CS 223B Project Proposal: Analysis and Characterization of a Color Edge Detection Algorithm

Kate Starbird / John Owens

Project Description
There are many existing edge-detection algorithms which produce a variety of results. Research continues to refine and increase the precision of these edge detectors. One area of improvement involves the use of color data to locate edges in a scene. The majority of edge detection algorithms deal mainly with the greyscale intensities, but including color differences between regions should allow for more precise edge detection, using more information to produce results. As computers become able to deal with larger quantities of data at higher speeds color edge detection becomes increasingly attractive.
We will implement an existing color edge detection algorithm and attempt to improve its performance. We also plan to study the parallelism of our chosen algorithm. We will explore both coarse grained (task-level) and fine grained (instruction-level) parallelism.

The final product will be a color edge detection application, written in C, which examines PPM (color) files and outputs a PBM (black/white) image file.
While we will implement the algorithm itself, the parallelization will not be implemented but instead described. If time permits, we will also describe an implementation of a suitable memory and cache layout strategy.

Koschan, A Comparative Study On Color Edge Detection.
Kumar, Gopalakrishnan, Kanal. Parallel Algorithms for Machine Intelligence and Vision, Springer-Verlag, 1990.
MIPS Extension for Digital Media with 3D. (as multimedia instruction set example)

Schedule of Work
4 February: Project proposal submitted
10 February: Specific color edge detection algorithm to implement identified.
19 February: First pass done of implementation of algorithm. Begin parallelization work.
27 February: Second pass done of implementation. Parallelization study complete.
13 March: Writeup complete.