Pattern recognition with parallel associative memory

Recognizing specific patterns plays an important role in digital photogrammetry. A common task is locating special targets in an image, such as control points. The paper explores the applicability of Parallel Associative Memory (PAM) for searching targets in aerial photographs. The PAM we used in our research is based on the Nearest-Neighbor algorithm and uses the Hamming distance as a measure of closeness to discriminate patterns. This approach is particularly successful if the patterns can be described logically. Our research focused on targets typically used for ground control points. We tried to develop a method which parallel to the data acquisition process sorts out the approximate target positions where the precise localizations are needed. A library with different targets was entered into the PAM. Image patches that were moved across the image were constantly compared with the library by determining the Hamming distance. The results are encouraging. The majority of control points in different images were correctly identified and only a few targets were wrongly matched.