JP6078431B2 - Video quality estimation apparatus, video quality estimation method and program - Google Patents

Description

  The present invention relates to a video quality estimation apparatus, a video quality estimation method, and a program, and more particularly to a video quality estimation technique for efficiently performing objective evaluation of video quality in a video distribution service performed via a network.

  With the increase in the speed and bandwidth of broadcast waves and Internet access lines, video communication services for transferring video media between terminals or between a server and terminals via broadcast waves and the Internet have become widespread.

  In broadcasting, CATV, and the Internet, since the original video is encoded and transmitted, the deterioration due to the encoding occurs. For this reason, the quality (user experience quality: QoE (Quality of Experience)) perceived by the user with respect to video media and the like deteriorates.

  Specifically, when the original video is encoded, the video signal in the frame is deteriorated by processing in units of blocks, or high-frequency components of the video signal are lost, so that the fineness of the entire video is lowered.

  As a result, the user perceives blur, blur, and mosaic distortion in the received video.

  In order to confirm that the video communication service having the above characteristics is provided to the user with good quality, the quality of the video experienced by the user is measured before or during the provision of the video communication service, It is necessary to monitor the quality of the video provided to the user.

  Therefore, there is a need for an objective video quality evaluation technique that can appropriately evaluate the quality of video experienced by the user.

  On the other hand, various subjective quality evaluation methods and objective quality evaluation methods have been proposed, and objective quality evaluation methods include techniques for deriving video quality evaluation values from feature values extracted from video pixel signals ( For example, see Non-Patent Document 1-3).

ITU-T recommendation J.341, January 2011 ITU-T recommendation J.247, August 2008 ITU-T Recommendation J.144, March 2004 ITU-T recommendation P.910, April 2008

  In the prior art disclosed in Non-Patent Documents 1-3, feature amounts are derived in units of video frames (pictures), and values obtained by performing statistical processing (such as averaging processing) on the basis of them are obtained for a certain period of time (for example, For example, 10 seconds).

  However, there is a problem that as the video size increases, the calculation amount necessary for calculating the feature amount increases. For example, in Non-Patent Documents 2 and 3, feature amounts are calculated with the original video size. If these feature amounts are calculated for large video sizes such as 4K (3840x2160) and 8K (7680x4320), the amount of computation is enormous. Become. As described above, the conventional technique calculates the feature amount with the original video size, and there is a problem that the calculation amount becomes enormous as the video size increases.

  In Non-Patent Document 1, the HD size (1920x1080 or 1280x720) is converted to 960x540, and the feature amount is calculated. When the feature size is calculated by reducing the video size as in Non-Patent Document 1, if the 4K or 8K video is made extremely small like 960x540, the feature amount calculated with the original size and the size of 960x540 The calculated feature quantity does not necessarily have a high correlation, and as a result, there is a problem that the quality estimation accuracy decreases.

  In order to explain the above-mentioned problem in detail, FIGS. 1 and 2 show examples of feature amounts when the image size is downsized and the feature amount is calculated and when the feature size is calculated with respect to the original size. FIG. 1 shows an SI (vertical and horizontal) defined in ITU-T recommendation P.910 (see Non-Patent Document 4), which is a video (degraded video) obtained by encoding a 10-second 3840x2160 / 60 fps original video at a certain bit rate. The 3x3 Sobel filter is applied to the original video in the direction, the square root of the square sum of the vertical and horizontal edge amounts is calculated for each pixel, and the standard deviation of the edge amount is calculated for each video frame. The average value (SI_i_Lx) for (600) is shown. FIG. 2 shows the result of calculating TI (calculating the standard deviation of the difference image of two adjacent video frames) for each video frame and calculating the average value (TI_i_Lx) for the number of differential video frames (599 frames). ing. Here, in the definition of ITU-T recommendation P.910, SI and TI are defined as having the maximum value in the measurement interval with respect to the SI value and TI value calculated for each video frame. Here, SI and TI of all video frames are averaged by the number of video frames. I represents R (original video) and D (degraded video), L1 is the original video size (3840x2160), L2 is 1920x1080, L3 is 960x540, L4 is 480x270, and L5 is 240x135. As can be seen from these figures, SI has a good correspondence between the feature amounts calculated from the resolutions of L1 and L2 (that is, a curve can be approximated), but has a bad correspondence with other resolutions. TI has a good correspondence with the feature amount calculated from the resolutions of L1, L2, and L3, but it has a bad correspondence with other resolutions. In other words, it can be said that the resolution level that can be reduced differs depending on the feature amount.

