An Efficient Low Bit-Rate Video-coding Algorithm Focusing on Moving Regions

Kwok-Wai Wong, Kin-Man Lam, Wan-Chi Siu

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, OCTOBER 2001

Introduction Conventional block-based coding schemes

do not take into account the arbitrary shape of moving objects.

Eight patterns are pre-defined to approximate the moving regions in a macroblock.

Experimental results show both the prediction efficiency and the compression efficiency are improved.

Low Bit-Rate Video Coding Using Patterns

The eight patterns for moving region approximation

Moving Region Detection

T: thresholding functionB:structuring element of morphological ope

rations, a square pattern of size 3*3

If < threshold,the value of M(x,y) is assumed to be zero. The value of the moving regions is nonzero.

Moving Region Detection Static MBs (SMBs)

the contents of the MB are all zero or the size of moving region is smaller than a threshold.

Active MBs (AMBs) The contents of all the sub-blocks have a nonzer

o value. Active-region MBs (RMBs)

The remaining MBs. One of the eight predefined patterns will be use

d to approximate the moving regions.

Moving Region Detection

Mk : the Kth MB in the processed frame M(x,y)

PN (i,j) : the pre-defined pattern number N

If the computed minimum Dk,N is greater than a threshold,the MB is assumed to be an AMB.

The rule for classifying the types of MB

AMB RMB

Architecture of the Encoder and Decoder Type of MB

SMB: encoded as a skipped MB AMB: the conventional block-based

encoding method is used RMB: “Pattern Matching” unit

Pattern Matching To find a pre-defined pattern which is the

best representation of the moving region in a RMB.

Uses 3 bits to represent PMB(pre-defined pattern information), and then encoded by VLC.

Architecture of the Encoder and Decoder Block Rearrangement

If RMB, “Block Rearrangement” unit rearrange the residual errors of the moving regions into a block of size 8*8

“Inverse Rearrangement” unit Motion Estimation

AMB: “Block Predictor” RMB: “Moving region predictor” Uses 1 bit to represent MPMODE(motion

predictor mode), together with the motion vector are encoded by VLC

Encoder

Decoder

Motion Estimation and Compensation

The contents of SMB are copied from the reference frame directly.

The matching process is more precise as the static region is neglected in motion estimation.

Conventional method: 256 “minus” ,255 “sum”, 256 “absolute” operations

New: 64 “minus”, 63 “sum”, 64 “absolute” operations

Detected moving regions

Results of pattern approximation

An example of pattern approximation for the sequence “News”

The principle of interframe coding for RMB

Prediction Error Encoding AMB: conventional DCT-based coding

method The prediction errors of the moving

region in a RMB are rearranged to a block of size 8*8, and DCT transform is then employed.

Only a block is needed to encode for a RMB,and the compression ratio is increased.

Rearrangement of the residual errors

Simulation Results

PSNR for the first 100 frames for the video sequence “Akiyo”

Encoding time saved per frame compared to H.263

Simulation results “Akiyo” with different bit-rates

Conclusion Eight pre-defined patterns were

chosen by experiments to represent moving regions.

This approach outperforms the H.263 in terms of the PSNR and the rum-time for sequences of smooth motion.