No-Reference Quality Assessment of JPEG Images via a Quality Relevance Map
S. Alireza Golestaneh, and Damon M. Chandler
This webpage serves as the online supplement of the paper "No-Reference Quality Assessment of JPEG Images via a Quality Relevance Map" submitted to IEEE Signal Processing Letter.
The following supplementary results and analysis of the NJQA algorithm are included on this page:
- More Qualitative Results
- Changing Theta and Length and Computing Quality relevance map
- Computing the correlation with different weight for NJQA
- Analysis of parameters τ1, and τ2
- Download NJQA MATLAB Code
|More Qualitative Results|
|Changing Theta and Length and Computing Quality relevance map|
In this section, for quality relevance map parameters, we compute the map of
a couple of images in good, average, and bad quality by changing the parameters.
As the results show, by changing parameters (θ and Length) with range ±20% the results are so close to each other, and the maps are significantly similar.
|Computing the correlation with different weight for NJQA|
In Eq. (4) of NJQA paper, we assume that areas with structures are more important than nearly uniform areas in estimating image quality. Therefore, in Eq. (4) we give lesser weight to the term which denotes he sum of zero-valued-DCT coefficient in naturally uniform areas.
We provide our results via different coefficients to compute NJQA (with range ±20%) and then provide the PCC correlation for the results of LIVE database.
As we can see, the PCC correlations are very close to each other.
In conclusion, the results are very close when the values are chosen within a ±20% range.
|Analysis of parameters τ1, and τ2|
The analysis of changing τ1 and τ2 are provided on the online supplements of our previous work onhttp://vision.okstate.edu/s3/ (Please see "Analysis of parameters τ1 and τ2" section).