Purchase this article with an account.
J. W. Warnicki, P. R. Montague; Detection of Compression Artifacts in JPEG Images. Invest. Ophthalmol. Vis. Sci. 2010;51(13):1805.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Ophthalmic digital image files for a single fluorescein angiogram with color fundus photographs can exceed 260 megabytes of data. These data must be stored on local computer systems, and must be transmitted via the Internet to reading centers for evaluation in clinical trials. We evaluated compression methods to reduce the amount of data sent and stored.
Compression factors were calculated for TIFF to PNG and TIFF to JPEG image conversions and the reductions in file size were found to be significant, ranging from 0.4105 (color PNG) to 0.1261 (monochrome highest-quality JPEG). Mathematical subtractions of TIFF and compressed images verified that PNG compression is 100% lossless. At different compression levels, JPEG images were mathematically compared to TIFF images to determine what percentage of the image changed during compression, the mean difference in pixel values that changed, the standard deviation among the changed pixels, and the range of change. In the highest quality monochrome JPEG compression, 40% of pixels were altered with a mean change of 1.08 gray values (SD=0.63) and a range of -3 to +5 gray values.Using custom image comparison software which presents images at full resolution, two sets of images were presented to 15 graders to determine if any differences between JPEG images and TIFF images were visually detectable. The graders compared original TIFF images with images compressed at six different levels. Images sets were presented in random order. A comparison of the TIFF image to itself was included in the set for validity testing.
Of the 30 evaluations (15 graders, two different image sets), 27 showed no observable difference between the highest quality JPEG and the TIFF image. Of the three grades where the grader indicated that a difference could be determined, two were from a grader that also recorded a difference when comparing the TIFF with itself, and one was from a grader who recorded no difference when comparing the next two higher compression levels of the same image.
We conclude that there is no visually detectable difference between highest quality JPEG images and TIFF, and that JPEG images can be reliably used for visual evaluation in clinical and research settings. The effects of compression on automated image analysis involving statistical image processing and pattern recognition was not tested and warrants further investigation.
This PDF is available to Subscribers Only