Codebook Edge Detection

Abstract

This paper deals with the problem of extracting edge structure from compressed image representations. As image coding techniques become more common in image manipulation systems, it is reasonable to develop methods of analyzing an image using operations on the compressed image representation rather than the reconstructed image. Such an approach can be computationally efficient and produce an overall gain in computing performance.In this paper a technique for detecting edges directly from vector-quantized image representations is developed. The technique is shown to provide good performance in comparison to other gradient-type edge detectors, requiring no computation beyond the initial coding of the image. This approach provides a method ofextracting edge information which may be useful in the processing of very large image datasets.