摘要：H.264/AVC视频压缩标准对信道错误的鲁棒性是重点由它错误控制技术进行评估的，我们称错误评估技术为FMO。我们调查研究了FMO作为一种H.264的误差控制技术是如何在一个容易产生错误的网络中对于传输误差和包丢失进行最好的处理。一种新的使用MBAmap将宏块映射到条带组的方法已经采用，这种方法可以帮助在解码中进行误差隐蔽技术并且提高了被接收视频的主客观性能。
关键词：FMO H.264/AVC 误差控制 MBAmap 视频传输
Abstract： The robustness of the H.264/AVC video compression standard against channel error is evaluated with the focus on one of its error resilience tools called FMO. We investigate how the FMO as an error resilience tool of H.264 can be deployed best for coping with the transmission errors and packet losses in an error-prone network. A new approach for mapping macroblocks into slicegroups using MBAmap has been introduced which helps error concealment at decoder and improves subjective and objective quality of the received video.
Keywords： FMO H.264/AVC error resilience MBAmap video transmission
1.Introduction
The syntax of H.264/AVC video coding on average reduces 50% bit rate as compared to previous standards keeping similar level of encoder optimization. This makes it an attractive candidate for wireless video transmission. Video transmission in mobile environments also requires a few error resilience features to protect bitstream from possible errors and packet losses. H.264/AVC offers a few error resilience tools， among which FMO relies on how effectively the adjacent macroblocks of a video frame scattered between different slicegroups. In FMO each MB is assigned to a slice according to an MBAmap (Macroblock Allocation Map) consisting of an identification number for each MB in the picture specifying the slice. In this paper， we provide a new pattern in MBAmap of FMO to disperse all possible errors for a better quality of received video.
2. CONVENTIONAL PATTERN
FMO is one of the most important tools introduced in H.264 which can be used to scatter possible errors throughout the frame. By using FMO， the MBs in a frame can be assigned into different slice groups with different patterns. The macroblocks are encoded in a non-scan order in a frame and a scan order in a slice. A slice group map is used for allocating MBs into different slicegroups. The scatter type of FMO in H.264/AVC produces a pattern (MBAmap) by using the following equation.
［i］ = (a + ((b * n)/2)) % (n)(i)
Where ‘i’ is the address of macroblock located in particular column (a) & row (b) and n as the number of slicegroups.
Fig.i.FMO dispersed pattern (H.264/AVC)
The procedure is followed to assign slicegroups to each of macroblock in the frame. This function is known in both， the encoder and the decoder. The pattern produced as a result provides a dispersed ordering where no any macroblock is surrounded by the other MB from the same slicegroup. The pattern for a QCIF sized frame with 8 slicegroups has been shown in Fig. i. The same pattern is followed in all different resolutions and does not change until different number of slicegroups is used. The current pattern lacks some of the features required to disperse errors throughout the frame in case more than one particular slicegroups are lost. There are repetitions of slice group pairs (0， 4)， (1， 5)， (2， 6) and (3， 7) occurring in columns. If any of these pairs is lost， the resulting error is accumulated in columnar area of limited regions as shown for the pair (1， 5) in Fig. i. (grey blocks). In such condition there always left a possibility to use maximum two available MBs for concealment which are adjacent neighbors to a lost MB lying in the horizontal direction. Furthermore， the factor becomes worse if a SG adjacent to the lost pair is also lost.
We define a new type of pattern to get maximum level of randomization in MBAmap keeping number of MBs in each SG equal. In the next section， we present this pattern that fulfills the requirements of a robust MBAmap.
3. PROPOSED METHOD FOR FMO
We improve scattered pattern of FMO by introducing an MBA-map that arranges MBs into SGs without repeating a particular order. We try to provide maximum combinations of distinct slicegroups in the neighborhood of each MB. Let?s suppose we have total 6 slicegroups， all possible combinations of neighboring slicegroups for each MB can be calculated as，
?nCr where n=6-1 &r=4(direct neighbors top， left， right， bottom)?nCr= ?5C4= 5!/(5-4)!4! = 5
There are total 5 possible combinations of neighboring pairs that can be introduced around a SG. For example， slice group ‘0’ can have neighbors as any of following 5 combinations.
(1， 2， 3， 4) (1， 2， 3， 5) (1， 2， 4， 5) (1， 3， 4， 5) (2， 3， 4， 5)
We introduce all of these neighboring pairs in the neighborhood of SG ‘0’ at different locations in the MBAmap (grey blocks in Fig. ii.). The neighboring pairs for other SGs (1 to 6) will follow the same criteria. The resulting MBAmap is shown in Fig. ii which doesn?t repeat a particular pattern and all the SGs almost have same number of macroblocks. It assures that MBs from same SG are put distant to each other， to spread possible errors when this particular SG is lost. In next section we evaluate the performance of these patterns for error resilient video transmission.
4. EVALUATION &RESULTS
For evaluating the objective and subjective quality， we use the sequence ‘foreman’ in a simulated error-prone network. The sequence is with QCIF resolution and a frame rate of 30fps. All the frames are intra coded with a moderate quality quantization parameter (Q=28). Experiments have been conducted for different number of slicegroups. For distinct analysis of the performance of each pattern， we consider Gilbert model with 50% probability that a packet gets lost if the previous packet was lost. Fig. iii shows PSNR values obtained by these two patterns. The proposed pattern shows better performance than the conventional pattern in getting objective quality. The current MBAmap guarantees for a better objective quality with a uniform distribution of errors， which is the main purpose of using FMO as error resilient tool in H.264/AVC. Fig iv. shows frames from sequence ‘foreman’. With a loss of slicegroups among repetitive pairs， the error is limited in columnar areas when H.264 pattern is used but is scattered well in case proposed pattern in MBAmap is employed. Our proposed pattern avoids accumulation of errors in particular regions and the process of concealment finds more error-free MB. It produces less distortion in the recovered block and results in more improved subjective quality.
5. CONCLUSION
In this paper， we propose a dispersed pattern for FMO in H.264/AVC that performs the distribution of MBs into slicegroups in an optimal way. We arrange MBs in the MBAmap with a defined pattern that fully provides isolation of a lost macroblocks decreasing their resultant distortion. The pattern fulfills all the requirements of a robust MBAmap for error resilient transmission of video contents. Our experiments consider the issue where conventional pattern can?t avoid error accumulation in limited areas in case more than one slicegroups are lost and the results show the effectiveness of proposed pattern over it.
(责任编辑：何秀秀)