07-01-2014, 03:12 PM
AN EDGE-ADAPTIVE BLOCK MATCHING ALGORITHM FOR ERROR CONCEALMENT
AN EDGE-ADAPTIVE BLOCK .pdf (Size: 1.32 MB / Downloads: 22)
ABSTRACT
A widely-used block matching algorithm (BMA) for error
concealment may suffer from the deteriorated quality of a
concealed block that includes multiple objects in different
motion directions. This paper proposes an edge-adaptive
BMA that decomposes a damaged 16x16 macroblock (MB)
into four 8x8 blocks and conceals the 8x8 blocks together
only when they belong to the same object. The edge-
adaptive BMA detects edges on MB boundaries and uses
the number and positions of the edges to determine which
8x8 blocks belong to the same object. The proposed
algorithm improves PSNR by an average of 0.27 dB
compared with the existing BMA for error concealment.
INTRODUCTION
Temporal error concealment is a technique to recover a
damaged area in a video by replacing it with the similar area
in the previous frames [1]. Among many temporal error
concealment methods proposed in literature, a BMA is one
of the widely used techniques. A BMA uses the neighboring
blocks of the damaged block and searches the blocks that
best match the neighboring blocks in the previous frames.
From the best-matched neighboring blocks, a BMA finds
the block that can replace the damaged block [2]. For the
selection of the neighboring blocks used for search
operations, various methods are proposed. Figure 1 (a)
shows the typical neighboring blocks used by a popular
BMA [2]. The square represented by E is the damaged
block of a 16x16 size, often called MB in video
compression standard terminology and the two surrounding
MBs represented by No, Ni are used to find the matched MB
in the previous frames. This technique achieves a good
concealment quality even when the error rate is high or the
movement of objects is fast. However, its search operations
to find the best-matched block require a large amount of
computation.
CASE II: A single edge detected
Figure 2 shows the concealment technique when one edge is
detected. The circle placed on the boundary between the
damaged MB and its neighboring blocks stands for the
position that an edge is detected. For example, Figure 2 (a)
shows that an edge is detected at the upper boundary of
block EO. In this case, block EO is concealed independently
using the upper and left neighboring blocks. The other
blocks, El, E2, and E3 are concealed together using the
remaining neighboring blocks. The number given inside a
neighboring block indicates the blocks used together for
concealment. In Figure 2 (a), the neighboring blocks
numbered '1' are used to conceal block EO and the
neighboring blocks numbered '2' are used together to
conceal blocks El, E2, and E3. All the other cases are
shown in Figure 2. The block that has an edge is concealed
using its two neighboring blocks (represented by '1') and
the other blocks are concealed together by six neighboring
blocks (represented by '2').
EXPERIMENTAL RESULTS
The evaluation uses five CIF-sized video sequences,
Foreman (FM), Table Tennis (TT), Coast Guard (CG),
Stefan (SF), and Mobile and Calendar (MC). These
sequences are encoded and decoded by the H.264/AVC
JMl 1.0 standard software [6]. For all sequences, 100 frames
are encoded at the frame rate of 30 fps.
Figure 7 shows the second frame of Foreman sequence
concealed by various techniques with 25% MB loss rate
(MLR). The 8x8 BMA represents the BMA that conceals
8x8 blocks independently while the 16x16 BMA represents
the BMA that conceals the entire 16x16 MB together [2]. R-
BMA is proposed in [8] and JM 11.0 software employs the
algorithm [9]. For fair comparison, the number of iterations
in R-BMA is limited to one.
CONCLUSIONS
The proposed edge-adaptive BMA uses the detected edges
to decide 8x8 blocks to be concealed together or
independently. Compared with the BMA that conceals the
entire 16x16 MB together [2], an average PSNR is
improved by 0.27dB. The PSNR is improved by 2.01dB
compared with the BMA that conceals each 8x8 block
independently.