Аннотация и ключевые слова
Аннотация (русский):
Объектом исследования является совершенствование и оценка качества цифровой обработки изображений с позиции устранения нежелательных дефектов (артефактов) в виде стробоскопических эффектов. Предложены новые способы оценки стробоскопических эффектов на изображениях, основанные на выделении контуров методами Canny и Prewitt. Рассматриваются возможности оценки стробоскопических эффектов на растровых изображениях, представленных цветовой моделью RGB. Оценка эффектов строба осуществляется путем сравнения контуров, выбранных методами Canny и Prewitt для отдельных цветовых компонентов (красный, зеленый, синий) и изображений серой шкалы. На выбранных контурах определяются их минимальные и максимальные координаты, по которым границы контуров дифференцируются. Как показали эксперименты, этот подход может служить численной метрикой для эффектов строба на растровых изображениях. Другой предложенный подход при оценке эффектов строба заключается в сравнении площади кругов, полученных из предыдущих расчетов. Отношение меньшей площади к большей площади может служить числовой метрикой для оценки эффектов строба на растровых изображениях. Предлагаемые метрики сопровождаются численными расчетами и графическими иллюстрациями.

Ключевые слова:
цветовая модель RGB, Canny, Prewitt, контурные границы, метрика, эффекты строба, артефакты, растровые изображения, двумерные матрицы изображений
Текст
The tasks of digital image processing are very diverse. First of all, it concerns the improvement of image quality. In this direction, a lot of experience is accumulated, both in the development of filters and their application. There are also tasks to assess the quality of images. This can be done both on the basis of comparison with existing samples, and on the basis of an analysis of non-standard (no-reference) measures to assess the quality of digital images [1, 2]. The emergence of such tasks is due to the fact that the images are different types of unwanted defects (artifacts). The problem of finding artifacts on images is quite relevant today. Particularly distinguished are the tasks of analysis without reference measures to evaluate the quality of digital images, i.e. in the absence of a comparable reference image. In this case, it is necessary to take into account the nature of the artifacts. Some progress has been made in non-preference evaluation of blocking artifacts or blocking artifacts on raster images. In this direction, there are certain results noted in [2-9]. At the same time, certain problems create artifacts in the form of strobe effects. The stroboscopic effect is a phenomenon of human visual perception. Fixation of strobe effects is expressed as blurred duplication of the main image. In this regard, there are tasks not only to eliminate strobe effects, but also to determine their computer vision systems. In addition, it is desirable to have a numerical measure of the value of strobe effects on the images. It is the solution of this problem that is devoted to certain studies in this paper. The proposed approach in obtaining a numerical metric for evaluating strobe effects is based on the consistent application of contour filters Canny and Prewitt. In addition, contours are selected by Canny and Prewitt methods for each channel of a full-color image, i.e. red, green, blue. Formulation of the problem In [2-9], methods and techniques for determining artifacts on raster images are considered. In works [5-8] it is noted that many ways to identify artifacts are sensitive to the type of artifacts. The algorithms described in [2-4] are quite good at evaluating the blocking artifacts, which are widespread. Computer experiments carried out on these algorithms have shown that they are insufficient to evaluate other artifacts; in particular, strobes provoke artifacts. Therefore, specialized algorithms and methods should be developed to evaluate the appropriate types of artifacts in images. In this paper, the author's opinion suggests new ways to evaluate strobe effects in images. Moreover, this assessment follows the approach of non-preference evaluation of artifacts [1-4]. The proposed approach is based on the allocation of image contours using the methods of Canny and Prewitt. A better method, the Canny contour detector allows you to determine all the significant changes in brightness in the image. The Prewitt method also highlights the outline of the image, but it seems to ignore many small details in the images. Here you can note other methods, for example, Sobel method. But carried out experimental studies have shown that to solve the task of estimating strobe effects, the most suitable methods were the methods of Canny and Prewitt. Solutions to tasks for image processing based on the separation of contours, analysis of artifacts are used in various fields [10-12]. This suggests that the tasks of image processing are relevant at the present time. In this paper, the problem of computer detection of artifacts such as strobe effects is considered and solved. The first suggested approach to evaluating strobe effects on images is to apply Canny and Prewitt detectors sequentially for each color channel of a full-color image. As shown by model experiments, the results of the selection of the main contours of the image are significantly different for the methods of Canny and Prewitt. This difference was the basis for the evaluation of strobe effects. Applying the Canny and Prewitt methods to individual color channels of