Abstract

Diffraction pattern sampling has been successfully applied to various problems in automatic pattern recognition, e.g., sorting of photographs based on scene content. In the current research we address the problem of sorting pictorial imagery according to whether a photograph was exposed upright (0°) or rotated 90°. This problem is of interest in the field of photographic processing, where along a continuous film strip, it is desirable to know whether a photograph was exposed in landscape or portrait mode. We limit the photographs in our study to those containing (1) landscape and houses only, (2) people only, and finally (3) groupings of houses and people. The optical transform is known to be independent of image translation but relates in a one-to-one fashion with image rotation. The detector used for sampling the optical transform is the Ring-Wedge detector, consisting of 32 wedge-shaped detectors and 32 concentric half-ring detectors. Using a feature set extracted from data obtained from wedge (angle-dependent) values, we illustrate the method of software development. Algorithms based on physical optics expectations are found to give reliable sorting in some cases. Discriminant functions using multiple feature algorithms are described. Results are presented in truth-table form for the three groupings listed above.

© 1987 Optical Society of America