AN APPROXIMATE REASONING TECHNIQUE FOR RECOGNITION IN COLOR IMAGES OF BEEF STEAKS

Visual information plays in important role in food science research and applications. Color and color distribution act as cues in many such discrimination problems. In the determination of degree of doneness in beef steaks, for example, it is the distribution of red and brown which serve as visual indicators. In previous work, we developed capabilities to perform the basic color processing of food images. In this paper we present a methodology, based on approximate reasoning, for automatically determining the degree of doneness from the color images. We define a meaning vector of fuzzy sets for the fuzzy variables representing doneness classes from several of the color histograms of the steak images. We then construct a decision function which produces a fuzzy degree of agreement between the meaning of vector of an unknown sample and the prototypical vector corresponding to each class. This approach produces good classification results when the final class memberships are converted to a crisp partition. However, the memberships themselves provide an indication of the strength of class assignment. The technique is compared to two crisp and fuzzy feature-based pattern recognition algorithms. © 1990, Taylor & Francis Group, LLC. All rights reserved.