Codec Evaluation

Feedback Channel Performance evaluation

In the DISCOVER monoview video coding architecture, the feedback channel has the role to adapt the bitrate to the changing statistics between the side information (an estimation of the frame to be encoded) and the frame to be encoded, i.e. to the quality (or accuracy) of the frame interpolation. Therefore, contrary to conventional codecs, it is the decoder's responsibility to perform rate control and, in this way, to guarantee that only a minimum number of parity bits are sent to correct the mismatches/errors present in each side information bitplane (thus the use of a channel coding tool like LDPC or turbo coding). Since this decoder rate control operation based on the feedback channel is central in the DISCOVER codec, it is important to be aware of its behavior and impact in order to design more efficient WZ video coding algorithms.

Measuring the Number of Requests

During the decoding of a given bitplane of a given DCT band bk, the decoder may request one or more times to the encoder for more parity bits. The number of requests depends mainly on the side information quality, on the bk DCT band number of bitplanes and on the accuracy of the initial rate estimation; the latter is quite important since the encoder estimates for each bitplane the initial number of parity bits (minimum rate) which are sent at once to the decoder, and requests only occur if more parity bits are necessary to decode the bitplane with a low amount of errors. To have an insight on how the number of requests varies with the temporal correlation of the video sequence and the accuracy of the rate estimation performed at the encoder, it is presented here the number of parity bits requests at the bitplane level of each band, and for each frame.

The average number of decoder requests at frame, D(Q), and bitplane, D(Q)ij, levels for a certain quality rank, Q, is computed from [1].

The following results were obtained:

1. Number of Requests

Measuring the Feedback Channel Rate

After the average number of requests per DCT band and bitplane is known, it is possible to measure the feedback channel rate for each DCT band and bitplane. In order to measure the feedback channel rate, it is assumed that only one bit is required by the decoder to inform the encoder if more parity bits are needed or not to successfully decode the current bitplane. If more parity bits are needed, the decoder sends the bit ‘1’ via the feedback, channel; otherwise, the bit ‘0’ is transmitted and the encoder, receiving such bit, sends parity bits for the next bitplane to be decoded. This is clearly a simplistic solution since more bits are needed depending on the transmission solution used; however, it serves to have an idea of the feedback channel rate.

Since only one bit is transmitted via the feedback channel for each decoder request, the total feedback channel rate at frame R(Q), and bitplane R(Q)ij levels for a certain quality rank, Q, is computed from [1].

The following results were obtained:

2. Feedback Channel Rate

Measuring the Number of Errors Versus the Number of Requests

Since WZ coding is very much about correcting ‘errors’ in the side information estimation, it is interesting to evaluate the amount of requests needed to correct a certain amount of ‘errors’. With this purpose in mind, in this section the number of errors corrected with the number of requests made for each bitplane is measured. It is important to keep in mind that since an initial number of bits is estimated at the encoder and sent at once to the decoder, there is no direct relation between the number of requests made and the number of bits used for the decoding of each bitplane. Therefore, the number of requests made basically reflects the accuracy of the initial rate estimation.

The following results were obtained:

3. Number of Errors vs Number of Requests

Measuring the Number of Requests Versus Side Information Quality

Since the parity bits are successively requested to correct the side information errors and improve the decoded quality, it is important to know how the number of requests varies with the side information frame quality. This will allow designing more adequate request strategies since the number of requests has a significant impact on the decoder complexity.

The following results were obtained:

4. Number of Requests vs Side Information Quality

References

1. J. Pedro, C. Brites, J. Ascenso, F. Pereira, "Studying the Feedback Channel in Transform Domain Wyner-Ziv Video Coding", 6th Conference on Telecommunications - ConfTele, Peniche, Portugal, 2007.