Digital Image Segmentation and On-Line Print Quality Diagnostics

Zuguang Xiao, Purdue University

Abstract

During the electrophotographic (EP) process for a modern laser printer, object-oriented halftoning is sometimes used which renders an input raster page with different halftone screen frequencies according to an object map; this approach can reduce the print artifacts for the smooth areas as well as preserve the fine details of a page. Object map can be directly extracted from the page description language (PDL), but most of the time, it is not correctly generated. For the first part of this thesis, we introduce a new object generation algorithm that generates an object map from scratch purely based on a raster image. The algorithm is intended for ASIC application. To achieve hardware friendliness and memory efficiency, the algorithm only buffers two strips of an image at a time for processing. A novel two-pass connected component algorithm is designed that runs through all the pixels in raster order, collect features and classify components on the fly, and recycle unused components to save memories for future strips. The algorithm is finally implemented as a C program. For 10 test pages, with the similar quality of object maps generated, the number of connected components used can be reduced by over 97% on average compared to the classic two-pass connected component which buffers a whole page of pixels. The novelty of the connected component algorithm used here for document segmentation can also be potentially used for wide variety of other applications. ^ The second part of the thesis proposes a new way to diagnose print quality. Compared to the traditional diagnostics of print quality which prints a specially designed test page to be examined by an expert or against a user manual, our proposed system could automatically diagnose a customer’s printer without any human interference. The system relies on scanning printouts from user’s printer. Print defects such as banding, streaking, etc. will be reflected on its scanned page and can be captured by comparing to its master image; the master image is the digitally generated original from which the page is printed. Once the print quality drops below a specified acceptance criteria level, the system can notify a user of the presence of print quality issues. Among so many print defects, color fading – caused by the low toner in the cartridge – is the focus of this work. Our image processing pipeline first uses a feature based image registration algorithm to align the scanned page with the master page spatially and then calculates the color difference of different color clusters between the scanned page and the master page. At last, it will predict which cartridge is depleted.^

Degree

Ph.D.

Advisors

Jan P. Allebach, Purdue University.

Subject Area

Electrical engineering

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS