1 July 2002 Univariant assessment of the quality of images
Author Affiliations +
To evaluate the quality of images, most methods compare a degraded image to a perfect reference. Nevertheless in many cases, a reference does not exist. We propose an original univariant (i.e., without a reference) method based on the use of artificial neural networks. The principle behind it is to first teach a neural network to assess image quality using images taken from a pool of known examples, then use it to assess the quality of unknown images. The defects considered are compression artifacts, ringing, local singularities, etc. To simplify, only images with defects that are not mixed with one another were first used. Two illustrative examples are presented: assessment of the quality of JPEG compressed images and detection of local defects. The quality of the images is assessed without a reference and with error less than 6%–7% compared to the bivariant method that was learned. Our method can even be used to model some very simple visual comportment.
© (2002) Society of Photo-Optical Instrumentation Engineers (SPIE)
Mathieu Jung, Mathieu Jung, Dominique Leger, Dominique Leger, Marc G. Gazalet, Marc G. Gazalet, } "Univariant assessment of the quality of images," Journal of Electronic Imaging 11(3), (1 July 2002). https://doi.org/10.1117/1.1482096 . Submission:


Univariant assessment of the visual quality of images
Proceedings of SPIE (June 01 2000)
Contrast gain control for color image quality
Proceedings of SPIE (July 16 1998)
New hybrid zerotree/DPCM image coder
Proceedings of SPIE (September 24 1998)

Back to Top