Learning Pseudo Metric for Multimedia Data Classification and Retrieval

This paper aims to develop a theoretical framework for learning pseudo metric (LPM) for multimedia data classification and retrieval. Neural networks are employed to approximate the LPM through learning feature examples. Training samples are generated by a k-MEAN clustering technique, and practical criteria for verifying the approximation performance are presented. The LPM metric is evaluated using a semantic image classification task where the database contains 11 categories of natural images with an accurate ground truth. Experimental results demonstrate the usefulness and effectiveness of the proposed techniques for multimedia data classification and retrieval.