CC BY
6APRIL 27-28, 2023
RESEARCH OF THE EFFICIENCY OF IMAGE RESOLUTION CHANGE METHODS
FOR THEIR BIDIRECTIONAL RESIZING.
1Akhmedova Anora Khalbaevna, 2Gavrilov Igor Aleksandrovich, 3Puzy Anastasia
Nikolaevna
1,2,3Tashkent University of Information Technologies https://doi.org/10.5281/zenodo.7859587
Abstract. The quality of television images after reducing their resolution by a factor of 2, and then restoring their initial resolution by various interpolation methods are studied in the article. A possibility of using the studied methods for bidirectional image scaling when compressing video data volumes in high-definition and ultra-high-definition television is assessed.
Keywords: Digital TV, digital image, image scaling, image resizing, interpolation, image quality.
INTRODUCTION
In modern broadcast television, high-definition formats are increasingly being introduced, and work to transmit ultra-high-definition TV images in 4K and 8K formats is in progress. However, TV transition to the formats greatly increases the amount of video data that needs to be encoded in real time. At the same time, the 4K format with a resolution of 3840*2160 contains about 8.3 megapixels, and the 8K format has 33 megapixels. They both require very high-speed and expensive encoders or the use of special methods that greatly minimize the original images video data. One of such interesting approaches is the method described in [1]. It is based on the idea of preliminary image size reduction before encoding and then restoring the original image size after decoding on the receiving side. For this purpose, it is proposed various interpolation methods for reducing and increasing of the images' size.
MAIN PART
Previous researches have shown that both simpler non-adaptive interpolation algorithms and more complex adaptive interpolators are used to resize digital images [2-5]. The difference between the methods is non-adaptive algorithms process all pixels in the same way, regardless of the image structure, while adaptive algorithms use pixel-by-pixel image analysis to detect edges and fine structure elements.
Such approach allows you to minimize distortion in reduced or enlarged images in those places where they are most noticeable. As a rule, these algorithms are primarily designed to save the maximum detail of the original local effects in increased images. In this case, the interpolation methods themselves in non-adaptive and adaptive interpolators can be the same.
Nowadays in practice such interpolation methods as bilinear, bicubic, splines, Lanczos, Mitchell and a number of others have received wide distribution. However, studies have shown [5] that a number of serious problems arise during bidirectional resizing of TV images, due to the fact that the restored images differ greatly in quality from the original ones.
First reason is the algorithms of reducing image sides by 2 (horizontally and vertically): transformation of original resolution images into images with a 4-fold reduced number of pixels demand to remove somehow the extra pixels. Usually this problem is solved by interpolation calculation of the values of the reconstructed pixel from the values of surrounding pixels. So,
APRIL 27-28, 2023
noticeable distortion or even loss of fine details may occur in the reduced image due to averaging or decimation the pixel values.
To assess the quality of reduced images we prepared a test black-and-white reference image of a segmented circle with a line thickness of 1 pixel (Fig. 1). It must be obtained with an ideal case of interpolrtion after double down of original image size. At the Fig. 1 you can see a table of pixel values of the upper part of the image for an objective quality assessment of the image after interpolation. The white color is represented by a maximum brightness value of 255, and black one by a zero brightness value.
Fig.1. The reference test image with its pixel values.
For researching a segmented circle twofold increased was reduced by 2 by various methods of non-adaptive and adaptive interpolation and compared with the standard (Fig. 1). Figures 2-5 show the results of a twofold reduction of images by non-adaptive Lanczos-3, Mitchell, bicubic and B-spline interpolation methods, and Figures 6-7 are the results of adaptive interpolation by the S-Spline and S-Spline Max algorithms, which provide the best results in most cases.
Fig.2. The result of a twofold reduced image by the Lanczos-3 method
Fig.3. The result of a twofold reduced image by the Mitchell method
APRIL 27-28, 2023
Fig.4. The result of a twofold reduced image by the Bicubic interpolation method
Fig.5. The result of a twofold reduced image by the B-spline interpolation method
Fig.6. The result of a twofold reduced image by the adaptive method of S-Spline
interpolation
APRIL 27-28, 2023
Fig.7. The result of a twofold reduced image by the adaptive method of S-Spline Max
interpolation
As can be seen from the presented Figures 2-7, none of the researched methods provides scaling without distortion of pixel values, especially black lines, which is replaced by gray ones. In addition, the Lanczos, bicubic and B-spline interpolation methods expand the width of the lines in the image. And the results of Mitchell and S-Spline interpolation methods are closest to a quality of the initial image, although they create gaps in the halon lines.
Thus, significant distortion of twofold reducing image creates a second bidirectional interpolation problem associated with the restoration of already distorted images, as shown in Figures 8-9.
Fig.8. The result of restoring the image size by the Mitchell method
Fig.9. The result of restoring the image size by the S-Spline Max. As you can see from the results of restoring of the original resolution from reduced images (Fig. 8, 9), all image lines have a double width, although in the reference test image the width was equal to one pixel. There are distortions of their brightness and clarity often
APRIL 27-28, 2023
noticeable. Also, important details missing in the reduced image cannot be restored by any interpolators.
CONCLUSIONS AND RECOMMENDATIONS
When researching the possibility of appllying bidirectional resizing to TV images for increasing the efficiency of encoding high and ultra-high definition images, it was found that a twofold reduction in image sizes using modern adaptive and non-adaptive scaling methods is accompanied by significant brightness and structural distortions of test images. In particular, an additional interpolation distortions of brightness and twofold size expansion of the structural elements of the images occur, when restoring the size of reduced images. It significantly reduces the clarity and contrast of the images obtained. Thus, images after their bidirectional resizing have a lower visual quality compared to the original one. Therefore, to implement the idea of bidirectional image resizing, it is necessary to develop more efficient methods for image size reducing, which will minimize their brightness and texture distortions. And methods restoring the image sizes should not double the structural elements in those places, where this is not supposed to be done. Then the idea of bidirectional image resizing can be used in high-definition and ultra-high definition broadcast television.
REFERENCES
1. Н. В. Соловьев, Г. В. Шифрис. Использование предварительного масштабирования для повышения качества видеопотока // Информационно-управляющие системы №4, 2011. С.2-8.
2. Гаврилов И.А., Пузий А.Н, Бабаян Р.И. Увеличение коэффициентов сжатия видеоданных на основе изменения размеров ТВ изображений. Научно -технический и информационно-аналитический журнал ВЕСТНИК ТУИТ. 2016. №1 (37). с. 46-54
3. Анисимова А.Г., Гаврилов И.А. Применение методов ресайза для повышения эффективности кодирования ТВ изображений. Статья в сборнике III Международной научно-практической конференции «Актуальные проблемы науки и образования в современном ВУЗе», Ч. 1. 2017 г. С. 246-250.
4. Gavrilov I., Puziy. A. ^e Investigation of Image Sizes Reduction Methods. European Science Review, № 11-12, 2017, Vienna, -P. 136-139.
5. Anastasia Puziy, Igor Gavrilov, Khabibullo Nosirov, Anora Akhmedova. Efficiency estimation of image resizing based on interpolating transformations. Internation conference on information science and communications technologies applications, trends and opportunities. ICISCT2019. 2019 (5/32).
