Proceedings of the World Congress on Engineering 2017 Vol I WCE 2017, July 5-7, 2017, London, U.K. Noise Reduction Techniques for Processing of Medical Images Luis Cadena, Alexander Zotin, Franklin Cadena, Anna Korneeva, Alexander Legalov, Byron Morales Abstract — An application of different techniques for Almost all contemporary image processing involves processing medical images is presented in this paper. The discrete or sampled signal processing. proposed of the techniques consists of application of digital Most of image processing filters can be divided into two spatial filters. Digital filters are used which in the last decades main categories: linear filters and nonlinear filters. Nonlinear have taken great impetus for the treatment of images in different fields of science and in the case of the medical image filters include order statistic filters and adaptive filters. The processing better results are obtained in order to make better choice of filter is often determined by the nature of the task interpretations. and the type and behavior of the data. Noise, dynamic range, Medical images can contain some noise therefore it makes color accuracy, optical artifacts, and many more details affect sense to suppress noise on preprocessing stage. The algorithms the outcome of filter in image processing [13-17]. of most often used filters have been considered, such as mean filter, Gaussian filter, median filter and 2d Cleaner. With the trend toward larger images and proportionally larger filter II. FILTERS kernels, the need for a more efficient filtering algorithm Filtering is a technique for modifying and enhancing an becomes pressing. Conducted comparison of optimized and image. Various filters are used for image preprocessing. The classical implementations of filters algorithms. It shows great primary purpose of these filters is a noise reduction, but filter speed improvement of optimized implementation. can also be used to emphasize certain features of an image or remove other features. In image processing, 2D filtering Index Terms— medical image, image processing, denoising, techniques are usually considered an extension of 1D signal mean filter, median filter, Gaussian 2D filter, 2DCleaner filter processing theory. Almost all contemporary image processing involves discrete or sampled signal processing. I. INTRODUCTION Most of image processing filters can be divided into two main Digital image processing consists of algorithmic processes categories [1-3]: linear filters and nonlinear filters. Nonlinear that transform one image into another in which certain filters include order statistic filters and adaptive filters. The information of interest is highlighted, and/or the information choice of filter is often determined by the nature of the task that is irrelevant to the application is attenuated or eliminated. and the type and behavior of the data. Noise, dynamic range, Thus, image processing tasks include noise suppression, color accuracy, optical artifacts, and many more details affect contrast enhancements, removal of undesirable effects on the outcome of filter in image processing. capture such as blurring or distortion by optical or motion One of the simplest linear filters is a filter which calculates effects, color transformations, and so on. arithmetic mean value of spectrum. The arithmetic mean Filtering is a technique for modifying and enhancing an filter is defined as the average of all pixels spectrum within a image. Various filters are used for image preprocessing. The local region of an image. Pixels that are included in the primary purpose of these filters is a noise reduction, but filter averaging operation are specified by a mask (Kernel). Kernel can also be used to emphasize certain features of an image or size can be different and depends on task. An arithmetic mean remove other features. In image processing, 2D filtering filter operation on an image removes short tailed noise such techniques are usually considered an extension of 1D signal as uniform and Gaussian type noise from the image at the processing theory. cost of blurring the image. Mathematically mean filter can be Manuscript received February 26 2017; This work was supported by described as follows: Universidad de las Fuerzas Armadas ESPE, Av. Gral Ruminahui s/n, 1 RH RW Cnew (y, x) Cold (y dy, x dx) Sangolqui Ecuador (RH 2 1) (RW 2 1) dy RH dx RW L. Cadena is with Electric and Electronic Department, Universidad de las where C , C – new and old values of the image pixels Fuerzas Armadas ESPE, Av. Gral Ruminahui s/n, Sangolqui Ecuador. new old (phone: +593997221212; e-mail: [email protected]; spectrum, respectively; RH, RW – constants defining the [email protected] ). rank of the filter vertically and horizontally. A. Zotin is with Department of Informatics and Computer