Content-based image filtering
Kudasov, Fedor (2013)
Kudasov, Fedor
2013
Master's Degree Programme in Information Technology
Tieto- ja sähkötekniikan tiedekunta - Faculty of Computing and Electrical Engineering
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2013-08-14
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tty-201308211292
https://urn.fi/URN:NBN:fi:tty-201308211292
Tiivistelmä
This paper presents an adaptive content-based image denoising technique. This technique uses image area classification for two purposes: perform more precise filtering and decrease computation complexity compared to modern filters of the same quality performance.
Overview of several top image filtering techniques was made. Spatial domain (LPA-ICI), transform domain (SW-DCT) and combined filters (SA-DCT and BM3D) were studied in order to understand basic principles of image denoising. Image area classification which gives reasonable division into classes with clearly distinguishable properties for image filtering was observed. We have chosen block-wise classification that maps each block to Texture , Smooth and Edge classes. Performance of discussed filters on image area classes was shown. Adaptive free parameters choise for filtering quality improvement was analysed. It was shown that for some classes best parameters set differs from the best parameter set for the entire image.
Methods to improve denoising algorithms speed which we were using in our adaptive solution were proposed. The most suitable algorithms with appropriate parameters set for each image area class were chosen. Modi ed classi cation algorithm applied to noisy images was developed. Whereupon, a modi ed BM3D-based adaptive denoising algorithm was proposed. Finally, multiple tests were performed and verification of speed and quality performances improvement compared to a baseline BM3D algorithm was obtained.
Overview of several top image filtering techniques was made. Spatial domain (LPA-ICI), transform domain (SW-DCT) and combined filters (SA-DCT and BM3D) were studied in order to understand basic principles of image denoising. Image area classification which gives reasonable division into classes with clearly distinguishable properties for image filtering was observed. We have chosen block-wise classification that maps each block to Texture , Smooth and Edge classes. Performance of discussed filters on image area classes was shown. Adaptive free parameters choise for filtering quality improvement was analysed. It was shown that for some classes best parameters set differs from the best parameter set for the entire image.
Methods to improve denoising algorithms speed which we were using in our adaptive solution were proposed. The most suitable algorithms with appropriate parameters set for each image area class were chosen. Modi ed classi cation algorithm applied to noisy images was developed. Whereupon, a modi ed BM3D-based adaptive denoising algorithm was proposed. Finally, multiple tests were performed and verification of speed and quality performances improvement compared to a baseline BM3D algorithm was obtained.