Artifact Detection FAQ

Artifact Detection FAQ

Q: I'm getting a score of about 20 for DC Blockiness, yet I don't see much blocking artifacts. Why not?

A: Make sure there is not a confusion of terms in this case: DC Blockiness is not a measure of percieved blocking artifact. It is an objective measure that quatifies how much of the signal from each processed block has been reduced to DC. For MPEG-2 where there is no deblocking filters in the standard decoder, DC blockiness tends to approximately correlate with percieved blockiness, For H.264, with deblocking filters, percieved blocking may be much lower DC blocking. In both cases, if the block grid is not well distributed throughout the picture, the DC-blockiness algorithm will less effective.

Blocking artifacts that we see have components that can be detected by all four types of artifact detection:

  • added edges along the block boundaries
  • removed edges inside the blocks
  • rotated edges along the block boundaries However, rotated edges tend to be more unique to blocking artifacts since other examples of added edges, such as mosquito noise, and lost edges, such as overall blurring can occur.
  • DC blockiness is a measure of how much variance is lost on average in processed blocks. This includes reduced lost edges, blurring, reduced contrast and other ways of describing lost variance (or "AC" signal) within the block.

    So here are some ways that you can get a small to moderate DC blockiness measurement result while not seeing any blockiness whatsoever:

  • Gain reduction: If the overall gain of the picture is reduced (say by a "contrast" control), both the AC and DC are reduced and the image just looks a little less bright and maybe has less contrast (depending on how the brightness control is set), and no blocking artifact is seen, but the DC blockiness will correctly measure that there has been a reduction of signal towards the mean of each block.
  • Actual contrast reduction: similar to gain change, except the average block values remain the same while the variance is reduced. The brights and darks within the block are closer to the mean brightness of the block. If contrast is reduced somewhat uniformly throughout the image, no blocking artifacts will be seen, but the DC blockiness will correctly measure that there has been a reduction of signal towards the mean of each block.

    For the problem you have described, assuming blocking is not uniform across the frame, I would like to suggest that you might try checking the

  • rotated (and/or added) edges measurement as an alternative objective measurement and
  • rotated (and/or added) edge artifact weighted DMOS (or PQR) as a predictor of relative subjective impact of this artifact.

    The rotated edges artifact measurement tends to pick up jaggies, where diagonal edges have been rotated to the horizontal or vertical. When combined with the human vision perceptual model, the measurement predicts how much of these rotated edges are seen.

    Copyright © 2008 Kevin Ferguson/
    All Rights Reserved.

    Any redistribution of information found at this site is prohibited.