  When the feature amount calculated from the reduced resolution and the feature amount calculated from the original size can be approximated with high correlation, the feature amount is calculated from the original resolution as in Non-Patent Documents 2 and 3. It turns out that it is inefficient from the viewpoint of quantity reduction. Further, as in Non-Patent Document 1, it may be inefficient from the viewpoint of the amount of calculation to calculate a feature amount with a fixed size regardless of the feature amount. Furthermore, when downsampling from a high-resolution video to a specific resolution, the feature quantity calculated from the downsampled video does not correspond to the feature quantity calculated from the original size, and the quality estimation accuracy is reduced.

  The above feature amount is explained with an example of the original video size of 3840x2160, but if the original video size changes, how much downsampling can be changed. For example, when a certain feature amount is calculated from the video size of 7680x4320, three stages of downsampling can be performed, but when a certain feature amount is calculated from the original video size of 640x480 video size, only two steps of downsampling can be performed. The size of downsampling varies depending on the original video size. For this reason, when deriving the feature amount, it is important from the viewpoint of quality estimation accuracy to derive the feature amount by changing the thinning number of pixels in consideration of the original video size.

  The present invention has been made in view of the above points, and derives a video quality evaluation value by using a feature amount calculated from a downsampled video according to a feature amount type used for objective video quality evaluation value estimation. The object is to provide video quality estimation technology. Specifically, the amount of calculation can be reduced by downsampling the original video size to an appropriate resolution according to the feature quantity to be calculated, calculating the feature quantity, and deriving the video quality evaluation value. A video quality estimation apparatus, method, and program are provided.

An image quality estimation apparatus according to an aspect of the present invention is
A video quality estimation device for evaluating video quality,
Based on the correlation between the feature value derived from the pixel signal at the original video size of the input pixel signal and the feature value derived from the down-sampled pixel signal, the number of pixels to be down-sampled is determined and input. A down-sampling unit for down-sampling the processed pixel signal;
A video quality estimation unit that derives a video quality evaluation value using a feature amount derived from the downsampled pixel signal;
It is characterized by having.

In addition, a video quality estimation method according to an aspect of the present invention includes:
A video quality estimation method in a video quality estimation apparatus for evaluating video quality,
The downsampling unit of the video quality estimation device is based on a correlation between a feature quantity derived from a pixel signal at an original video size of an input pixel signal and a feature quantity derived from the downsampled pixel signal. Determining the number of pixels to be down-sampled and down-sampling the input pixel signal;
A step in which a video quality estimation unit of the video quality estimation device derives a video quality evaluation value using a feature amount derived from a downsampled pixel signal;
It is characterized by having.

A program according to an aspect of the present invention is
A program for causing a computer to function as a video quality estimation device for evaluating video quality, comprising:
Based on the correlation between the feature value derived from the pixel signal at the original video size of the input pixel signal and the feature value derived from the down-sampled pixel signal, the number of pixels to be down-sampled is determined and input. Down-sampling means for down-sampling the pixel signal obtained, and video quality estimation means for deriving a video quality evaluation value using the feature quantity derived from the down-sampled pixel signal,
It is made to function as.

  According to the present invention, it is possible to derive a video quality evaluation value using a feature amount calculated from a downsampled video according to a feature amount type used for objective video quality evaluation value estimation. Since the video quality evaluation value is estimated using a video obtained by down-sampling the original video size to an appropriate resolution, it is possible to reduce the amount of calculation without reducing the accuracy of the quality evaluation.

