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. Its central idea is to combine advanced inductive learning techniques with background domain knowledge in parallel multi-level image analysis and interpretation. The methodology is illustrated by a problem of learning how to conceptually interpret natural scenes. In the experiments described, three learning programs were employed: AQ15c-for learning decision rules from examples, NN-for neural net learning, and AQ-NN-for multistrategy learning that combines symbolic and neural net methods. Presented results illustrate the performance of the learning programs in Yearning to interpret natural scenes in terms of recognition accuracy, training time, recognition time, and complexity of the created descriptions. The methodology has proven to be very promising for the presented application. Overall, the experiments indicate that the multistrategy learning approach AQ-NN appears to be the most advantageous.