Resilient transmission of H.264/AVC video sequences using Probabilistic Neural Networks

H.264/AVC is expected to become an essential component in the delivery of wireless multimedia content. While achieving high compression ratios, this codec is extremely vulnerable to transmission errors. These errors generally result in spatio-temporal propagation of distorted macroblocks (MBs) which significantly degrade the perceptual quality of the reconstructed video sequences. This paper presents a scheme for resilient transmission of H.264/AVC streams in noisy environments. The proposed algorithm exploits the redundant information which is inherent in the neighboring MBs and applies a Probabilistic Neural Network (PNN) classifier to detect visually impaired MBs. This algorithm achieves Peak Signal-to-Noise Ratio (PSNR) gains of up to 14.29 dB when compared to the standard decoder. Moreover, this significant gain in quality is achieved with minimal overheads and no additional bandwidth requirement, thus making it suitable for conversational and multicast/broadcast services where feedback-based transport protocols cannot be applied.