proposed method first divides the document image into non overlap windows and extracts text characters as connect component in each window. The Euler number of each text character is then calculated and used as topological feature to identify halftone text character. After all the halftone text characters are identified, the document image is segmented into halftone text region and non-halftone text region. Each region is then binarized using its own pixel value statistics respectively. Comparing to global histogram based thresholding methods; the proposed method produced better binarization result on scanned document images containing both halftone text and non-halftone text.