The figure which shows the relationship between the feature quantity (SI) calculated from the video frame in the original video size and the feature quantity (SI) calculated from the down-sampled video frame The figure which shows the relationship between the feature-value (TI) calculated from the video frame in the original video size, and the feature-value (TI) calculated from the down-sampled video frame The block diagram of the video quality estimation apparatus which concerns on 1st Example of this invention. The flowchart of the video quality estimation method according to the first embodiment of the present invention. The figure which shows the relationship between the feature quantity (PSNR) calculated from the video frame in the original video size and the feature quantity (PSNR) calculated from the down-sampled video frame Configuration diagram of video quality estimation apparatus according to the second embodiment of the present invention The flowchart of the video quality estimation method according to the second embodiment of the present invention. Flowchart of video quality estimation method according to the third embodiment of the present invention.

  Examples of the present invention will be described in detail below.

<First embodiment>
FIG. 3 is a configuration diagram of the video quality estimation apparatus 100 according to the first embodiment of the present invention. The video quality estimation apparatus 100 is an apparatus that objectively evaluates video quality. As will be described below, the video quality estimation apparatus 100 estimates the video quality evaluation value using a video obtained by down-sampling the original video size to an appropriate resolution, thereby calculating the amount of calculation when evaluating the video quality. Can be reduced.

  The video quality estimation apparatus 100 includes a downsampling unit 101 and a video quality estimation unit 102.

  The downsampling unit 101 downsamples the input signal (pixel signal). For example, the downsampling unit 101 may downsample the original video size in half by thinning out even-numbered or odd-numbered pixels in the horizontal and vertical directions.

The video quality estimation unit 102 estimates a video quality evaluation value in a desired measurement section from the downsampled pixel signal. The video quality estimation unit 102 includes a video frame quality estimation unit 103 and a video quality evaluation unit 104. The video frame quality estimation unit 103 derives a feature quantity F _ij (i: feature quantity number, j: video frame number) for each video frame. Next, the video quality evaluation unit 104 derives a video quality evaluation value V by weighting the derived feature amount _Fij .
A specific calculation processing example is as follows.

ただし、a_j、b_iは重み係数、Jは測定区間の映像フレーム数、Iは特徴量の数である。

Here, a _j and b _i are weighting factors, J is the number of video frames in the measurement section, and I is the number of feature quantities.

  Next, the operation of the video quality estimation apparatus 100 according to the first example will be described. FIG. 4 is a flowchart of the video quality estimation method according to the first embodiment of the present invention.

  The downsampling unit 101 downsamples pixel information of the original video size. For example, the resolution of 3840x2160 is thinned out in the horizontal and vertical directions, and even-numbered pixels are thinned out, and converted into a video having a resolution of 1920x1080 (S101). In this embodiment, even-numbered pixels are thinned out. However, odd-numbered pixels may be thinned out, and instead of down-sampling in half in the horizontal and vertical directions, one pixel at a time for 3 pixels, 4 pixels, etc. You may leave and downsample.

  The video frame quality estimation unit 103 derives a feature amount for each video frame from the downsampled video (S102). Examples of feature amounts and calculation methods will be described in detail below.

  The video quality evaluation unit 104 weights the feature amount calculated for each video frame, derives a video quality evaluation value V, and ends the process. (S103). In the present embodiment, an example is shown in which the feature amount is weighted and the video quality evaluation value is derived. However, the scope of the present invention is not limited to this derivation method.

Derivation of the feature amount in the video frame quality estimation unit 103 will be described. Here, PSNR (Peak-Signal-to-Noise Ratio) will be described as an example of the feature amount for each video frame. Similar to FIGS. 1 and 2, FIG. 5 shows a correlation diagram of PSNR calculated from the resolution of L2-L5 with respect to PSNR calculated for the resolution of original L1. As shown in FIG. 5, the PSNR calculated from the resolutions of L1 and L2 can be approximated, and may be calculated by down-sampling in half in the horizontal and vertical directions (that is, to L2 level video). Recognize. Thus, the downsampling unit 101 downsamples the resolution in half in the horizontal and vertical directions, the video frame quality estimation unit 103 calculates the PSNR for each video frame from the resolution of L2 (1920x1080), and the video quality evaluation unit 104 A quality evaluation value is calculated. As a result, the calculation amount can be reduced without degrading the quality estimation accuracy.
A specific calculation processing example is as follows.

