Title:
Java Applets for LPS-Based Image Morphology

Author:
Mark DeRoller

Date:
Month? 1998

Abstract:
abstract

e-mail: mderolle(at sign)cldx.com

Visit project's home page.

Committee:
1, Chairman
2, Reader
3, Observer
Title:
Bitonal Image Compression Using Linear Pixel Shuffling

Author:
Elaine Meadows

Date:
November 1997

Abstract:
Linear pixel shuffling is a method based on an arithmetic progression for visiting all of the pixels in an image in a very mixed up order. The algorithm is designed so that when a fraction of the total pixels are chosen, every subrectangle of the fraction contains the same number of pixels. The total number of visited pixels is smoothly distributed over the entire image [1]. Because of the very even distribution of pixels over the image, it is also possible to limit the number of pixels visited based on the distance between visited pixels. If this distance is presented as using a "fat" pixel, it is possible to approximate the original image with "fat" pixels. This technique may be especially useful in representing and transmitting large images over a very busy network.

The purpose of this project is to use Linear Pixel Shuffling with the "fat" pixel technique to implement a lossy compression of a black and white image. The implementation will be done in Java, and will be suitable for running from an applet on the World Wide Web.

The shuffling algorithm is taken from [2]. The algorithm is based on a two-dimensional Fibonacci sequence.

References used in the above:

[1] Anderson, P.G., "Advances in Linear Pixel Shuffling", presented at the Conference on Fibonacci numbers and their applications, Pullman, Wa., July 1994.

[2] Anderson, P.G., "Linear Pixel Shuffling for Image Processing, an Introduction", The Journal of Electronic Imaging. April 1993, pp. 147-154.

Visit project's home page.

e-mail: emk3594(at sign)cs.rit.edu

Committee:

Dr. Peter G. Anderson, Chairman
Dr. Stanislaw P. Radziszowski, Reader
Title:
LPS Error Diffusion Half-Tone

Author:
John Szybist

Date:
April 1997

Abstract:
A new method of digital halftoning is described that combines Linear Pixel Shuffling order of visiting pixels in an image with diffusion of error in all directions. LPS is based on an arithmetic progression with wrap-around that re-orders the pixels to produce a smooth progressive probing of the image. The initial proposal experienced problems with high contrast because of error being diffused to pixels already rendered. Described are several solutions to this problem leading to an optimal technique that eliminates the problem. We compare the results of this method to the traditional Floyd-Steinberg error diffusion algorithm.

e-mail: szybist(at sign)kodak.com

Committee:
Dr. Peter G. Anderson, Chairman
Prof. Frank Cost, School of Printing, Reader
Prof. J. A. Biles, Information Technology Department, Observer
Title:
Document Image Orientation

Author:
Herb Chapman

Date:
TBA

Abstract:

Orienting document images can be a valuable step in a document processing system. This capability eases the document preparation process, and allows other processing steps to expect images in a known orientation. This project addresses document image orientation by concentrating on the text appearing in the document.

This system will capture characters from a scanned document, and attempt to recognize the characters and their orientations. The recognition will be performed by Optical Character Recognition (OCR) Neural Networks (NNs). Several systems will be configured by training a variety of NNs. The relative performance of each configuration will be evaluated.

Committee:
Dr. Peter G. Anderson, Chairman
Dr. Roger S. Gaborski, Eastman Kodak, Reader
Mr. John D. Hanson, Eastman Kodak, Observer
Title:
A File Format Based on Linear Pixel Shuffling

Author:
Jenny K. Pan

Date:
January 1998

Abstract:
There is a saying of "graphics file formats are immortal." Many kinds of graphics file formats have been in the market already, such as TIFF, GIF, JPEG etc., nevertheless, we always need a way to create a better ones and always need a way to read, understand, and display them.

Dr. Anderson has been doing research work on LPS (Linear Pixel Shuffling) theory for many years and he would like to implement his theory to a LPS Image (Graphics) Format, including LPS compression algorithm which is implemented by Elaine Meadows.

In my project, the following specs will be implemented:

(1) An html file invokes a Java applet file that brings out an ImageViewer frame.

(2) From the ImageViewer frame, the user may choose to open a GIF image file as original file or LPS image file with desired percent image shown.

(3) The user may save the image as a regular GIF file or in the LPS file format.

(4) This converter (display) project is implemented in JAVA and HTML and is able to displayed on WWW.

(5) The LPS converter will handle bilevel, monochrome, and color images.

Visit Jenny's home page.

Committee:
Dr. Peter G. Anderson, Chairman
Dr. Stanislaw P. Radziszowski, Reader
Prof. 3, Observer
Title:
Web-based LPS and Fractals

Author:
Wenlai Tian

Date:
Month? 1998

Abstract:
abstract

e-mail: wxt2966(at sign)cs.rit.edu

Visit project's home page.

Committee:
Dr. Peter G. Anderson, Chairman
tba, Reader
tba, Observer
Title:
Experiments with Neural Networks and Genetic Algorithms

Author:
Sanjay Raghavendra

Date:
September 1995

Abstract:
This report presents a series of experiments performed using novel training techniques for feed-forward neural networks. One set of experiments describes an algorithm in which the first layer of weights in the network are determined using a quasi-random number generator and frozen, that is, they are not trained. Only the second layer of weights are trained, and this can be done using a linear discriminator such as the perceptron training algorithm. The necessarily large number of hidden units can be pruned using several techniques detailed here, including a genetic algorithm search. Results from testing the algorithm on several problems, such as the classic two-spirals problem and higher dimensional cases, are presented.

In the second set of experiments, the problem is to train N-2-N compression networks to learn patterns of bits such that input equals output. The pattern to be learned has just one of the N bits "on." It turns out that this is a difficult problem to train for large N. Instead, we experimented with two indirect approaches. The report concludes with a summary of results from these experiments.

Committee:
Dr. Peter G. Anderson, Chairman
Dr. Roger S. Gaborski, Reader
Prof. John A. Biles, Observer

e-mail: anderson(at sign)cs.rit.edu
Title:
Mandelbrot LPS Demonstration.

Author:
Lee Hurst

Date:
May 1998

Abstract:
Mandelbrot set demonstration using Java.

See the demo.
Title:
LPS Error Diffusion for Color Images

Author:
Qi Xie

Date:
July 1998

Abstract:
We implement a new method of color image halftoning. This method combines Linear Pixel Shuffling order of visiting pixels in a image with error diffusion. Linear pixels shuffling is a quasi-random ordering of pixels, a technique with benefits of pseudo-random ordering, but described by an extremely simple rule based on an arithmetic progression. Digital halftoning of grayscale images using LPS and error diffusion has been studied and the performance is quite good. In this project, this new method will be applied to color images. The results will be tested and compared to the traditional error diffusion algorithm such as Floyd-Steinberg algorithm.

e-mail: qxx3451(at sign)cs.rit.edu

Visit project's home page.

Committee:
Dr. Peter G. Anderson, Chairman
Dr. Roger S. Gaborski, Reader
Dr. Feredoun Kazemian, Observer

e-mail: anderson(at sign)cs.rit.edu