How to measure the restored MB in corrupted picture?
BMA(boundary matching algorithm):
Very simple strategy, just to get the absolute value sum(or variation) of external and internal pixel . The best matching MB is with the smallest sum.
Easy but with low computation complexity!
OBMA(Outer BMA)
Difference is both outer of corrupted and ref MB respectively.
STBMA(spatial-temporal BMA)
A little improvement of BMA, the distortion is calculated by two parts:
the absolute sum of ref out MB pixel with cur corrupted out MB pixel;
use Laplacian estimator to calculate horizontal and vertical direction sum by cur corrupted external MB pixels.
Then combine these two parts together, using a weighted sum.
This method takes temporal information into account.
DSTBMA(directional STBMA)
Add diagonal and anti-diagonal information to get spatial boundary.
Edge Direction Estimation:
Employ Sobel Operator on pixels surrounding the lost MB external edge to get the direction and if the max number of one direction overhead the Threshold, just think the four 8x8 block has a dominant direction. Otherwise dealing with four blocks independently!
SSIM
As above criteria uses MSE or MAE, just change selected block. But this method uses a new criteria——human vision.
NEMA
Neighbors Matching Criterion.
Using valid extension pixel of corrupted to select MV compensation method.
Above estimation methods have their limits!
Most of concealment techniques are suitable for FMO and in hardware, FMO not only hard to achieve but also needs lots of sources and time.