Visual Quality Assessment Databases, MMSP Konstanz
  • Home
  • VQA Databases
    • KonViD-1k Database
    • KonIQ-10k Database
    • KADID-10k Database
    • IQA-Experts-300
    • KonPatch-30k Database
    • KoSMo-1k Database
    • StudyMB 2.0 Database
    • Picture-wise JND Data
  • About
  • Contact

Finding the
intrinsic image resolution

Definition

The intrinsic image resolution is the largest size at which
an image
​​ shows no quality degradation and is of excellent quality.
​Wrong focus, graininess, processing artefacts can detract from the appeal of an image.

Size matters.

​Viewing an image at a smaller size or zoom factor, can make degradation artefacts less visible and annoying, improving the image's perceived quality.
Picture
Shrunk version (20% zoom) of an original image. The quality is much higher, no annoying defects are visible.
Picture
A small crop from the original 6000x4000 pixels photo (100% zoom). The image appears of low quality, blurry, and grainy.
Picture
​A high-resolution, but out of focus, low quality  image is downsized to a lower resolution.

At the intrinsic resolution the image shows no quality degradation, and image details are sharp and clear.

Further downsizing does not help further improve the quality of the image.

What are image quality degradations?

Picture
A pristine quality image (left) is distorted by several degradation types to exemplify possible image degradations.
Degradations make an image look unpleasant, and low-fidelity.

The degradations shown are intentionally exaggerated to a level that would be considered almost always unpleasant, even if the image is shrunk substantially. This means that the intrinsic resolution of the distorted images is in fact very low.

Under normal circumstances, degradations are much less obvious.

Before
​proceeding ...

    Please adjust the zoom level of your browser to 100%. This is usually done by pressing the key combination CMD+0 (or CTRL+0).
    Do not use a mobile phone to view images for this experiment. The screen size is too small to notice the fine details required to judge the optimal resolution of an image.
Submit

How to assess the ​​intrinsic resolution

In the following examples, each line shows crops taken from the same source image. Each crop is extracted after previously zooming out (downscaling) the source image by a different amount. You can often notice strong degradations at the original size (leftmost), whereas the lowest resolution is of excellent quality (rightmost image). The INTRINSIC resolution is often somewhere in between.

​Note: it is best to view each image as is, do not click to enlarge.
slightly blurry
INTRINSIC
similar quality to the one on the left, but lower resolution
very blurry and grainy
still slightly out of focus
INTRINSIC
out of focus
slightly out of focus
INTRINSIC
slightly over-exposed
INTRINSIC
INTRINSIC
INTRINSIC
slightly blurry
INTRINSIC
INTRINSIC
INTRINSIC
over-sharpened
INTRINSIC

Participating in the experiment

You will annotate images using a browser-based interface (Zovi). This allows browsing a collection of images, zooming to select the intrinsic resolution for each, and submitting the selection automatically.

Zovi controls

Zoom: use the zoom slider at the top of the screen ​
Pan: click and drag the image​
Next image: press the green arrow buttons or the right arrow key.


Picture
Zovi annotation interface for selecting the intrinsic resolution.

Steps to complete the work

  1. Open Zovi in your web-browser
  2. ​Select the intrinsic resolution for each image
  3. Move to the next image and repeat​
  4. Submit your results at the end
2. ​Select the intrinsic resolution
  • explore the image to find any visible defects
  • zoom out until the defects are not visible anymore
  • repeat if necessary until no annoying defects are found​
The intrinsic resolution is determined by the largest zoom-level at which ​
​the image quality is excellent, and no annoying defects are visible.

I takes about 15 seconds to find the intrinsic resolution/zoom for each image. 
Thank you for participating in our study!
Imprint | Privacy policy | About | Contact
Copyright "VQA Group at Universität Konstanz" © 2019-2021
  • Home
  • VQA Databases
    • KonViD-1k Database
    • KonIQ-10k Database
    • KADID-10k Database
    • IQA-Experts-300
    • KonPatch-30k Database
    • KoSMo-1k Database
    • StudyMB 2.0 Database
    • Picture-wise JND Data
  • About
  • Contact