Abstract— Number of works covers a topic of denoising of digital images affected by an additive white Gaussian noise (AWGN), which are formed by typical devices that contain of lenses and semiconducting sensors which capture a projected scene. These constructive elements inevitably add numerous distortions, degradation and noise. Some tasks require high- quality digital images, this leads to the development of denoising algorithms which also sharpen an image and perform its colour correction. In this paper we present our results of applying several image filtration algorithms based on Principal Component Analysis (PCA) and non-local processing. Work is focused on a discussion of experimental data which is aimed to uncover best practices of use for the studied filtration algorithms. Index Terms—Image filtration, principal component analysis, non-local processing, applications I. INTRODUCTION S it was shown by Chatterjee and Milanfar in 2010, the theoretical limit of image reconstruction hasn’t been yet achieved [1]. There still are debates on how to increase performance of filtration techniques used today. Among the widest spread methods of cancelling an AWGN in digital images, according to [2], are the algorithms which base on: (1) local processing, (2) non-local processing, (3) pointwise processing and (4) multipoint processing. All the named methods evolved through the years and now each of them has sophisticated implementations which compete with each other on the test metrics. That is why most of researchers consider their own evaluations of image reconstruction for specific textural, edge and contrast regions. The main problems with the quality of reconstructed images which researches try to evaluate are: a Gibbs effect, which becomes highly noticeable on images containing This work was supported in part by the Russian Foundation for Basic Research under Grant № 12-08-01215-а "Development of methods for quality assessment of video". A.L. Priorov is with the Yaroslavl State University, Yaroslavl, Russia 150000 (e-mail: andcat@yandex.ru). K.I. Tumanov is with the Yaroslavl State University, Yaroslavl, Russia 150000 (corresponding author, phone: +7-910-814-2226; e-mail: tumanov@susqu.edu). V.A. Volokhov is with the Yaroslavl State University, Yaroslavl, Russia 150000 (e-mail: volokhov@piclab.ru). E.V. Sergeev is with the Yaroslavl State University, Yaroslavl, Russia 150000 (e-mail: sergeev@piclab.ru). I.S. Mochalov is with the Yaroslavl State University, Yaroslavl, Russia 150000 (e-mail: yar_panda@yahoo.com). objects with high brightness contrast on their outer edges, and an edge blurring of objects on an image being processed. Both of these effects highly degrade an image perception and could not be suited for high demands. List of the most successful solutions to the stated problems includes the following digital image reconstruction algorithms: (1) algorithm based on block-matching and 3D filtering (BM3D) [3]; (2) algorithm based on shape-adaptive discrete cosine transform (SA-DCT) [4]; (3) k-means singular value decomposition (K-SVD) [5]; (4) non-local means algorithm (NL-means) [6]; (5) algorithm based on a local polynomial approximation and intersection of confidence intervals rule (LPA-ICI) [7]. In our previous work [8], we proposed a parallel filtration scheme algorithm based on PCA and non-local processing. In the present study we compare sequential and parallel filtration schemes in terms of their work principles and the results of their use in modern digital image filtration tasks. Literature on digital images noise cancelling shows that modern AWGN filtration methods used for greyscale images may be successfully transferred to other digital image processing tasks. So, this work in addition to the primary use of the methods shows how they may be used for: (1) denoising AWGN-affected colour images; (2) filtration of mixed noises from greyscale images; (3) suppression of blocking artefacts in compressed JPEG images; (4) filtration of mixed noises from colour images. Usage of the AWGN model may be explained with the help of statistics theory, namely – central limit theorem. It has an important practical value and is suitable for describing the work of devices containing numerous independent additive noise sources, each of which has its own random distribution, which may be unknown. Resulting sum of these noise distributions is best described as a Gaussian distribution. On practice AWGN model well suits to simulate a thermal noise which is inevitably observed in digital devices such as charge-coupled devices (CCDs) or CMOS matrixes. Filtration of colour images is an issue of the day for various practical applications. That is why there are numerous solutions to it. As a possible approach, in this work we did no transition from RGB image to an image with separated brightness and colour information, and added an AWGN separately to each channel with the same characteristics. This method was used for simplicity and for further research it may be extended by using specific noise models and applying them to each image layer in a variation of interest. Applications of Image Filtration Based on Principal Component Analysis and Nonlocal Image Processing Andrey Priorov, Kirill Tumanov, Member, IAENG and Student Member, IEEE, Vladimir Volokhov, Evgeny Sergeev and Ivan Mochalov A IAENG International Journal of Computer Science, 40:2, IJCS_40_2_02 (Advance online publication: 21 May 2013) ______________________________________________________________________________________