PI: Shahram Latifi
Affiliation: Electrical and Computer Engineering Department, UNLV
Contact Information
Shahram Latifi, Professor
Electrical and Computer Engineering Dept.
University of Nevada, Las Vegas
Las Vegas, NV 89154-4026
Phone: (702) 895-4016
Fax : (702) 895-4075
Email: latifi@ee.unlv.edu
List of Supported Students and Staff (optional)
Dr. Emma Regentova, Research Scientist
Mr. Dongsheng Yao, M.S. Student
Ms. Leslie Sabo, Undergraduate Student
Project Award Information
Award Number: IIS-9616206
Duration: 09/01/2000 -- 08/31/2001.
Title: Direct Processing of Compressed Data
Keywords
ITU Group 3/4, JBIG, Wavelet transform, Document Image Analysis, Image Segmentation.
Progress in computer and communication technologies allows a large volume of digital information to be exchanged and archived. While the need for data compression is evident, many operations may have to be performed on the uncompressed data.
This research investigates the following:
· develop new algorithms for processing compressed data without fully decompressing them,
· develop new compression algorithms that allow given operations to be rapidly performed on compressed data,
· use wavelet transforms for document image analysis, specifically document image segmentation.
Publications and
Theses in the Past Year
· Deng S, Latifi S and Regentova E, “Document Segmentation Using Polynomial Spline-Wavelets”, Journal on Pattern Recognition, to appear.
· S. Deng, S. Latifi and E. Regentova, ``The Effectiveness of Polynomial Wavelets in Text and Image Segmentation", Document Recognition and Retrieval VII, Proceedings of SPIE, pp. 259-266, San Jose, January 2000.
·
Regentova
E, Latifi S and Deng S, “Connected Components Detection for ITU Group3/4 Coded
Document Images”, submitted to the IEEE
Transactions on PAMI, September
2000.
· Regentova E, Latifi S and Deng S, “Image Similarity Estimation by Processing Compressed Data”, to appear in the Journal of Image and Vision Computing.
·
Regentova
E, Latifi S, Yao D, Deng S, Zhao J,
“Compressed-Domain Techniques for JBIG-Encoded Documents Images”,
submitted to IJDAR, May 2000
·
Regentova E, Latifi
S, and Deng S, Wavelet-based Techniques for Image Similarity Estimation, ITCC 2000, LV, NV, March 26-29, 2000.
·
Deng S, Latifi S
and Regentova E, The Effectiveness of Polynomial Wavelets in Text and Image
Segmentation, IST/SPIE's 12th Annual International Symposium, Electronic
Imaging Science 2000, 23 – 28 January 2000, San Jose, California.
· Dongsheng Yao, Emma Regentova and Shahram Latifi, The Wavelet Probing For Normalized Cuts Framework, submitted to CISST'2001, Las Vegas, July, 2001.
· Dongsheng Yao, Emma Regentova and Shahram Latifi, An Experiment on Wavelet Domain Segmentation using Normalized Cuts Framewor, submitted to Computer Vision and Image Understanding.
· Shaojie Wu, Shahram latifi " Spatial Image Denoising in Discrete Wavelet Transform domain" submitted to CISST'2001, Las Vegas, July, 2001.
· Shulan Deng, “Techniques for Document Image Processing in Compressed Domain”, December 2000.
· Shaojie Wu, “Image Interpolation in the Compressed Domain”, Master’s Thesis, May 2001.
· Dongsheng Yao, “Document Segmenation using Wavelets”, Master’s Thesis, May 2001.
Demos:
· http://www.ee.unlv.edu/~regent/jbig_segm.html; http://www.ee.unlv.edu/~dyao/DOC/seg.html
i) Research Experiences for Undergraduates
We have been able to support two undergraduate students- June Light and Quentin Slate in the past year. This year we have involved a senior undergraduate- Leslie Sabo. These students have been working closely
with our NSF research team and plan to publish their technical work in student-oriented technical forums such as the IEEE Potential.
ii) Int'l Conf. on Information Technology (ITCC) http://www.cs.clemson.edu/~srimani/itcc2001/cfp.html
In an effort to broaden research activities on data compression and document analysis, the PI founded an international conference on information technology in March 2000. Sponsored by the IEEE Computer Society, this conference focuses on exploring techniques that utilize the synergy arising from combing
coding and computing. In its second year, the ITCC has
attracted many researchers and will be held on April 2-4 in Las Vegas. The PI
is the General Chair for this conference.
iii)
Information Technology Lab
As one of the spin-offs of our NSF project, we established
this lab last year with the mission of exploring innovative approaches to
knowledge discovery in Information Technology. The research team responsible
for this lab are currently: Shahram Latifi (PI), Emma Regentova (Investigator),
Dongsheng Yao (Master's Student), Shaojie Wu (Master's Student) and Leslie Sabo
(Senior Undergraduate).
iv)
A Ph.D. Graduate
Ms. Shulan Deng graduated from UNLV in Fall 2000 after completing all the requirements for the Ph.D. degree. Since 1997, Shulan had been working as a research assistant on this project.
v) Hire
Dr. Emma Regentova who was working on this project for the past two years in the post-doc's capacity, has now secured a Tenure-track position in the Electrical and Computer Eng. Dept. This hire was mainly due to the success we had in her collaboration with us, and improves stability of our research efforts.
