JP5350300B2 - Transcode video quality objective evaluation apparatus, method and program - Google Patents

Description

The present invention relates to a transcoded video quality objective evaluation apparatus, method and program, and in particular,
The present invention relates to a transcoded video quality objective evaluation apparatus, method and program for objectively evaluating the quality of transcoded video in an IPTV service and video distribution service performed via an IP (Internet Protocol) network such as the Internet.

  With the increase in speed and bandwidth of Internet access lines, the spread of video communication services that transfer video media including video and audio between terminals or between servers and terminals via the Internet is expected.

  Since the Internet is a network whose communication quality is not always guaranteed, when communication is performed using audio and video media, the bit rate is lowered and the line is congested due to the narrow line bandwidth of the network between user terminals. As a result, packet loss and packet transfer delay occur, and the quality perceived by the user for audio, video media, etc. (user experience quality: QoE (Quality of Experience)) deteriorates.

  Specifically, when the original video is encoded or when the already encoded video is re-encoded (transcoded), the video signal in the frame may be deteriorated by processing in block units, or the high-frequency of the video signal The loss of components reduces the fineness of the entire video.

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

  In order to confirm that the video communication service as described above is provided with good quality, the quality of the video experienced by the user is measured before the service is provided or is provided to the user. It is important to monitor the high quality of the video.

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

  Conventionally, as a method for evaluating video quality, there are a subjective quality evaluation method (for example, see Non-Patent Document 1) and an objective quality evaluation method (for example, Non-Patent Document 2).

  The subjective quality evaluation method is a quality scale (very good, good, normal, bad, very good) of the quality experienced by multiple users who actually watched the video and experienced the quality (may be 9 or 11 levels) Bad) and disturbance scale (no degradation is observed, degradation is observed but not bothered, slightly concerned about degradation, concerned about degradation, very concerned about degradation), etc. Average the video quality evaluation value for each condition (for example, packet loss rate 0% and bit rate 20Mbps) by the number of users, and define the value as MOS (Mean Opinion Score) value or DMOS (Degradation Mean Opinion Score) value ing.

  However, subjective quality evaluation not only requires special dedicated equipment (such as a monitor) and an evaluation facility that can adjust the evaluation environment (such as room illuminance and room noise), but many users actually evaluate the video. There is a need. For this reason, it takes time for the user to actually complete the evaluation, which is not suitable for evaluating the quality in real time.

  Therefore, it is desired to develop an objective quality evaluation method that outputs a video quality evaluation value using a feature amount (for example, bit rate, bit amount in frame units, packet loss information, etc.) that affects video quality.

  One of the conventional objective quality estimation methods is that the original video before encoding and the degraded video after encoding are input, and the video signals (that is, pixel values) of both are compared, and the feature quantity that affects the video quality There is a technique for deriving a video quality evaluation value from the above (for example, see Non-Patent Document 2).

  In addition, in one of the conventional objective quality evaluation methods, the original video before encoding is not used, but the deteriorated video after encoding is input, and the feature quantity that affects the video quality is derived from this deteriorated video signal, There is a technique for deriving a video quality evaluation value (see, for example, Non-Patent Document 3).

  Further, as one of the conventional objective quality evaluation methods, there is a technique that receives transmitted packets as input, derives a feature quantity that affects video quality from these packets, and derives a video quality evaluation value (for example, (See Patent Documents 4 and 5).

  Many conventional objective quality evaluation methods estimate a video quality evaluation value using a packet or a video signal (pixel value) as described above.

ITU-T recommendation P.910 ITU-T recommendation J.247 J. Yang, H. Choi, and T. Kim, "Noise estimation for blocking artifacts reduction in DCT coded images," IEEE Trans. On CSVT, vol. 10, no. 7, pp. 1116-1134, Oct. 2000. K. Yamagishi and T. Hayashi, "Non-intrusive Packet-layer Model for Monitoring Video Quality of IPTV Services," IEICE Trans. Fundamentals, vol. E92-A, no. 12, pp. 3297--3306, Dec. 2009 . K. Watanabe, K. Yamagishi, J. Okamoto, and A. Takahashi, "Proposal of new QoE assessment approach for quality management of IPTV services," IEEE ICIP 2008, pp. 2060-2063, Oct. 2008. Stephane Pechard, Dominique Barba and Patrick Le Callet: Video Quality Model based on a spatio-temporal features extraction for H.264-coded HDTV sequences, in Proceedings of the IEEE Picture Coding Symposium, PCS2007, 2007.

