Master of Science
Patch-based approaches such as Block Matching and 3D collaborative Filtering (BM3D) algorithm represent the current state-of-the-art in image denoising. However, BM3D still suffers from degradation in performance in smooth areas as well as loss of image details, specifically in the presence of high noise levels.
Integrating shape adaptive methods with BM3D improves the denoising outcome including the visual quality of the denoised image; and also maintains image details. In this study, we proposed a framework that produces multiple images using various shapes. These images were aggregated at the pixel or patch levels for both stages in BM3D, and when appropriately aggregated, resulted in better denoising performance than BM3D by 1.15 dB, on average.
Summary for Lay Audience
Noise in images usually occurs during image acquisition and/or transmission, when image information can be lost, and because of this, research in denoising digital images focuses on improving image information. BM3D the most prominent image denoising algorithm for the last decade, it utilizes a process of searching for matching patches to improve image quality. Similar to BM3D, Our proposed framework uses different shapes next to the square shape to improve patch matching. However, Instead of obtaining only one output image, our framework combines various obtained images. The combination of these images improves the numerical and visual quality of the denoised image to a greater extent than BM3D.
Massoud, Mena Abdelrahman, "Framework For Kernel Based BM3D Algorithm" (2020). Electronic Thesis and Dissertation Repository. 7314.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.