An architecture for enhancing image processing via parallel genetic algorithms and data compression

This paper improves the parallel genetic algorithm (PGA) by applying data compression techniques to the image in order to minimise the data manipulated by the chromosomes. Image registration is performed in the compressed domain and the chromosomes encode the transform in this domain. The transform must then be converted back to the real world domain for practical use. Consequently the chip area can be decreased by the compression factor, and the processing time for the image can be improved. A variety of compression techniques can be used, for example JPEG, Fractal, various forms of Discrete Cosine Transform (DCT), run length encoding, Huffman encoding, and Arithmetic encoding. For image registration purposes the compression method must be fast and provide a good compression ratio. The compression method does not have to be lossless. DCT is a lossy technique that has a good compression ratio. DCT is the easiest method to implement efficiently on-chip. Consequently the DCT compression method was chosen as an effective lossy compression technique for reducing the image size. Additional benefits to using a lossy algorithm include the ability to compress the image to a fixed compressed image size. This permits a variety of sizes of image to be processed on a chip with limited memory per chromosome. Consequently this system is very flexible. The technique and its implications are described along with simulated results for a number of images. The design was evaluated using a 1μm ES2 CMOS process, in which an individual chromosome could be processed in approximately 2 milliseconds