Auditory-Visual Speech Processing 2005

British Columbia, Canada
July 24-27, 2005

A Coding Method for Visual Telephony Sequences

Edson Bárcenas, Mauricio Díaz, Rafael Carrillo, Ricardo Solano, Carolina Soto, Luis Valderrama, Javier Villegas, Pedro Vizcaya

Pontificia Universidad Javeriana, Bogotá-Colombia

Usually the design of a vector quantizer involves the minimization of a distortion measure such as the MSE. In this paper we present a new paradigm applied to the design of a visual telephony coding scheme: the synthesis of credible image sequences. In other words, the creation of smooth and coherent transitions between the images to be reproduced. This new paradigm requires the redefinition of the samples representative of each class in the Lloyd- Max algorithm. In our case the design criterion is the minimization of the maximum error within the class samples and their representative. The results obtained from the proposed method are compared to those obtained from the Lloyd-Max original algorithm.

Video transmission over asynchronous networks without real time control frequently suffers from information losses that can cause the loss of entire images. The present paper introduces a method based on the interpolation of the received images to estimate the lost images. The interpolation uses the search of credible image sequences by means of the Viterbi algorithm and a modified sequence distance.

Full Paper

Bibliographic reference.  Bárcenas, Edson / Díaz, Mauricio / Carrillo, Rafael / Solano, Ricardo / Soto, Carolina / Valderrama, Luis / Villegas, Javier / Vizcaya, Pedro (2005): "A coding method for visual telephony sequences", In AVSP-2005, 87-92.