There have been many tests of image formats earlier before, but this time DSSIM was introduced for the more objective and more scientific comparison between lossy formats including JPEG, WebP and AVIF. Optimisation for 🔞Reconquest at Night was also introduced for reducing colour banding and adding a tremendous shallow depth of field.
DSSIM is a tool for comparing two images and determining the dissimilarity of them. The higher value means the more dissimilar, and 0 means no difference.
The standard for this comparison was JPEG (optimised with MozJPEG) at the quality 75 (so called medium quality) as the same or similar quality is used by many social networking services. The sharp YUV option for WebP was enabled this time for the correct colours overall. Behold that the higher value of quality (quantiser) for AVIF means the worse image.
The WebP image was larger than the JPEG one, and its DSSIM was even significantly higher. And for WebP, Quality 99 was preferred as Quality 100 unexpectedly resulted in the higher DSSIM. AVIF achieved the much smaller file size. Unfortunately, colour banding was visible on all of the lossy images.
The shallow depth of field (DoF) was introduced, but it worsened the colour banding issue. Thus, the grain (noise) adjustment was applied so colour banding was eliminated, but the PNG file size increased 59.51%. JPEG and WebP still saved the similar percentage of file sizes, but AVIF in this test was not that impressive comparing to the previous one. The WebP image remained larger than the JPEG one.
It remains unimpressive that WebP images could be larger than JPEG ones at the similar DSSIM so it seems that AVIF is still the better choice for the future. For artworks, adding the grain adjustment is the effective way to eliminate colour banding, and the shallow DoF can create a phenomenal effect for a background that could be drawn simpler. Artists could experiment with these approaches for the better quality, file size, and overall appearance.