A Novel Noise Estimation Algorithm Based on Quadtree Decomposition

• Main Authors: Jun-Hung Wang, 汪俊宏
• Other Authors: Shen-chuan Tai
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/55424846219739249962

碩士 === 國立成功大學 === 電腦與通信工程研究所 === 95 === Digital images may be contaminated by noise in various steps such as image acquisition, recording, compression, and transmission through faulty camera sensors or noisy communication channels. In general, noise would corrupt digital images, cause some loss of information details, and degrade the image quality severely. Therefore, noise reduction has become more imperative to suppress noise quantities and recover the corrupted pixels. Moreover, in order to improve the performance of noise removal, it may combine other image processing steps, like edge detection, image segmentation, or object recognition. For many noise reduction algorithms, either they are implemented in frequency or in space domain, the noise level is assumed to be known in priori, which is not valid in practical cases and influences the filtering performance hugely. Due to this fact, noise estimation is a required research topic beforehand to apply actual noise removal operations to noisy image. Usually, noise estimation algorithms in the spatial domain are categorized in two main approaches: block-based and filter-based. In this thesis, we address an novel algorithm which combines quadtree decomposition with both the two foregoing approaches to estimate the noise level with acceptable computational load. The simulation results show that our algorithm has more outstanding performances in comparison with several existing algorithms for highly noisy and good-quality digital images. Additionally, the proposed algorithm also could be applied to noise removal in commercial image-based or video-based applications such as digital cameras and television for its performance and simplicity.