A New Similarity Measure for Non-local Means Denoising
Non-local means (NLM) denoising algorithm is a good similarity measure based denoising algorithm for images with repetitive textures. However, NLM cannot handle the large rotation. In this paper, we propose a rotation-invariant and noise-resistant similar
- PDF / 834,912 Bytes
- 11 Pages / 439.37 x 666.14 pts Page_size
- 24 Downloads / 236 Views
Institute of Intelligent Machines, Chinese Academy of Sciences, Hefei 230031, Anhui, China [email protected] 2 School of Information Science and Technology, University of Science and Technology of China, Hefei 230027, Anhui, China
Abstract. Non-local means (NLM) denoising algorithm is a good similarity measure based denoising algorithm for images with repetitive textures. However, NLM cannot handle the large rotation. In this paper, we propose a rotationinvariant and noise-resistant similarity measure based on improved LBP operator, and use it to search for similar image patches. In addition, in order to speed up the algorithm, an automatic selection strategy of similar patches is proposed. Consequently, the self-similarity can be used to obtain more similar patches for denoising. Experiment results demonstrate that the proposed method achieved higher peak signal-to-noise ratio (PSNR) and more visual pleasing results than some state-of-art methods. Keywords: Rotation-invariant · Similarity measure · NLM · PSNR
1
Introduction
The goal of image denoising methods is to recover the original image from a noisy measurement. Several methods have been proposed to remove the noise and recover the true image. Most of them can be divided into two parts, spatial filtering algorithm and transform domain filtering algorithm. The former mainly includes the mean filtering, median filtering, wiener filtering and non-local means (NLM) filtering, etc [1-3]. The latter mainly includes wavelet threshold filtering [4-6], and filtering method based on dictionary learning [7-9], etc. The NLM [3] algorithm extends the local calculation model to nonlocal and it has been proved to have better performance than other classic denoising algorithm. This denoising filter searches similar patches and uses them in a weighted average, which the weights depend on the amount of similarity. So, the similarity measurement between patches is the most important part in NLM denoising algorithm. In order to obtain better filtering performance, many researchers have conducted the thorough research on the basis of NLM [10-15]. By sparse 3D transform-domain collaborative filtering, the BM3D algorithm obtains very good filtering effect. For the research on the speed of operation, researchers mainly use the pre-selection method [16-17]. Although these methods in a certain extent, improve the filtering performance, there are still some shortcomings. Most improved filtering algorithms cannot handle rotation or mirroring. © Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part I, CCIS 546, pp. 306–316, 2015. DOI: 10.1007/978-3-662-48558-3_31
A New Similarity Measure for Non-local Means Denoising
307
Local binary pattern (LBP) operator was proposed by Ojala et al. [18]. Although it can capture the very local structure of the texture, the original LBP codes are sensitive to noise and image rotation. Therefore, we propose an improved LBP operator, and put forward an improved method for searching for similar image patches on the basis of th
Data Loading...
