%0 Conference Proceedings %F dvmmPub86 %A Jaimes, Alejandro %A Mintzer, Fred %A Ravishankar Rao, A. %A Thompson, Gerhard %T Segmentation and Automatic Descreening of Scanned Documents %B IS&T/SPIE Color Imaging: Device-Independent Color, Color Hardcopy, and Graphic Arts IV %C San Jose, CA %X The advent of digital libraries and advances in scanning and printing technologies has opened up new possibilities for reproduction of printed materials. In one application, IBM printing equipment and software are being used for an on-demand book printing service. Books that are out of print are being scanned and stored in a digital library for on-demand printing. When a book is ordered, in any quantity, it is printed directly from the digital library. In this paper we address image processing and work flow issues in creating a solution for this on-demand printing problem. Once an image is generated by scanning, the following steps are taken: segmentation of the image into text and photograph areas; image skew angle detection; detection and determination of screen frequencies in photograph areas; removal of photograph screens (descreening). One of the major challenges encountered in terms of quality is the supresion of moire patterns or artifacts that appear in photographs when the book is printed from the library. When photographs in the original books are offset printed, a screening process takes place. When those screened photographs in the book are scanned and printed from the digital library moire effects are very likely to appear. These effects can be eliminated if the screen in the scanned photograph is removed. We present novel algorithms which are robust and computationally efficient for this processing, placing special emphasis on the effective detection and removal of screens in photograph areas. Effective descreening is a crucial step in removing the moire effects that appear in the photographs when printed. For segmentation we use simple a technique based on gradient density which yields different values for photograph and text areas. Photographic areas typically exhibit high gradient densities and text areas exhibit low gradient density. From the segmentation we obtain bounding boxes for the different areas. The orientation of text lines is used to determine the orientation or skew angle of the scanned page. It is necessary to determine the skew angle to compensate for any rotation that occurs during scanning, in order to accurately find screen frequencies. Screen patterns in the photograph areas are then detected by applying a Fast Correlation function in both the horizontal and vertical directions. Since the function acts as an accumulator in each dimension, this yields two one dimensional vectors for which dominant frequencies are found by analysis of the corresponding Fourier power spectra. Removal of the screen is then performed by applying a smoothing filter chosen according to the detected frequencies in the horizontal and vertical directions. Once the photographs are descreened, they can be enhanced to improve their contrast and sharpness. We demonstrate the robustness of our approach by providing quantitative performance measures and applying them under different noise conditions in the detection of the halftone frequency and in the removal of screens produced by different halftone patterns. The use of these image processing techniques in conjunction with operator review provide an acceptable solution to the on-demand printing problem described above %U http://www.ee.columbia.edu/dvmm/publications/99/spie-des.pdf %8 January %D 1999