Optimal resynchronization marker positioning method using a novel information measure

https://doi.org/10.1016/S0923-5965(02)00059-0Get rights and content

Abstract

This paper presents a new resynchronization marker (RM) positioning technique to effectively block the error propagation caused by channel errors. The optimal technique can be defined as the one minimizing the information loss under an error condition. For realization of the optimal technique, the amount of the information lost by errors needs to be measured. Thus, we introduce a novel measure that is successful in reflecting the actual importance of the lost data, which is measured by using error concealment techniques. That is, the amount of information lost by an error at a certain location is numerically specified by the difference between the image recovered by the error concealment and the image normally decoded without any error. Since such a measurement method is appropriate for reflecting the actual damage by errors, it is very useful for determining the RM positions. With the new measure, the optimal RM positioning technique is proposed and verified by experiments.

Introduction

It is clear that wireless video will play an important role in emerging and future generations of communications systems. For wireless video applications, the error resilience has to be enforced since compressed video is especially vulnerable to errors. It can be achieved either in video encoder or in system layer or both [5]. In this paper, the error resilience in video encoder for wireless environment is handled, since the error resilience in system layer, such like a retransmission method, cannot be easily applicable to real-time video applications that require strict delay constraints.

High coding gain for video signals is achieved by eliminating the spatial and temporal correlations. Under noisy channel, however, the spatial and temporal dependencies cause the error propagation, which results in severe quality degradation in the reconstructed images. To cope with the problem, the coded bitstream is partitioned into video packets by resynchronization markers (RMs) and the data in the packets do not refer to each other. For example, in the process of the motion vector prediction and the DC/AC prediction in MPEG4, a packet boundary is considered as if it is a picture boundary. The picture start code (PSC) and the group of block start code (GBSC) in H.263 and VOP start code in MPEG-4 correspond to RMs. By need of reinforcing error resilience, the standards provide the options that RMs can be more frequently inserted.

A group of block (GOB) in H.263 [4] includes the fixed number of MBs, which will be called a fixed positioning method (FPM), as long as the slice structure mode is not indicated. Since the encoding process works in the manner of variable rate, RMs will most likely be unevenly spaced throughout the bitstream. Therefore, certain portions of the scene, such as high motion areas, will be more susceptible to errors. In the meanwhile, the video packet approach adopted by MPEG4 [3] is based on providing periodic RMs throughout the bitstream. In other words, if the number of bits contained in the current video packet exceeds a predetermined threshold, then a new video packet is created at the start of the next MB. In [7], [6], Yoo et al. have made an effort in finding the optimal RM position, but their analysis is valid only under the assumption that all MBs other than the not-coded MBs have equal importance. In general, the assumption is not valid for practical situations. In [1], Cote et al. solved more general problem that determined simultaneously the MB mode and RM positions in viewpoint of R-D optimization. In this paper, we deal with a specific problem to determine the optimal RM positions in case that coded bitstream is already prepared and the number of RMs is given. We introduce a cost function that represents the information loss by channel errors. The quantity of information loss is evaluated by summing the distortion from the location where error occurs to the end of the video packet. At this point, our method is different from Cote's method where it was assumed that if an error occurs at any location within a slice, the complete slice is discarded.

In this paper, the analysis in [6] is extended to the general case in which each of MBs retains a different importance. As a result of the extended analysis, the optimal RM positioning method is proposed. In contrast to H.263 and MPEG4, the proposed method is based on the actual importance of each of MBs, which is measured by considering the quality degradation caused by the loss of the MB. That is, the importance of a certain MB is measured as follows: after error concealment for the MB which is assumed to be lost, the quality of the MB recovered by an error concealment process is evaluated, referring to the MB normally decoded without any error. When the recovered MB is seriously degraded, it is considered that the MB is important or has a large amount of information. Since the important data have the large amount of information, in this paper, the importance is considered as the amount of information.

This paper is organized as follows. In Section 2, when a coded bitstream is partitioned into packets, the optimal RM positions are analytically derived. In Section 3, for the realization of the optimal positioning method, a novel measurement method that measures the amount of information of MBs is introduced. The amount of information of a certain MB, in this paper, means the degree of quality degradation by the loss of the MB. In Section 4, the proposed method is experimentally evaluated, compared with FPM and MPEG4s. Finally, in Section 5, the conclusions and discussions are described.

Section snippets

Optimal resynchronization marker positions

Consider a bitstream with length B that generated by encoding a frame (or visual object plane in MPEG4), where the length B can be either the number of bits or the number of MBs in the frame. The bitstream is partitioned to M video packets for transmission. Suppose that an error occurs at the position t during transmission. Then, the amount of information loss, L(t), is represented byL(t)=tNiα(x)dx=A(Ni)−A(t),forNi−1⩽t<Niand1⩽i⩽M,whereA(y)=−∞yα(x)dx,N0=0andNM=B.In Eq. (1), Ni and α(t) denote

Information measurement of MBs

The amount of information loss was previously defined as the importance of an MB or a bit. Accordingly, we try to quantify quality degradation by the loss of the MB or the bit. For example, when an MB at nth frame is lost, the current frame as well as the future frames are degraded by the lost MB but the past frames are not done. To accurately measure the importance of the MB, we have to reflect the degradation of the current frame and the future frames. If the current frame and the future

Experimental results

The theoretical results are first investigated. Suppose that α(t)=βeβt/B/(1−eβ), for 0⩽t<B and β>0, which is a monotonically decreasing distribution and approaches to a uniform distribution as β→0. On the other hand, the sum of α(t), i.e., the total amount of information, has a constant value B regardless of β, i.e., 0Bα(t)dt=B for β>0. The optimal RM positions for the distribution obtained by the proposed algorithm are shown in Table 2 where B=1000, the number of packets=4, and the number

Conclusions

We propose a new reasonable measure that measures the amount of information in MB unit as well as the resynchronization marker positioning technique that minimized the information loss using the measure. Since the measure is obtained based on the error concealment technique, it is successful in reflecting the actual importance of MBs. Thus, the proposed method is useful for MPEG4 in use of the RM enable mode and H.263 in use of the slice structure mode. The new method based on the measure is

References (7)

  • G Cote et al.

    Robust video communications over error-prone network

    J. Select. Area Comm.

    (June 2000)
  • T.R. Gardos, Video Codec Test Model, Near-Term, Version 10 (TMN-10), Draft 1, ITU-T Q.15/16, Document # Q16-D-65d1,...
  • ISO/IEC 14496-2 FDIS, Information technology – generic coding of audio-visual objects, Part 2: Visual, December...
There are more references available in the full text version of this article.

Cited by (3)

View full text