USING SPEECH ANNOTATION FOR HOME DIGITAL IMAGE INDEXING AND RETRIEVAL

Abstract:- We present in this paper a system for image indexing and retrieval through speech annotation made directly on a digital camera. The annotations are based on a pre-defined structured syntax and transcribed by an automatic speech recognition engine. N-best lists are incorporated in the index generation and query expansion process to introduce probabilistic model and compensate recognition errors. We also apply a practical noise cancellation algorithm to enhance the speech signal collected directly through the built-in microphone on a digital camera. Experiments show that by utilizing automatic query expansion and noise cancellation together, the overall system performance improve significantly to be comparable to the lab environment collected data