| Popis: |
Edge detection is an important first step in many vision tasks where its improvements in speed and efficiency present a continuous challenge for developers of high-speed image recognizers. Classical techniques for accurate detection of edge features, as exemplified by Canny operator, demands such expensive operations as the iterative use of Gaussians and Laplacians, and their designs are largely sequential. Alternatively a variety of complex and edge-preserving filters have been developed to reduce the effects of noise without significantly distorting the edge loci. This paper describes a cascaded precursor approach for edge detection based on selective local contrast modifications which combine point- wise image operators and non-linear transformation. A principal advantage of the approach lies in its simplicity and uniformity of operations; the latter is a characteristic blueprint for efficient (parallel) low-level image processing algorithms. Further, unlike many enhancement algorithms, the characteristics of the proposed precursor can be studied analytically, thus allowing the independent adjustments of detector parameters for maximum performance in the specific environment. |
