LAUSR

In this paper, a Chebyshev polynomial-based functional link artificial neural network (CFLANN) technique for Wyner–Ziv (WZ) frame estimation in a distributed video coding framework is proposed. The estimated WZ frame at the decoder is also referred to as the side information (SI). The proposed scheme (CFLANN-SI) works in two phases, namely, training and testing. The network is trained offline, and to achieve better generalization, the training (input, target) patterns are created across several video sequences constituting varied motion behavior. It estimates the SI frame using adjacent key frames as inputs. The training convergence characteristics of CFLANN-SI is observed to be faster with reduced mean square error as compared to a multi-layer perceptron-based prediction scheme. It is also observed that once the model is trained, it is capable of estimating SI for rest of the incoming WZ frames of the video sequences as well as for the video sequences which are not considered during the learning phase. The proposed scheme is evaluated with respect to different parameters, namely, rate-distortion, peak-signal-to-noise-ratio, the number of parity requests made per estimated frame, decoding time requirement and so on. Comparative analysis shows that the present CFLANN-SI technique generates better SI in resemblance to the competent schemes, in terms of the subjective quality improvement as well as the objective quality gains. Further, to substantiate that the present scheme provides a significant improvement over that of the benchmark techniques, a statistical analysis tool is used with a significance level of 5%.