Doctor of Philosophy (PhD)
Electrical and Computer Engineering
First Committee Member
Number of Pages
The main objective for image compression is usually considered the minimization of storage space. However, as the need to frequently access images increases, it is becoming more important for people to process the compressed representation directly. In this work, the techniques that can be applied directly and efficiently to digital information encoded by a given compression algorithm are investigated. Lossless compression schemes and information processing algorithms for binary document images and text data are two closely related areas bridged together by the fast processing of coded data. The compressed domains, which have been addressed in this work, i.e., the ITU fax standards and JBIG standard, are two major schemes used for document compression. Based on ITU Group IV, a modified coding scheme, MG4, which explores the 2-dimensional correlation between scan lines, is developed. From the viewpoints of compression efficiency and processing flexibility of image operations, the MG4 coding principle and its feature-preserving behavior in the compressed domain are investigated and examined. Two popular coding schemes in the area of bi-level image compression, run-length and Group IV, are studied and compared with MG4 in the three aspects of compression complexity, compression ratio, and feasibility of compressed-domain algorithms. In particular, for the operations of connected component extraction, skew detection, and rotation, MG4 shows a significant speed advantage over conventional algorithms. Some useful techniques for processing the JBIG encoded images directly in the compressed domain, or concurrently while they are being decoded, are proposed and generalized; In the second part of this work, the possibility of facilitating image processing in the wavelet transform domain is investigated. The textured images can be distinguished from each other by examining their wavelet transforms. The basic idea is that highly textured regions can be segmented using feature vectors extracted from high frequency bands based on the observation that textured images have large energies in both high and middle frequencies while images in which the grey level varies smoothly are heavily dominated by the low-frequency channels in the wavelet transform domain. As a result, a new method is developed and implemented to detect textures and abnormalities existing in document images by using polynomial wavelets. Segmentation experiments indicate that this approach is superior to other traditional methods in terms of memory space and processing time.
Compressed; Document; Document Image; Domain; Image; Image Compression; Processing; Techniques; Wavelet Transform
University of Nevada, Las Vegas
If you are the rightful copyright holder of this dissertation or thesis and wish to have the full text removed from Digital Scholarship@UNLV, please submit a request to firstname.lastname@example.org and include clear identification of the work, preferably with URL.
Deng, Shulan, "Techniques for document image processing in compressed domain" (2008). UNLV Retrospective Theses & Dissertations. 2442.