Computer Vision Architecture using Fusion Technique

Abstract

Humans want to communicate with the computers in the same way as they communicate with other humans. Speech is the most natural and spontaneous form of communication. Speech is bimodal in nature and it combines audio and visual information to enhance speech recognition rate especially under poor audio conditions. This paper proposes novel computer vision architecture using fusion technique. This architecture combines or fuses more than one modality using multi-agents. In this we have used two modalities-audio and video. The audio part extracts the speech of a person and the video part extracts the face and lip information of the person. Here, different agents process the modalities and the fusion agent fuses these modalities for effective and efficient automatic speech recognition.