Introduction
Blocking artifact is the most obvious artifact associated with JPEG (or other
block-based coding), especially at very low bit rate. Some post-processing methods have
been proposed to reduce this annoy artifact by smoothing, which, unfortunately, will incur
blurring artifact. Different implementations may use different strategy to balance the
trade off. To compare the performance of different implementations, we need to find a way
to evaluate the distortion in processed image.
Subjective tests are the most reliable but they are too time-consuming for wide
application. Although criticized for not taking into account of the property of human
vision system (HVS), MSE and its inherences (PSNR or so) are still the dominant objective
image quality metrics in the field. Many other objective image quality metrics, which
imitate more or less of HVS, have been proposed. As we know, there is still no such a
magic objective image quality metrics that can solve most of the problems.
Most successful metrics are targeted at a specific (or limited) application filed. Since
blocking and blurring are the most common artifacts in JPEG, it will be very helpful if we
can find a better objective distortion metrics that is more closely related to our
subjective impression of blocking and blurring artifacts.
In this project, I
1) Implement several image quality metrics: RMSE, Liu et als BMR(blocking-to-masking
ratio) and Eskicioglus EOBD
2) Evaluate them through a simple but very efficient subjective test.
3) Propose a new metrics, which is a combination of pixel-based Root-Mean-Square-Error
(RMSE) and block-based distortion metrics. This metrics proves superior to others.