Page 250 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience
P. 250

Section 9.8.  Interactive Techniques                          227


            concealment is successful and the error of a given block is su$ciently small,
            then the encoder may decide against INTRA encoding. Note that this method
            requires  the  encoder  to  perform  the  same  error  concealment  process  that  was
            used  at the decoder.


            9.8.2.2  Restricted Prediction Based on Feedback
            Based  on  the  error  trace,  prediction  of  the  current  frame  is  restricted  to  use
            only error-free areas in the reference frame. For example, in Figure 9.9(c) the
            a ected pels in the reference frame, n + d − 1, are not used for predicting the
            current frame, n + d. This stops error propagation to the next frame, n + d +1.
            This  restricted  prediction  based  on  feedback  and  error  tracking  was  proposed
            by Wada in the  selective  recovery  technique [203].
               Again,  this  technique  can  also  bene t  from  the  reduced-complexity  error-
            tracking  algorithm  of  Steinbach  et  al.  [202],  and  the  coding  e$ciency  can
            also be improved by performing error concealment in the encoder so that both
            encoder and decoder  use  the same  reference  frames for prediction.


            9.8.3  Reference Picture Selection

            In reference picture selection (RPS), both the encoder and decoder store mul-
            tiple previous frames to be used as reference frames. When the encoder learns,
            through  feedback  messages  from  the  decoder,  that  the  most  recent  reference
            frame  contains  errors,  the  encoder  switches  to  use  another  older  reference
            frame that is known to be error free. Provided the alternative reference frame
            is not too far away from the current frame, the loss in coding e$ciency is not
            signi cant.  In  particular,  this  technique  is  more  e$cient  than  the  INTRA  re-
            fresh technique. The RPS technique has been adopted by H.263+ in annex N,
            and  an  enhanced  version  of  the  technique  has  been  included  in  annex  U  of
            H.263++.
               Figure 9.10 shows the RPS technique with two types of feedback messages.
            In  the  negative  acknowledgment  mode,  illustrated  in  Figure  9.10(a),  the  de-
            coder  sends  a  negative  acknowledgment  (NACK)  message  whenever  errors
            are  detected  in  a  frame.  In  the  example  shown,  the  decoder  detects  an  error
            in  frame  3  and  sends  a  NACK(3)  message  to  the  encoder.  At  the  encoder,
            the  encoding  operation  proceeds  in  the  normal  way  (i.e.,  using  the  most  re-
            cent reference frame for prediction) until the NACK(3) message arrives before
            encoding  frame  6.  Based  on  this  message,  the  encoder  knows  that  errors  oc-
            curred  in  frame  3  and  propagated  up  to  the  most  recent  reference  frame  5.
            To stop this error propagation, the encoder uses the older error-free reference
            frame  2  instead  of  the  most  recent  reference  frame  5  to  encode  the  current
            frame 6.
   245   246   247   248   249   250   251   252   253   254   255