Abstract
Determining the compression quality of an image
is important for photo forensics and image enhancement algorithms.
Unfortunately, there are a number of issues involved in
determining the compression quality of an image from its metadata
or quantization tables. A compression quality estimation
algorithm based on visual inspection of detected compression
artefacts is presented. This method detects and extracts feature
samples around compression block corners. These feature
samples are then pre-filtered to enhance the discontinuities
produced by compression artefacts. The feature samples are then
classified using a constricted Neural Network. The local quality
estimations are then combined using robust statistics to estimate
the maximum likelihood compression quality. This method was
shown to accurately estimate the compression quality of an image
without prior knowledge of the original uncompressed image.