  However, since the technique of Non-Patent Document 2 is based on the premise that the original video signal is used, only the encoded video and the re-encoded (transcoded) video are used to transcode the video after transcoding. When evaluating quality, there is a problem that the quality evaluation accuracy is remarkably lowered.

  Specifically, since the video before transcoding encodes the original video, it includes noise such as block noise, blur noise, mosquito noise, and flicker noise. For this reason, when the original image can be used, there is a problem that the extracted edge cannot be extracted, and the quality evaluation accuracy is significantly lowered.

  Non-Patent Documents 3, 4, and 5 are techniques for evaluating video quality using received video and received packets, but using only received video and received packets, encoded video before re-encoding. Therefore, there has been a problem that it is impossible to catch the deterioration in quality of the encoded video before re-encoding.

  The present invention has been made in view of the above points, and in order to solve the above-described problem, a transcoded video is obtained by deriving a video quality evaluation value from a feature amount derived from encoded video before and after transcoding. On the other hand, an object of the present invention is to provide a transcoded video quality objective evaluation apparatus, method and program capable of estimating a video quality value with high accuracy.

  FIG. 1 is a principle configuration diagram of the present invention.

The present invention (Claim 1) is a transcoded video quality objective evaluation apparatus for objectively evaluating the quality of transcoded video,
Edge / texture amount extraction means 100 for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistical amount derived from the edge amount and the texture amount derived from the edge / texture amount extraction means , and the average value of all the video frames with respect to the ratio of the edge amount and the texture amount per video frame, a certain interval A first feature amount that derives at least one first feature amount of a feature amount that is a maximum value and a standard deviation value of a set of average values for each time or a motion amount that is a feature amount derived from an encoded video before transcoding; Feature quantity extraction means 200;
Second feature quantity extraction means 400 for deriving a second feature quantity from difference information using the number of pixels in the horizontal and vertical directions of the input encoded video before transcoding and transcoded video after transcoding; ,
First video quality means for deriving, as a first video quality evaluation value, a video quality evaluation value of the input encoded video before transcoding from the first feature quantity output from the first feature quantity extraction means 300,
A second video quality evaluation value indicating a quality evaluation value of the transcoded video after transcoding is derived from the first video quality evaluation value and the second feature quantity derived from the second feature quantity extraction unit 400. 2 video quality estimation means 500.

The present invention (Claim 2) is a transcoded video quality objective evaluation device that objectively evaluates the quality of transcoded video,
Edge / texture amount extraction means for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistical amount derived from the edge amount and the texture amount derived from the edge / texture amount extraction means , and the average value of all the video frames with respect to the ratio of the edge amount and the texture amount per video frame, a certain interval A first feature amount that derives at least one first feature amount of a feature amount that is a maximum value and a standard deviation value of a set of average values for each time or a motion amount that is a feature amount derived from an encoded video before transcoding; Feature amount extraction means;
Second feature quantity extraction means for deriving a second feature quantity from difference information using the number of pixels in the horizontal and vertical directions of the input encoded video before transcoding and transcoded video after transcoding;
A second evaluation value indicating the quality evaluation value of the transcoded video after transcoding from the first feature quantity derived from the first feature quantity extraction means and the second feature quantity derived from the second feature quantity extraction means. Second video quality estimation means for deriving a video quality evaluation value.

Further, the present invention (Claim 4) is the first feature amount extraction means according to Claim 1 or 2,
Based on the edge amount and texture amount extracted in units of video frames by the edge / texture amount extraction means, the total value of the edge amounts for all video frames of the estimation target video, and similarly, for all video frames of the estimation target video Video frame average feature quantity deriving means for calculating a total value of the texture quantities and deriving an average video frame feature quantity obtained by dividing the total value of the edge quantities by the total value of the texture quantities is included.

