| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系所 === 100 === Digital images are often corrupted by noise during image captured by digital cameras due to many factors, including temperature, exposure time encountered or image sensor (e.g. CCD). To overcome this problem, many methods were proposed. In this study, we focus on high ISO noise. High ISO noise image are affected by Gaussian noise and low frequency noise. It has a coarse-grain noise characteristics that called low frequency noise. It is difficult to distinguish between real signal and low frequency noise. To reduce low frequency noise, a denoising method is proposed to use multiresolution framework. Therefore, it have been used to analyze image at low resolution layer and high resolution layer . To use multiresolution characteristics to eliminate noise. Thus, using the wavelet transform to decompose the noise image into low and high frequency components and reduce noise, respectively. To reduce noise effectively and preserve image edge.
In this study , we propose a multiresolution local extrema filtering. It can separates into noise and image effectively than bilateral filter. And we use a two-level extrema-based multi-scale decomposition(EMD) framework to extract more detail. The detail layer contains edge as well as noise.The edge extraction was added to the initial denoised image. In experiment result, we can observe that our proposed method efficient remove noise while preserving image edge.
|
|---|
