Top3.1 Error Detection, Concealment And Recovery
A general diagram of error detection, concealment and recovery communication system using watermarking is illustrated in Figure 1
Figure 1. General error detection, concealment and recovery communication system
The watermark, which carries characteristics information of host signal, is embedded into host signal with watermark encoder in an un-intrusive way so that the introduced watermark is not perceivable. After being transmitted through the noise communication, the signal reached at the receiver is a distorted host signal. The watermark is extracted by the watermark decoder from the contaminated signal. This watermark then is used to help recover the host signal by correcting or concealing the errors.
Packet loss or delay usually occurs when transmitting multimedia in wireless and Internet environment. Protocols like UDP and TCP either leads to partial representation or requires re-transmission, which introduces intolerable time delay.
Lin et al. (2001) proposed an error detection and concealment in UDP environment using self-authentication-and-recovery images (SARI). Watermarks which contain content based authentication and recovery information are embedded into SARI image or video frames prior to transmission. At the receiver side, the embedded authentication information in the watermark is used to detect the locations of corrupted image blocks and the recovery information is employed to approximately restore the lost blocks.
The embedded watermarks are claimed to be compatible with quantization –based lossy compression like JPEG / MPEG. Since the watermarks are embedded prior to transmission, no extra control over the transmission or encoding process is needed. At the receiver side, the recovery is not based on adjacent image / video blocks, thus making it possible to restore packet loss in large areas or high variant areas.
Chen et al. (2005) proposed a fragile watermark error detection and localization scheme called “force even watermarking” (FEW) for wireless video communications, where compressed video streams are extremely sensitive to bit errors including random and burst bit errors, which hinder correct playing of the steaming video. In the proposed FEW algorithm, a fragile watermark is forcedly embedded on the DCT coefficients at the encoder side. During the decoding process, the same watermark is checked first for error detection and the locations of error bits can be accurately localized at macro block level. Since error recovery methods are usually employed at MB level, the localization ability of this fragile watermark scheme greatly improves the success of conceal erroneous MBs.
The proposed FEW watermarking encoder embeds fragile watermark on an 8 x 8 DCT macro block by forcing certain DCT coefficients quantized to even values, i.e. all the 64 quantized DCT coefficients after the typical zigzag scan position are rounded to the nearby smaller even values. This embedding procedure happens right after the normal video encoding quantization process. At the decoder side, if any of the received DCT coefficients in 8 x 8 block has odd value, then the system knows errors have occurred to that block during transmission.
According to their experimental results, compared to the traditional syntax-based error detection schemes, the proposed FEW scheme doubles the error detection probability, greatly improves the correctly locating an error by 3 to 13 times and yields a low image degradation of less than 0.44 dB.
Transmitting block-coded images through wireless channels, which is error-prone, usually results high packet losses. Gür et al. (2007) invented an error concealment algorithm to provide a close approximation of the lost packets, thus covering up the packet losses.
In their algorithm, replicas of the host image’s M x M macro blocks in the subbands with two wavelet schemes namely wavelet-tree and pyramid-structured wavelet LL subbands are left untouched to limit the visual degradation. In order to achieve blind detection, a shared-key-dependent pseudo-random sequence (PN sequence) is used to choose the host macro blocks where watermarks (replicas in this case) are inserted.
The watermark encoder works as follows: