Hybrid Pyramid / Neural Network Vision System

摘要

We have discussed with the NEL the types and variety of images and targets neededto evaluate our technology for aiding image analysts in real-world applications. We decided on the problem of detecting bomb-damaged structures. We have since visited the NEL to use the Directed-Search Testbed for selecting the positive examples for training and testing. We are in the process of preparing the data at Sarnoff for training. Training should begin early in June. NIDL funding is expected for applying the HPNN technology to two separate problems in breast-cancer screening: (1) Finding microcalcifications (evidence of tumors) in mammograms. This is a continuation of our previous work which reduced the false-positive rate obtained by the Rossman laboratory of University of Chicago by a factor of two. (2) Detecting masses in mammograms. The Joint Warfare Analysis Center has made some additional progress (to that described in our last report) in evaluating the HPNN technology. They have applied it to their own imagery, and report that in the initial tests it works well.