Detection and Location of Reseaux Using Pictorial Pattern Recognition

Plus-like images called reseaux and found on aerial photographs provide the capability of determining the extent of geometric distortion present in these photographs. Accurate location of these images within a photograph is, at present, a manual operation, An automatic system to perform this operation using a minicomputer system and microdensitometer is presented.

A model is generated from the statistics of actual reseau samples to create a test vehicle for the performance of various picture processing algorithms. A method of including various photographic effects into the model is presented.

Various combinations of scene analysis schemes are presented with results for the center location problem. Preprocessing techniques for noise removal are described. Also developed is a scheme for noise removal based on Bayes decision theory taking advantage of the relative slow scanning speed of the microdensitometer. Two methods of center location are described and comparative speed and accuracy results presented. The operation of the final system is also described.