Detection of Frames with Single Complete Signs of Finger Alphabet in Video Sequence Using Visual Attention Approach

Visual information is very important in human communication. It is used in any type of sign language communication, and in non-verbal communication of the entire population, as well. Therefore, visual information is crucial for communication of hearing impaired people. Video is the most common way to capture this type of information and it is very important to correctly process it. In this paper we propose a method for finding video frames representing single sign in the finger alphabet. The single sign is identified using standard video quality metrics. The calculations of the metrics are performed only within a region, which is determined by combination of object tracking and salient regions detection method based on human visual attention. For key frame selection, combination of sliding system for finding local extreme and adaptive threshold based on local averaging and variation is used. Proposed method is effective and achieves significantly better results in comparison with other commonly used methods.