A method of processing digital image data to detect edges includes the use of fuzzy reasoning. The method is completely adaptive and does not require any advance knowledge of an image.
During initial processing of image data at a low level of abstraction, the nature of the data is indeterminate. Fuzzy reasoning is used in the present method because it affords an ability to construct useful abstractions from approximate, incom- plete, and otherwise imperfect sets of data. Humans are able to make some sense of even unfamiliar objects that have imperfect high-level rep- resentations. It appears that to perceive unfamiliar objects or to perceive familiar objects in imperfect images, humans apply heuristic algorithms to understand the images.
Fuzzy reasoning is a suitable means of heuristic processing of incomplete and otherwise imperfect data. Most prior edge-detection methods require the selection of parameters (e.g., thresholds in gradient edge-detection algorithms) — a difficult task when little or nothing is known about an image in advance. Moreover, prior edge-detection methods based on mathematical models can detect only specific kinds of noticeable edges: For example, an optimal mathematical- model-based step edge detector can be ineffective in detecting ramp edges. Relative to methods that involve mathematical models and advance selection of parameters, the present method and possibly other methods of processing image data in partial imitation of image processing in the human brain and eye offer greater flexibility and the potential for superior performance.
In the present method, a window of 3 by 3 pixels is scanned over the whole image. An optimal intensity gradient based on the central pixel is found through a heuristic analysis. A crisp central-pixel value is generated after a fuzzy membership function is evaluated by use of the optimal intensity gradient.
The method has been implemented in a C-language computer program. The method was tested by applying the program to an image of a compact disk. As shown in the figure, this method performed better at detecting edges than did computer programs that implemented two prior edge-detection methods known as the Sobel and Prewit methods. The program of the present method even detected a dark central spot containing narrow edges that the programs of the Sobel and Prewit methods did not detect at all.
This work was done by Jesus A. Dominguez and Steve Klinko of ASRC Aerospace Corporation for Kennedy Space Center/strong>.