the image allows you to obtain binary images. In this case, the visible boundaries are encoded by value one and invisible (black) zero. Further, the coordinates of the boundaries are calculated, their extreme values. The corresponding differences of the horizontal and vertical boundaries obtained by the methods of Canny and Prewitt are adopted as metrics for estimating strobe effects in the image. These operations were implemented in the MATLAB system. Elements of the program code will be given below in the text. The following proposed method for determining the artifacts of stroboscopic effects also uses the methods of Canny and Prewitt. In this case, a full-color RGB image is converted to a grayscale, gray image. Then, the radii of the outer contours of the binary images are determined. From these radii, the smallest is determined and the corresponding circle is constructed. The ratio of the smaller area of the corresponding circle to the larger area of the other circle is used as a metric of strobe effects. If there are no strobes effects on the image, this ratio (metric) approaches unity. Numerical results of analysis and computer simulation of strobe effects estimation Verification of the efficiency of the proposed approaches for evaluating strobe effects on bitmaps was carried out on a number of test images. Among them were images with obvious strobe effects, as well as images with blocking artifacts and ordinary, «clean» images. This choice is stipulated by the fact that analysis of blocking artifacts was described in works [2-8]. In particular, the NPBM method, which considers the Perceptual Blockiness Metric and the algorithm of which is described in [2], does not give a positive result in determining the artifacts of the strobe effect type. The author of the NPBM algorithm was implemented in the MATLAB system and on the Win32 API of C ++. To evaluate blocking artifacts, both software implementations give similar positive results. Examples of images with strobe effects are shown in Fig. 1-3 (in the article, these images are presented as grayscale or gray images, accordingly, their scale was changed). Fig. 1. Strob1 Fig. 2. Strob2 Fig. 3. Strob3 Looking at the presented figures in Fig. 1-3, it can be noted that horizontal strobe effects predominate in the first two drawings, and in Fig. 3 there are elements of the vertical strobe effect. In this connection it makes sense to determine the estimates of the strobe effects horizontally and vertically of the given image. In the future, the proposed estimates of strobe effects will be called the strobe effect metrics on raster images represented by the RGB color model. The basis of image analysis with strobe effects strobe put artifacts selection border «flat» images by methods Prewitt and Canny [13, 14]. As shown in [15], the selection of boundaries using the Prewitt method gives certain advantages. Figure 4 shows the combined image obtained after the image processing methods in Fig. 1 isolation boundaries Canny and Prewitt. Fig. 4. The result of the image processing Strob1 Canny and Prewitt methods Image processing is performed after converting RGB color image into a grayscale image. As seen from Fig. 4, according to the method Prewitt fixed main image boundary. Already at this stage it is possible to determine the coordinates of the minimum border contours allocated to compare between these coordinates for Canny method and Prewitt method. More precise boundaries can be obtained by processing each color channel of an RGB image. The approach to decomposition of the image into its color components was used in [8], where the task of detecting blocking artifacts on raster images was solved. More precise boundaries can be determined programmatically after the decomposition of the color channels of the original full-color image. Example Strob1 blue component image shown in Fig. 5. Fig. 5. Blue component image processing Strob1 Canny and Prewitt methods For other images, the red or green components of the RGB image may be the more defining color components, respectively. As can be seen from Fig. 5, the left picture contains the artifacts of the effect strobe, and the right one is bordered only by the base image, i.e. images without strobe effects. In this case, it is possible to isolate the boundary values of the image coordinates with strobe effects and the boundary values of the coordinates, as it were, without strobe effects. The difference between them will indicate the magnitude of the so-called strobe effect metric. Since after the image processing methods such as Canny and Prewitt obtained binary image matrix, some of these matrices can be determined all the nonzero elements, which then define the minimum coordinate values with nonzero elements. Let us elaborate on the numerical definition of the proposed metrics for evaluating strobe effects. Using the syntax of the programming language of the MATLAB system, you can write the following program instructions: - [xc, yc] = find(cny); % Canny; - [xp, yp] = find(prwt); % Prewitt; - XcMin = min(xc); YcMin = min(yc); - XpMin = min(xp); YpMin = min(yp); - MetricH = abs(XcMin - XpMin); - MetricV = abs(YcMin - YpMin). This fragment of the program code can be used as a «gray» image, and for the channels of red, green, blue. Probably, it is obvious that if the strobe effects in the image are located on the right relative to the main part of it, then it is necessary to determine the maximum values of the coordinates of the contour boundaries. The same