ただし、PSNR_L2_jはjフレーム番目の映像フレームに対し，L2（1920x1080）の解像度から算出したPSNR、a_jは重み係数、Jは測定区間の映像フレーム数である。

Here, PSNR_L2 _j is the PSNR calculated from the resolution of L2 (1920 × 1080) for the j-th video frame, a _j is a weighting factor, and J is the number of video frames in the measurement section.

  Next, an example of the feature amount using the SI shown in FIG. 1 is shown. From the result of FIG. 1, the video size is down-sampled to the L2 level, and the feature amount is calculated by the following formula.

ただし、SI_R_L2_jとSI_D_L2_jは原映像と劣化映像のjフレーム番目の映像フレームに対し、L2（1920x1080）の解像度から算出したSI_R_L2とSI_D_L2、a_jは重み係数、Jは測定区間の映像フレーム数である。

However, SI_R_L2 _j and SI_D_L2 _j are SI_R_L2 and SI_D_L2 calculated from the resolution of L2 (1920x1080) for the j-th video frame of the original video and degraded video, a _j is a weighting factor, J is the number of video frames in the measurement section It is.

  In this example, feature values based on PSNR and SI are used as examples, and feature values are calculated from downsampled video, and video quality evaluation values are derived.However, other feature values are derived from downsampled video. The method for deriving the video quality evaluation value is also within the scope of the present invention.

<Second embodiment>
In the second embodiment of the present invention, a configuration in which a low-pass filter unit 110 is added before the downsampling unit 101 will be described.

  FIG. 6 is a configuration diagram of the video quality estimation apparatus 100 according to the second embodiment of the present invention. As shown in FIG. 6, a low-pass filter unit 110 is added before the downsampling unit 101. Other configurations are the same as those of the video quality estimation apparatus 100 according to the first embodiment shown in FIG.

  The low-pass filter unit 110 eliminates discontinuities in pixel values and smoothes the video. In general, the original image captured by the camera has a minute noise. Therefore, it is possible to remove these noises by performing filtering in the low-pass filter unit 110.

  Next, the operation of the video quality estimation apparatus 100 according to the second embodiment will be described. FIG. 7 is a flowchart of the video quality estimation method according to the second embodiment of the present invention. In the video quality estimation apparatus 100 according to the present embodiment, the video is not simply downsampled, but is downsampled after the video is smoothed by a filter (S100).

  The processing after downsampling is the same as in the first embodiment. As described above, when pixels are simply thinned out by the down-sampling unit, the discontinuity of the pixel value becomes remarkable. Therefore, the pixel discontinuity can be eliminated by filtering the pixel value by the filter unit 110.

  For example, the low-pass filter unit 110 uses an average value of 4 pixels in horizontal and vertical 2 × 2 as a pixel value. As another filtering process in the low-pass filter unit 110, for example, a Gaussian filter or another filtering process is also included in the scope of the present invention.

<Third embodiment>
In the third embodiment of the present invention, a configuration in which the downsampling unit 101 determines the number of pixels to be downsampled according to the original video size will be described.

  The video quality estimation apparatus 100 according to the third embodiment of the present invention is the same as the video quality estimation apparatus 100 according to the first embodiment shown in FIG. 3 or the video quality estimation apparatus 100 according to the second embodiment shown in FIG. is there.

  However, the downsampling unit 101 of this embodiment determines the number of pixels to be downsampled based on the original video size of the input pixel signal. The number of pixels to be down-sampled differs depending on the type of feature quantity used for objective video quality evaluation estimation. The feature quantity derived from the pixel signal at the original video size and the characteristic derived from the down-sampled pixel signal It may be determined based on the correlation with the quantity. For example, as shown in FIG. 1, since the feature quantity based on SI corresponds to the feature quantity calculated from the resolutions of L1 and L2, the feature quantity may be thinned out to the video size of L2. Further, as shown in FIG. 2, since the feature quantity based on TI has a good correspondence with the feature quantity calculated from the resolutions of L1, L2, and L3, the feature quantity may be thinned out to the video size of L3.

  Next, the operation of the video quality estimation apparatus 100 according to the third embodiment will be described. FIG. 8 is a flowchart of the video quality estimation method according to the third embodiment of the present invention. The original video size is input to the downsampling unit 101, and the number of pixels to be downsampled is determined before derivation of each feature amount based on the original video size (S104).

