Talk:Error Level Analyser

From Tiny Appz

Noah: This is excellent! Here's a few thoughts: (1) The resave error level should be less than the original image quality. For example, if the original was saved at 90%, then you should resave it with nothing higher than 89%. And don't go too low (e.g., from 90% to 25%) or everything will look like it has changed. A resave that is 5% to 25% less than the original should be a wide enough range, with around 10% being best. (These are my own qualitative findings.) (2) My own tool (jpegana) treats the R, G, and B channels separately. However, I like how you combine them to generate a temperature. (3) Consider combining your tool with an edge detection algorithm. Regions of high adjacent contrast (think "Zebra") will appear to have a higher error level than the rest of the image. Edge detection algorithms allow you to identify these areas.

- Neal Krawetz