Personal mobile video communication based on string image description

The semantic content of videos is mainly contained in objects and their time evolutions (rotations, translations and deformations). Literature states that the semantic content of an image is mainly concentrated in edges. In this paper, basing on the local dominant orientation information of images, a string signature has been introduced, trying to exploit similarities between pairs of frames of a given video. The image is partitioned into squared blocks, and for each block the main orientation information is quantized to a set of four possible ones. Labels are substituted to each block orientation information. String processing is used to compare the semantic content of any given couple of images of a sequence. The application framework of the proposed semantic content is perceptual video representation. This signature has been used in a communication application designed for (wireless) personal frameworks. The definition of a personal video codebook (PVC), stored and assumed known both on the personal handset and on the receiver, allows perceptual video coding with very low bandwidth. In this approach, the transmission of only the pointer to the PVC frame is required, as a first approximation. At the receiving link end the video is synthesized using the closer frame of the codebook instead of the true frame, at the generic time instant. High quality image can be obtained even in very low bit rate applications, as coding information is poor. Freezes in the video reproduction can almost be completely avoided, even for ABR links. Preliminary tests have been conduced on a panel of about 50 students. Results seem promising.