  The processing after the determination of the number of pixels to be downsampled is the same as in the first embodiment or the second embodiment, and then the downsampling unit downsamples based on the number of pixels to be downsampled (S101). For example, in the first embodiment, the number of pixels to be down-sampled is fixed, the video size of 3840x2160 is L1, and the feature quantity based on PSNR and SI is thinned out in half with respect to horizontal and vertical, and the video size of L2 ( The feature value was calculated from 1920x1080). For example, in the third embodiment, when the video size of the input video is 7680x4320, the video size of 7680x4320 is L1, and the feature quantity based on PSNR and SI is the feature quantity from the video size of L2 (3840x2160). A method of thinning out the number of pixels, such as calculation, is determined for each feature amount with respect to the input signal, and downsampling is performed.

<Effect of the embodiment of the present invention>
As described above, according to the embodiment of the present invention, the input video is down-sampled, the feature amount of the video frame unit is derived, the feature amount is weighted, and the video quality evaluation value of the plurality of video frames is derived. Therefore, in addition to being able to appropriately evaluate the video quality evaluation value, it is possible to contribute to a reduction in the amount of calculation when deriving the feature amount.

  Therefore, it is possible to easily determine whether the service being provided maintains a certain level of quality for the user, and it becomes possible to grasp and manage the actual quality of the service being provided in real time. . In addition, by reducing the amount of computation, the apparatus can be slimmed down and the apparatus cost can be reduced.

  Note that the video quality estimation apparatus 100 in the above embodiment is realized by installing a computer program in a computer including a CPU (Central Processing Unit), a memory, and an interface, and the various functions of the video quality estimation apparatus 100 described above are as follows. The various hardware resources of the computer and the computer program (software) are realized in cooperation.

  The operation of the constituent elements of the video quality estimation apparatus 100 shown in FIG. 3 or FIG. 6 is constructed as a program and installed in a computer used as the video quality evaluation apparatus for execution or distributed via a network. Is possible.

  Furthermore, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and installed in a computer or distributed.

  The present invention is not limited to the above-described embodiments, and various modifications and applications can be made within the scope of the claims.

  It can be used for a video quality estimation device that estimates a video quality evaluation value of video communication such as an IPTV service and a video distribution service performed via an IP network.

DESCRIPTION OF SYMBOLS 100 Video quality estimation apparatus 101 Downsampling part 102 Video quality estimation part 103 Video frame quality estimation part 104 Video quality evaluation part 110 Low pass filter part

Claims (5)

A video quality estimation device for evaluating video quality,
Based on the correlation between the feature value derived from the pixel signal at the original video size of the input pixel signal and the feature value derived from the down-sampled pixel signal, the number of pixels to be down-sampled is determined and input. A down-sampling unit for down-sampling the processed pixel signal;
A video quality estimation unit that derives a video quality evaluation value using a feature amount derived from the downsampled pixel signal;
A video quality estimation apparatus.   The video quality estimation apparatus according to claim 1, wherein the video quality estimation unit includes a video frame quality estimation unit that derives a feature amount in units of video frames.   The video quality estimation apparatus according to claim 1, further comprising a low-pass filter unit that filters the input pixel signal. A video quality estimation method in a video quality estimation apparatus for evaluating video quality,
The downsampling unit of the video quality estimation device is based on a correlation between a feature quantity derived from a pixel signal at an original video size of an input pixel signal and a feature quantity derived from the downsampled pixel signal. Determining the number of pixels to be down-sampled and down-sampling the input pixel signal;
A step in which a video quality estimation unit of the video quality estimation device derives a video quality evaluation value using a feature amount derived from a downsampled pixel signal;
A video quality estimation method comprising: A program for causing a computer to function as a video quality estimation device for evaluating video quality, comprising:
Based on the correlation between the feature value derived from the pixel signal at the original video size of the input pixel signal and the feature value derived from the down-sampled pixel signal, the number of pixels to be down-sampled is determined and input. Down-sampling means for down-sampling the pixel signal obtained, and video quality estimation means for deriving a video quality evaluation value using the feature quantity derived from the down-sampled pixel signal,
Program to function as.

Images