can be said if the strobe effects are positioned vertically relative to the main image. Then, in the first place should be treated the upper and lower boundary contours obtained by Canny and Prewitt methods. It should be noted that the proposed approach to determine the metric of a bitmap image with strobe effects does not distinguish blocking artifacts, identifying them as a «good» image. For example, take the images shown in Fig. 6-8. Fig. 6. Image with blocking artefacts - img138 Fig. 7. Good picture - img168 Fig. 8. Good picture - lena In Fig. 6, there are obvious blocking artifacts and the rest are good pictures. As a result of model experiments, it was observed that «ideally-good» images have a gating effect metric equal to zero. With increasing strobe effects, the metric also increases. Calculation of strobe effect metrics was also carried out taking into account horizontal and vertical strobe effects, both for «gray» images, and for the corresponding three color channels. Averaging of the results of estimating the horizontal and vertical boundaries of the selected contours was also carried out using the methods of Canny and Perwitt. In accordance with the proposed technique for determining the metrics of the art-facts of the effect strobe, the images presented in Fig. 1-3, Fig. 6-8 were considered. Model experiments were performed in the MATLAB system with the Image Processing Toolbox. The results of the experiments are listed in Table 1. Table 1 Numerical metric estimates of test images Gray scale image, RGB image Images and metrics of strobe effects Metrics Strob1 Strob2 Strob3 img138 img168 lena From gray image Horizontal 27 16 9 1 0 1 Vertical 0 1 57 1 1 2 Average 13.500 0 8.500 0 33.000 0 1.000 0 0.500 0 1.500 0 From RGB image Red Horizontal 13 16 9 1 0 1 Red Vertical 0 1 57 1 1 2 Red Average 6.500 0 8.500 0 33.000 0 1.000 0 0.500 0 1.500 0 Green Horizontal 29 13 9 1 0 1 Green Vertical 3 1 74 1 1 4 Green Average 16.000 0 7.000 0 41.500 0 1.000 0 0.5 2.500 0 Blue Horizontal 29 44 9 1 0 1 Blue Vertical 104 0 75 1 1 2 Blue Average 66.500 0 22.000 0 42.000 0 1.000 0 0.5 1.500 0 Horizontal average 23.666 7 24.333 3 9.000 0 1.000 0 0.000 0 1.000 0 Vertical average 35.666 7 0.666 7 68.666 7 1.000 0 1.000 0 2.666 7 Average of horizontal and vertical 29.666 7 12.500 0 38.833 3 1.000 0 0,500 0 1.833 3 As can be seen from Table 1, their metrics are noticeably different for images obvious artifacts strobe effect upward from the respective images of metrics, which seemed to contain no strobe effects, at least visually. Developing the approach of calculating the coordinates of the extreme boundaries, you can build circles with radii obtained from the extreme coordinate values of the boundaries of the primary artifact and image with strobe effects. Further you can define the areas of the corresponding circles that were obtained when the contours were selected by Canny and Prewitt. By making the ratio of smaller to larger area it will be possible to notice what this ratio is for good images and images with strobe effects. This relationship can also serve as a quantitative measure for evaluating strobe effect artifacts. Also, there are options for calculating the radii of circles, depending on the location of strobe effects on the image. Accordingly, the ratio of the areas of these circles and the numerical metric of determining the strobe effects will change. Software can calculate relatively the largest metric assess gate effects on bitmap images. Figures 9 and 10 show the construction of circles whose areas are involved in the calculation of the strobe effects metrics. Fig. 9. Drawing circles for the Strob3 image Fig. 10. Drawing circles for the lena image As you can see from Fig. 9, 10, images with strobe effects have different circles in comparison with conditionally clean images. For ideal images, the corresponding circles should practically merge (Fig. 10). Table 2 shows the calculated metrics of the relative areas of circles for all the images studied. Table 2 Metrics in the form of the ratio of the areas of circles Metrics Strob1 Strob2 Strob3 Img138 Img168 lena Sminmax 0.799 02 0.819 61 0.668 73 0.991 67 1.000 00 0.996 07 Still, the «clean» images (img168, lena) and the image with blocking artifacts (img138) differ markedly in the introduced metric, which is intended for evaluating strobe effects on images (Strob1, Strob2, Strob3). The proposed metric in Table 2 is denoted as Sminmax, meaning the ratio of the smaller area of the circle to the larger. Conclusion The approaches considered in the author's opinion allow us to identify strobe effects in images. It may also be necessary to determine the scale of belonging to good images and images with strobe effects. As follows from the description of the proposed metrics, they should not cause difficulties for their programming. In this regard, the author dares to propose a developed technique for determining the numerical metrics of the quality of raster images for the presence or absence of artifacts such as strobe effects. The author of the article calculations and constructions performed in the MATLAB system with the Image Processing Toolbox. The author will be grateful to everyone who will respond to this article with possible comments and suggestions. Considering the great attention to digital image processing, as well as the detection of various artifacts on them, the author hopes that the results of this article will find practical application in real production.
Список литературы

