Image quality index details

[Johnny]

Hi,

I've been testing the image quality index tool, which I think is very useful when processing video frames. Based on the information given I know it is based on texture characteristics, such as heterogeneity/richness and blurriness. However I don't really understand where this index comes from, and also in the map there are red signs only where texture-less areas are, but nothing about blur-motion effects. Finally, I know there is not an specific range but is there any top value, which would represent a highest image quality?

Best regards,
Johnny

[Roberto]

Hi Johnny,

the image quality index tool produces low/red values both for textureless and blurry areas. It doesn't really distinguish textureless area from and blurry ones, since both of them will equally give you bad results in Zephyr. It benefits from some of the algorithms that Zephyr uses during the reconstruction process to make a plausible quality estimation. Please also note that the quality index is not an absolute value. Basically, you will usually get higher values when the image resolution is higher, but an higher image resolution doesn't necessarily mean an higher image quality. It is very useful to find the worst/best images from a set of pictures taken with the same sensor/resolution.

[Johnny]

Hi Roberto,

Thank you very much. My problem is that I have some images which have a low quality index value (0.8) and are in general blurred (the whole image), however red zones are small. Well but I think this is mainly related to the range since also in average the score is 2.35 for the whole data set. I have one more question please, is image blurriness then measured based on image comparison between one image and the whole data set, or by a single-image analysis algorithm in order to detect edges for example.

Best regards,
Johnny

[Roberto]

The red/green representation is relative to the considered image and it's just a visual reference to highlight the worst/best parts of the considered image. The quality index value is more significative than the visual representation in order to evaluate the overall quality of an image. The index is computed in an independent way for each image by using specific image-processing algorithms. It's not a direct edge detection algorithm, but a bit more complex multiscale approach with frequency analysis.

[Johnny]

Hi Roberto,

Please clarify me this: Is this index dependent on the image resolution? Since I want to compare two data sets of different resolution and according to what you told, me the higher the resolution the higher the range of values. then I imagine the one with higher resolution will have a higher mean value, even when image quality in general is not that good. In summary, it will work when you analyze a data set with the same resolution, but not for comparing data sets of different resolutions, right?

Kind regards,
Johnny

[Roberto]

Yes, that's right. The easiest thing you can do if you want to compare data with different resolution is to downscale the high resolution pictures to match the lower ones. If their quality indices are still higher, they are certainly better. However, if their quality index is lower, you couldn't really tell which one is better (because they have a lower quality "per pixel", but in theory the high original pictures contain more information).