12/12/2002 09:00 AM
Electrical Engineering
A new technique for generating and losslessly compressing region boundaries in digital images is presented. Region boundaries are generated by finding a minimal set of pixels required to differentiate each region from neighboring regions. An enhanced, pixel-based, differential chain code is used to encode the generated boundaries. We utilized run-length encoding of links and introduce the techniques of dynamic template switching and extended vector templates. These features allow accurate region boundary representation using a minimal number of links while retaining the inherent simplicity of traditional chain codes. The results show that the boundary generation and encoding approach is able to encode all possible region boundary conditions when pixels are used as region boundaries. The results compare well to other pixel-based encoders (in bits-per-symbol) despite the fact our encoder contains extra information embedded in the chain code. The bit-encoding rates of the boundaries are relatively low compared to the number of pixels in the images over a test image set. Despite the addition of multiple features to the boundary encoder, the chain coder remains relatively simple. The technique of dynamic template switching was shown to improve the bit-encoding rate of the generated region boundaries.