CC BY
23
Puziy Anastasiya Nikolaevna,
Gavrilov Igor Aleksandrovich, Tashkent University ojInnformation Technologies named after Muhammad Al-Khwarizmi, E-mail: puziy-2008@mail.ru
^E INVESTIGATION OF IMAGE SIZES REDUCTION METHODS
Abstract: This article answer the question about the most effective algorithm for pre-encoded resize video frames to reduce the amount of data transmitted through the communication channels. An experimental study, in which the main criterion was the quality of reconstructed video frames, is described. There are also possibilities of many times applying the most effective resize algorithm are discussed.
Keywords: TV image, compression, redundant information, Lanczos, B-spline filter, video stream, resize, compression ratios.
The need for enhancing in the quality and quantity of TV programs significant increases within development of digital television processes. The number of transmitting programs of high-definition (HD) standards with a much larger amount of data is growing. Therefore, in conditions of limited frequency resource, it is possible to preserve image quality only by creating more effective methods of compressing TV images with high compression ratios of the video stream.
Compression of the video stream is affected by a number of factors. First of all, it is the presence of redun-
dant information (code, structural, statistical, psycho-visual, temporal ones), the value of which depends on the structure of the video. So videos with a relatively homogeneous background contain a large amount of redundant information and are compressed better than videos with small details and significant brightness differences (Fig. 1). That's why poorly compressible video data is usually additionally compress by quantizer due to loss of some useful information, what significantly reduces the quality of images. Hence, we need more efficient coding methods to preserve the visual quality of TV images [1].
a) Compression in 25 times b) Compression in 3 times
Figure 1. The results of different structure images compression without loss of quality
One of the promising increasing data compres- pixels horizontally and vertically before encoding in the sion in TV approaches can be able a reducing the size of original image (Fig. 2), then the amount of its data will the original image before encoding and restore its size be reduced by 4 times and, accordingly, the amount of during decoding, i. e. if we reduce twice the number of encoded image data will decrease.
Figure 2. The original, reduced and restored image of the video stream
However, we can irretrievably lose the part of use- cant deterioration in the quality of the reduced images,
ful information after the reduction of images size. It and hence, of the recovered ones [2]. (Figure 3) shows
greatly affects the quality and intelligibility of the re- an original test image with a shallow structure and the
constructed images. Thus, the removal of individual result of twice reducing its size by simply thinning out
pixels of fine-structure image elements leads to a signifi- the pixels.
O I □ O I □
CO
□ □ c o
OID
O XX
□ TO
□ X o
OID
O X □
O X □
Figure 3. Test original and twice reduced by the thinning method images.
As we can see from the figure above, after thinning out pixels many thin lines of the test image are completely removed. So, in practice, more complex methods and scaling algorithms are used back better results.
Various interpolation methods can be used for this purpose: bilinear, bicubic ones, based on Lanc-soz, Gauss, B-spline filters in Mitchell and Catmull-Rom variants etc. They are different in the computational complexity, the speed of processing, the accuracy
of pixel recovery, the presence or absence of a metadata array etc. [3].
The aim of research was selecting the method that minimizes structural distortions into TV images after twice reduced with the best reconstructed image quality. The reconstruction was made by the bilinear, bicubic interpolation methods, B-splines, Mitchell filter and Lanczos filter of the third order (Lanczos-3). The special graphic and the real TV images were used.
Based on the results of the studies, the Lanczos and B-spline interpolation methods provide the smallest distortions of fine-structure TV image elements. Other considered scaling methods provide less image quality.
In (Fig. 4 and 5) we can see reconstructed images after reduction by the methods of B-spline and Lanczos of the third order.
-^- □ I o □ I o o x a o x □
□ X o □ X o O X □ o xn
□ n
O
o
□ X
□ X
O X □ O I n
OID
O X □
a) B-spl ine b) Lanczos-3
Figure 4. The quality of the reconstructed test image after twice reducing it with B-spline and Lanczos-3 methods
a) B-spl ine b) Lanczos-3
Figure 5. The quality of the reconstructed real image after twice reducing it with B-spline and Lanczos-3 methods
As can be seen from the results of the experiments, the scaling methods using B-splines and Lanczos-3 provide a sufficiently high image quality without noticeable structural distortions. B-spline method, based on the fourth-order piecewise cubic algorithms, practically does not change the image structure, but suppresses high frequencies, what leads to blurring of the image and loss of clarity. And the Lanczos method of the third order based on the window sine-filter, on the contrary, fairly well transmits the clarity of the im-
age, but introduces some distortions and can create the Gibbs's effect.
Then, an additional study was carried out to evaluate the quality of TV images after a twofold reduction and increasing their size by the Lanczos method.
To evaluate the effectiveness of bidirectional image resizing we used 3 test images of different genres with different content of fine-grained elements. In the experiment each test image was first consistently twice reduced by the Lanczos-3 algorithm in along the hori-
zontal, vertical and the field, and then also restored to its original size. In this case, the evaluation of the quality of the reconstructed images was carried out
both subjectively (Fig. 6) and by calculating the mean square error (MSE) of the pixel values of the original and reconstructed image (Fig. 7) [4].
Figure 6. The quality of the original and reconstructed TV image after bidirectional (the field) resize using the Lanczos-3 method
Figure 4. Histograms of introduced distortions when scaling test images
using the Lanczos-3 method
As we can see from the results of the experiments, the Lanczos method shows good efficiency. So even on fine-grained images the visual degradation of quality is practically unnoticeable, and the objective estimation
of distortions doesn't exceed 11%, which is quite good result. Thus, the application of the Lanczos method may allow increase the efficiency of video streams compression of TV programs.
References:
1. Gavrilov I. A., Rakhimov T. G., Puziy A. N., Nosirov Kh.Kh., Kadirov Sh. M. "Цифровое телевидение", «Тор Image Media» - Tashkent c., - 2016. - 380 p.
2. Porev V. N. "Компьютерная графика", St. Petersburg: BHV - Petersburg, - 2002. - 432 р.
3. Kubasov D., Vatolin D. "Review of motion compensation methods", URL: http://cgm.graphicon.ru/con-tent/view/76/65/.
4. Richardson Ian "Video coding. MPEG-4 - next generation standard", Moscow, Vol. «Technosphere», - 2005.
5. Gavrilov I. A. Otto S. E., Kim M. V. "Improving the efficiency of the compression of the video stream based on the zoom image", Proceedings of the international conference "Actual problems of development of info-communications and the Information Society». - Tashkent. - 2012. - P. 157-162.
