Automatic segmentation and descreening of scanned color documents

Printing books-on-demand is a new technology that is revolutionizing the bo ok printing and publishing industry. One of the biggest bottlenecks in this process is the conversion of existing books into digital form. This typica lly involves digitization of original books through scanning, which is a sl ow and labor-intensive process. Careful attention must be paid to maintain the quality of the reproduced books and in particular of the images they co ntain. Halftoned image areas in the original books cause the most reproduct ion problems, as there is the potential that moire patterns may form when t hese image areas are re-screened. In order to avoid these moire patterns, i t is necessary to detect the image areas of the document and remove the scr een pattern present in those areas. In the past, we have presented techniqu es to perform these operations in the case of grayscale images. In this art icle, we extend these techniques to handle color images. We present efficie nt and robust techniques to segment a color document into halftone image ar eas, detect the presence and frequency of screen patterns in halftone areas and suppress the detected screens. Halftoned image areas are segmented by using a measure of image activity; image activity is low in text areas and high in halftoned areas. We use 2-D Fourier spectral analysis to identify t he screen frequencies present. The screens are then suppressed by low-pass filtering. Our technique speeds up the conversion process of books to digit al form, and overcomes quality problems in the reproduction of halftoned im ages.