1. Starovoitov V. V., Starovoitov F. V. Sravnitel'nyi analiz bezetalonnykh mer otsenki kachestva tsifrovykh izobrazhenii [Comparative analysis of artifact-free measures of evaluating quality of digital images]. Sistemnyi analiz i prikladnaia informatika, 2017, no. 1 (13), pp. 24-32.

2. Gu K., Zhou J., Zhai G., Lin W., Bovik A. C. No-reference quality assessment of screen content pictures. IEEE Transactions on Image Processing, 2017, vol. 26, no. 8, pp. 4005-4017.

3. Zhang Lin, Zhang Lei, Bovik A. C. A feature-enriched completely blind local image quality analyzer. IEEE Transactions on Image Processing, 2015, vol. 24, no. 8, pp. 2579-2591.

4. Xue W., Zhang L., Mou X., Bovik A. C. Gradient magnitude similarity deviation: A highly efficient perceptual image quality index. IEEE Transactions on Image Processing, 2014, vol. 23, no. 2, pp. 684-695.

5. Al'-Askari M. A., Fedosin S. A., Afonin V. V. Analiz kachestva rastrovykh izobrazhenii [Analysis of quality of raster images]. Nauchno-tekhnicheskii vestnik Povolzh'ia, 2018, no. 1, pp. 107-109.

6. Al'-Askari M. A., Fedosin S. A., Afonin V. V. Veroiatnostnaia otsenka kachestva rastrovykh izobrazhenii [Probabilistic assessment of quality of raster images]. Nauchno-tekhnicheskii vestnik Povolzh'ia, 2018, no. 4, pp. 62-65.

7. Al-Askari M. A. The definition of blocking artifacts in raster images. Aktual'nye problemy i dostizheniia v estestvennykh i matematicheskikh naukakh: sbornik nauchnykh trudov po itogam Mezhdunarodnoi nauchno-prakticheskoi konferentsii № 5. Samara, NN: ITsRON Publ., 2018. Pp. 49-53.

8. Al-Askari M. A. The definition of artifacts of the type of blocking on raster images. Informatsionnye tekhnologii. Problemy i resheniia. Ufa, Izd-vo UGNTU, 2018. Vol. 1 (5). Pp. 83-87.

9. Egiazarian K., Astola J., Ponomarenko N., Lukin V., Battisti F., Carli M. New full-reference quality metrics based on HVS. Proceedings of the Second International Workshop on Video Processing and Quality Metrics. USA, Scottsdale, 2006. 4 p.

10. Dronnikova S. A., Gurov I. P. Uluchshenie kachestva izobrazhenii pri obrabotke videokadrov s razlichnym vremenem ekspozitsii [Improvement of image quality during video frame processing with different time of exposure]. Nauchno-tekhnicheskii vestnik informatsionnykh tekhnologii, mekhaniki i optiki, 2017, vol. 17, no. 3, pp. 424-430. DOI:https://doi.org/10.17586/2226-1494-2017-17-3-424-430.

11. Ponomarev S. V. Issledovanie metodov detektirovaniia granits poverkhnostei v zadache sovmeshcheniia izobrazhenii trekhmernykh stsen [Study of methods of detecting surface boundaries in the problem of overlapping images of 3D scenes]. Nauchno-tekhnicheskii vestnik informatsionnykh tekhnologii, mekhaniki i optiki, 2017, vol. 17, no. 4, pp. 694-701. DOI:https://doi.org/10.17586/2226-1494-2017-17-4-694-701.

12. Varlamova A. A., Kuznetsov A. V. Obnaruzhenie vstraivanii na izobrazheniiakh putem analiza artefaktov, obuslovlennykh parametrami sensora registriruiushchego ustroistva [Detecting imbeddings on images by analyzing artifacts stipulated by parameters of sensor of recording unit]. Komp'iuternaia optika, 2017, vol. 41, no. 6, pp. 920-930.

13. Gonsales R., Vuds R., Eddins S. Tsifrovaia obrabotka izobrazhenii v srede MATLAB [Digital processing of images in MATLAB environment]. Moscow, Tekhnosfera Publ., 2006. 616 p.

14. Gonsales R., Vuds R. Tsifrovaia obrabotka izobrazhenii [Digital processing of images]. Moscow, Tekhnosfera Publ., 2012. 1104 p.

15. Umaa Mageswari S., Sridevi M., Mala C. An experimental study and analysis of different image segmentation techniques. International Conference on Design and Manufacturing (IConDM2013) (Chennai, India, 18-20 July 2013). Procedia Engineering, vol. 64, part 1, pp. 46-55.


Войти или Создать
* Забыли пароль?