Adaptive filtering techniques for acquisition noise and coding artifacts of digital pictures

Yang, J 2010, Adaptive filtering techniques for acquisition noise and coding artifacts of digital pictures, Doctor of Philosophy (PhD), Electrical and Computer Engineering, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Yang.pdf Thesis application/pdf 5.54MB
how_to_use_jviewer.txt Jviewer instructions Click to show the corresponding preview/stream 403Bytes
jviewer.exe Jviewer application Click to show the corresponding preview/stream application/octet-stream 280KB Images for chapter 6 Click to show the corresponding preview/stream application/zip 28.99MB
Title Adaptive filtering techniques for acquisition noise and coding artifacts of digital pictures
Author(s) Yang, J
Year 2010
Abstract The quality of digital pictures is often degraded by various processes (e.g, acquisition or capturing, compression, filtering process, transmission, etc). In digital image/video processing systems, random noise appearing in images is mainly generated during the capturing process; while the artifacts (or distortions) are generated in compression or filtering processes. This dissertation looks at digital image/video quality degradations with possible solution for post processing techniques for coding artifacts and acquisition noise reduction for images/videos.

Three major issues associated with the image/video degradation are addressed in this work. The first issue is the temporal fluctuation artifact in digitally compressed videos. In the state-of-art video coding standard, H.264/AVC, temporal fluctuations are noticeable between intra picture frames or between an intra picture frame and neighbouring inter picture frames. To resolve this problem, a novel robust statistical temporal filtering technique is proposed. It utilises a re-descending robust statistical model with outlier rejection feature to reduce the temporal fluctuations while preserving picture details and motion sharpness. PSNR and sum of square difference (SSD) show improvement of proposed filters over other benchmark filters. Even for videos contain high motion, the proposed temporal filter shows good performances in fluctuation reduction and motion clarity preservation compared with other baseline temporal filters.

The second issue concerns both the spatial and temporal artifacts (e.g, blocking, ringing, and temporal fluctuation artifacts) appearing in compressed video. To address this issue, a novel joint spatial and temporal filtering framework is constructed for artifacts reduction. Both the spatial and the temporal filters employ a re-descending robust statistical model (RRSM) in the filtering processes. The robust statistical spatial filter (RSSF) reduces spatial blocking and ringing artifacts whilst the robust statistical temporal filter (RSTF) suppresses the temporal fluctuations. Performance evaluations demonstrate that the proposed joint spatio-temporal filter is superior to H.264 loop filter in terms of spatial and temporal artifacts reduction and motion clarity preservation.

The third issue is random noise, commonly modeled as mixed Gaussian and impulse noise (MGIN), which appears in image/video acquisition process. An effective method to estimate MGIN is through a robust estimator, median absolute deviation normalized (MADN). The MADN estimator is used to separate the MGIN model into impulse and additive Gaussian noise portion. Based on this estimation, the proposed filtering process is composed of a modified median filter for impulse noise reduction, and a DCT transform based denoising filter for additive Gaussian noise reduction. However, this DCT based denoising filter produces temporal fluctuations for videos. To solve this problem, a temporal filter is added to the filtering process. Therefore, another joint spatio-temporal filtering scheme is built to achieve the best visual quality of denoised videos. Extensive experiments show that the proposed joint spatio-temporal filtering scheme outperforms other benchmark filters in noise and distortions suppression.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Electrical and Computer Engineering
Keyword(s) Temporal Filtering
Video Denoising
Post Filtering
Robust Statistics
Version Filter Type
Access Statistics: 733 Abstract Views, 1446 File Downloads  -  Detailed Statistics
Created: Fri, 08 Apr 2011, 12:18:03 EST by Guy Aron
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us