Click here for full text:
Characterizing and Estimating Block DCT Image Compression Quantization Parameters
Samadani, Ramin
HPL-2005-190
Keyword(s): compression; artifact reduction; image processing; sequential estimation
Abstract: This paper describes an algorithm for estimating quantization matrices from raster images previously compressed with JPEG. First, the space of commonly used quantization matrices is statistically characterized using over 15000 image files. The insights from this characterization are used to design the algorithm. The two stage algorithm first applies a new sequential estimation process to determine some individual Q matrix entries, and then applies shape- gain vector quantization to recover complete quantization matrices. Low average absolute error values are found for 124 images not used during training. In addition, the estimated Q matrices work well when applied to compression artifact reduction. Notes:
5 Pages
Back to Index
| 
Printable version
