The all-digital ring-wedge detector applied to automatic object recognition

An all-digital ring-wedge detector is presented, which when coupled with neural network software produces a highly flexible recognition system, sharing many of the advantages of the optoelectronic hybrid system for obtaining the same data format. While less capable in terms of space-bandwidth product, the all-digital system is relatively simple to use, much less expensive, and readily applicable both with hard-copy images and digital images, offering the possibility of interesting variations on what one is able to do with the hybrid system.; Several problems dealing with automatic object recognition are studied with contributions including the first detailed descriptions of two separate algorithms for calculating the ring-wedge data format. Also, the use of spatial frequency domain information in combination with image domain information for improved recognition is emphasized through the introduction of a novel localized ring-wedge transform, which extracts localized spatial frequency information as a function of position within a larger scene.; In fingerprint recognition we demonstrate a very effective system applicable with either gray-scale or binary images, including ring-only (orientation independent) and wedge-only (scale-size independent) sortings. Further, we present a successful fingerprint verification system based on a user carrying a data encrypted key and being allowed entrance by comparing stored data to data collected directly from the subject's fingerprint. In the application of the localized ring-wedge transform to fingerprint imagery, direct examples are presented demonstrating fingerprint quality detection and local ridge orientation determination.; In the operator-independent, automatic assessment of image quality, we report a high accuracy in the classification of both linear (image blur level) and nonlinear (image compression artifacts) degradations in a manner that is widely independent of scene content. Using interesting system variations, we present several experiments in which the end goal is to classify images according to numerical quality scales. In addition, the localized ring-wedge transform is found to be valuable for estimating local perceptual fidelity. We emphasize that our method is using information from the degraded image alone, without specific data from the original scene; however, the original scene can be incorporated as a reference when appropriate.