The Most Significant Edges: An Efficient Image Description for Machine Vision Applications

Efficient edge detection is accomplished by means of a hierarchical processing scheme which implements spatial frequency channels similarly to processing in biological visual systems. We use such a pyramidal data structure to efficiently create the channels and thereby extract a set of edges from the various scales of the image. The set of selected edges serves as a compact and simple representation of the image, and as such appears to be most useful in image description for machine vision applications. To assess the importance of the information represented by the most significant edges thus extracted vis-avis image structure, images are reconstructed from these edges and compared with the originals based on human subjective (perceptual) criteria.