Connected components are used for morphology analysis in OCR and for document segmentation based on feature aggregation. Connected components detection and their geometrical properties are extremely time consuming procedures. The problem of connected component extraction directly in the Group IV domain is being investigated. The idea is to analyze coding modes and check the conditions of overlap and touch of the black runs in two adjacent lines of raster image (coding/decoding line and the reference line) and to apply propagation type labeling for the connected runs.
JBIG compression scheme uses two- and three-lines context for prediction and indexing of arithmetic coders and progressive transmission of low resolution images. Two lines context can be used to derive information about connection of black runs in two raster lines and to label them. What is important, JBIG uses sophisticate rules for resolution reduction such as exception handling for isolated pixels, lines and dithered patterns that allows for connected components detection on ,low resolution images. Geometrical properties of connected components derived from low resolution JBIG images can be employed for aggregation and layout analysis rather than for character recognition. This approach allows to speed-up computations for feature extraction and implement layout analysis and segmentation on the reduced set of data, that is faster. The second task being explored is the conversion of raster images into vector forms. This operation is of interest in many applications ranging from cartography to character recognition. Corner detection can be implemented along with connected components detection while connection is analyzed. Also, information about the local structure can be derived in the course of connection detection and labeling of the black runs directly in G4 domain, while two-lines template of JBIG scheme can serve to detect corners. In the meantime, we have considered applying the emerging technology, wavelet transform, to document image processing. Most segmentation techniques rely on a priori knowledge or assumptions about the generic document layout structure and textual and graphical attributes. However, in many applications it is desirable to have segmentation methods that do not assume any a priori knowledge about document layout. Wavelet transforms have been widely used as effective tools in texture segmentation in the past decade. Segmentation of document images, which usually contain three type of texture information: text, picture and background, can be regarded as a special case of texture segmentation. Thus, wavelet transform can be exploited efficiently to distinguish between these three types of textures.
· James S. Walker, A Primer on Wavelets and their Scientific Applications, Man & Hall/CRC, 1999
· R.M.Rao, A.S.Bopardikar, Wavelet Transforms, Addison-Wesley, 1998
· Jia Li, Robert M. Gray, Text and Picture Segmentation by the Distribution Analysis of Wavelet Coefficients, International Conference on Image Processing, Chicago, USA, 1998.
· K. Jain, S. Bhattacharjee, Text Segmentation Using Gabor Filters for Automatic Document Processing, Machine Vision.
Processing digital images has become widespread recently due to remarkable advances in hardware and software industry. Such images can be produced by converting photographs, printed text, and other media into digital form. Direct acquisition of data in digital form has also been desirable because of the reliability and precision involved in handling digital information. The CCITT G3/G4 standard, developed based on run-length data, is widely accepted for binary document compression in general and facsimile transmission in particular. By means of this standard, users can interchange and share information efficiently and conveniently. This also leads to the desire to process the image to satisfy specific needs. JBIG compression is based on progressive transmission that allows to receive documents of the lowest resolution and request for increased resolution version until decent quality of image that makes characters recognizable will not be obtained. Traditional algorithms for image processing are based on pixel, which are computationally expensive. This has motivated researchers to consider processing image data in the compression domain.
The algorithms for run-length-based coding schemes are well understood and proven to be efficient. In the CCITT G4 domain, white pass modes can be used to detect skew and image matching, while vertical modes can be searched for possible bar codes. Therefore one can see that the features and structures existing in coded data can be utilized and accessed directly to facilitate image operations.
· K. R. Castleman, Digital Image Processing, Prentice Hall, 1996.
· K. Sayood, Introduction to Data Compression, Morgan Kaufmann Publishers, Inc., 1996.
· H. Witten, A. Moffat, and T. C. Bell, Managing Gigabytes, Van Nostrand Reinhold, 1994
· J. Shi and J. Malik, ``Normalized Cuts and Image Segmentation'', IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(8), 888-905, 2000.
· J. Malik, S. Belongie, T. Leung, J. Shi ``Contour and Texture Analysis for Image Segmentation'', International Conference in Computer Vision, Greece, 1999.
Potential Related Projects
Data base operations could be done more efficiently if data were stored (perhaps in some sort of compressed form) and queried differently from the classic methods. Specific compression techniques may prove useful in improving performance of data intensive applications such as data mining.