A method for obtaining neural network training sets in video sequences

A method for video sequence classification is presented as a technique for enhancing the performance of neural network classification systems beyond that obtained by manually selecting the training data. The method uses a stochastic representation of each image frame in a sequence set, to form the basis for texture-based classification of each image. Each frame is allocated into statistically similar bins. The bins are used for discriminating large amounts of data and enhance the possibility of selecting representative vectors when neural network training sets are formed. The technique is shown to be a powerful method for operator independent classification of large sets of video data. The resulting classification is not limited to training sets selection. It can also be used for data mining, video indexing, "ground truth" for algorithm evaluation and classification as well as data profiling.