The MIST Methodology and Its Application to Natural Scene Interpretation

The MIST methodology (Multi-level Image Sampling and Transformation) provides an environment for applying diverse machine learning methods to problems of computer vision. The methodology is illustrated by a problem of learning how to conceptually interpret natural scenes. In the experiments described, three learning programs were used: AQ15c—for learning decision rules from examples, NN—neural net learning, and AQNN—multistrategy learning combining symbolic and neural net methods. Presented results illustrate the performance of the learning programs for the chosen problem of natural scene interpretation in terms of predictive accuracy, training time, recognition time, and complexity of the induced descriptions. The MIST methodology has proven to be very useful for the presented application. Overall, the experiments performed indicate that the multistrategy learning program AQ-NN appears to be the most promising approach. ___________________________ 1 This research was supported by the Advanced Research Projects Agency under grants F4962092-J-0549 and F49620-95-1-0462, administered by the Air Force Office of Scientific Research. The development of machine learning systems used in this research was supported in part by the Advanced Research Projects Agency under grant No. N00014-91-J-1854, administered by the Office of Naval Research, in part by the Office of Naval Research under grant N00014-91-J-1351, and in part by the National Science Foundation under grants IRI-9020266 and CDA-9309725.