Techniques, Pixels that are included in the averaging operation are Reshetnev Siberian State Aerospace University, 31 krasnoyarsky rabochу specified by a kernel. The larger the filtering kernel becomes pr., Krasnoyarsk 660014, Russia Federation (e-mail: [email protected] ) F. Cadena is with College Juan Suarez Chacon, Quito, Ecuador (e-mail: the more predominant the blurring becomes and less high [email protected] ) spatial frequency detail that remains in the image. A. Korneeva is Siberian Federal University, 79 Svobodny pr., 660041 In the case of using the descriptions in the form of Krasnoyarsk, Russia ( e-mail: [email protected] ) convolution filter the computation takes the following form: A. Legalov is Siberian Federal University, 79 Svobodny pr., 660041 RH RW Krasnoyarsk, Russia ( e-mail: [email protected] ) Cnew (y, x) Ady RH ,dx RW Cold (y dy, x dx) B. Morales is Universidad de las Fuerzas Armadas ESPE, Av. Gral dy RH dx RW Ruminahui s/n, Sangolqui Ecuador. (e-mail: [email protected]) where A – filter kernel. ISBN: 978-988-14047-4-9 WCE 2017 ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online) Proceedings of the World Congress on Engineering 2017 Vol I WCE 2017, July 5-7, 2017, London, U.K. Defining the KHs as vertical kernel size (KHs=RH×2+1) and y is the distance from the origin in the vertical axis, and σ is KWs as horizontal kernel size (KWs=RW×2+1). The kernel the standard deviation of the Gaussian distribution. coefficients of mean filter are calculated according to Since the image is represented as a collection of discrete formula: pixels it is necessary to produce a discrete approximation to 1 the Gaussian function before perform the convolution. A i, j K K Depends on kernel size and σ some of coefficients can be out Hs Ws Due to arithmetic mean filter property of using equal range of kernel. Theoretically the Gaussian distribution is weights it can be implemented using a much simpler non-zero everywhere, which would require an infinitely large accumulation algorithm which is significantly faster than convolution kernel. In practice it is effectively zero more than using a sliding window algorithm. Thus the accumulation of about three standard deviations from the mean. Thus it is the neighborhood of pixel P(y,x), shares a lot of pixels in possible to truncate the kernel size at this point. Sometimes common with the accumulation for pixel P(y,x+1). This kernel size truncated even more. Thus after computation of means that there is no need to compute the whole kernel for Gaussian Kernel, the coefficients must be corrected that way all pixels except only the first pixel in each row [4]. that the sum of all coefficients equals 1. Once a suitable Successive pixel filter response values can be obtained with kernel has been calculated, then the Gaussian smoothing can just an add and a subtract to the previous pixel filter response be performed using standard convolution methods. The value. Thus, the filter computation can be considered the convolution can in fact be performed fairly quickly since the following way: equation for the 2-D isotropic Gaussian is separable into y 1 RH RW and x components [5] (Figure 2). In some cases the Cold (y dy, x dx), if x 0 K Hs KWs dy RH dxRW approximation of Gaussian filter can be used instead of C new (y, x) 1 RH 1 RH classic version [6,7]. C (y, x 1) C (y dy, x RW 1) C (y dy, x RW), othervise new old old K Hs dy RH K Hs dy RH Figure 1 shows difference in processing time of classical and optimized variant of mean filter on image with size 512×512. Comparison of filters we conducted on following PC: Intel core i5 3.1GHz 8 GB RAM. Fig. 2. Isotropic Gaussian is separable into y and x components Difference in processing time of classical 2D and double 1D implementations of Gaussian filter shown on Figure 3. Fig. 1. Comparison of classical and optimized implementation of mean filter The Gaussian filter (also known as Gaussian blur) is a smoothing filter, which used to blur images and remove detail and noise. In this sense it is similar to the mean filter, Fig. 3. Comparison of classical 2D and double 1D implementation of but it uses a different kernel. The Gaussian filter uses a Gaussian filter Gaussian function (which also expresses the normal distribution in statistics) for calculating the transformation to apply to each pixel in the image. The equation of a Gaussian The best-known order-statistics filter is the median filter, function in one dimension is which, as its name implies, replaces the value of a pixel x 2 spectrum by the median of the spectrum levels in the 1 2 G(x) e 2 neighborhood of that pixel: 2 Cnew (y, x) med Cold (ky,kx) in two dimensions, it is the product of two such Gaussians, (ky ,kx )Kxy one in each dimension: where K – kernel window, with dimensions KHs×KWs x 2 y 2 centered at (x, y).