Research of Image Affective Semantic Rules Based on Neural Network

Abstract

To bridge the semantic gaps between the low-level image visual features and the high-level emotional semantics, the paper describes image features using texture and completes the semantic mapping through BP neural network. On the premise of keeping the accuracy of classification unchanged, the trained feedforward neural network is pruned using RX algorithm. Finally, the rules of IF-THEN which can be understood easily are extracted from pruned neural network model. The experiment shows that the method is effective and the rules extracted are comprehensible.