Further, the present invention (Claim 5) is the first feature amount extraction means according to Claim 1, 2, or 4,
A video frame feature for deriving a video frame feature amount obtained by dividing the total value of the edge amount by the total value of the texture amount for each video frame based on the edge amount and the texture amount extracted for each video frame by the edge / texture amount extraction means A quantity derivation means;
Deriving an average feature value by deriving an average feature value obtained by summing the video frame feature values derived from the video frame feature value deriving means for a plurality of video frames in a certain section of the video to be estimated and dividing by the number of video frames Means,
The maximum feature amount indicating the maximum value of the plurality of average feature amounts of all the estimation target sections is calculated with respect to the average feature amount of a plurality of video frames in a certain section of the estimation target video derived by the average feature amount deriving unit. A maximum feature amount deriving means for deriving;
A standard deviation feature value obtained by taking a standard deviation of a plurality of average feature values in all sections to be estimated with respect to an average feature value of a plurality of video frames in a certain section of the estimation target video derived by the average feature quantity deriving means. And standard deviation feature quantity deriving means for deriving.

Further, the present invention (Claim 6) is the first feature amount extraction means according to Claims 1, 2, 4, or 5,
It further has a motion amount deriving means for deriving a motion amount indicating the motion of the video from the input encoded video before transcoding.

Further, the present invention (Claim 7) is the second feature amount extraction means according to Claim 1 or 2,
PSNR deriving means for deriving a peak-signal-to-noise ratio (PSNR) from the input encoded video before transcoding and transcoded video after transcoding is included.

  FIG. 2 is a diagram for explaining the principle of the present invention.

The present invention (Claim 8) is a transcoded video quality objective evaluation method for objectively evaluating the quality of transcoded video,
Transcode video quality objective evaluation device
An edge / texture amount extraction step (step 1) for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistical amount derived from the edge amount and the texture amount derived from the edge / texture amount extraction step , and an average value of all the video frames with respect to the ratio of the edge amount and the texture amount per video frame, a certain interval A first feature amount that derives at least one first feature amount of a feature amount that is a maximum value and a standard deviation value of a set of average values for each time or a motion amount that is a feature amount derived from an encoded video before transcoding; Feature amount extraction step (step 2),
A second feature amount extraction step (step for deriving a second feature amount from difference information using the number of pixels in the horizontal direction and the vertical direction of the input encoded video before transcoding and transcoded video after transcoding 3) and
First video quality estimation for deriving, as a first video quality evaluation value, a video quality evaluation value of an input encoded video before transcoding from the first feature quantity output from the first feature extraction step Step (step 4);
A second video quality evaluation value indicating a quality evaluation value of the transcoded video after transcoding is derived from the first video quality evaluation value and the second feature quantity derived from the second feature quantity extraction step. The video quality estimation step (step 5) is performed.

The present invention (claim 9) is a transcoded video quality objective evaluation method for objectively evaluating the quality of transcoded video,
Transcode video quality objective evaluation device
An edge / texture amount extraction step for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistical amount derived from the edge amount and the texture amount derived from the edge / texture amount extraction step , and an average value of all the video frames with respect to the ratio of the edge amount and the texture amount per video frame, a certain interval A first feature amount that derives at least one first feature amount of a feature amount that is a maximum value and a standard deviation value of a set of average values for each time or a motion amount that is a feature amount derived from an encoded video before transcoding; A feature extraction step of
A second feature amount extraction step of deriving a second feature amount from difference information using the number of pixels in the horizontal direction and the vertical direction of the input encoded video before transcoding and transcoded video after transcoding;
A second evaluation value indicating a transcoded video quality evaluation value from the first feature value derived from the first feature value extraction step and the second feature value derived from the second feature value extraction step. And a second video quality estimation step for deriving a video quality evaluation value.

Further, according to the present invention (Claim 10), in the edge texture amount extraction step of Claim 8 or 9,
A clustering step of capturing an input encoded video before transcoding to create an edge image, performing clustering based on the value of each pixel of the edge image, and deriving a clustered image;
Filtering the clustering image derived by the clustering step to create a filtered image;
A difference clustering image derivation step for creating a difference clustering image from a difference value between the clustering image derived by clustering and the filtering image derived from the filtering step;
The counting step of counting the edge amount and the texture amount of the difference clustering image derived by the difference clustering image deriving step for each video frame is performed.

The present invention (Claim 11) is characterized in that, in the first feature amount extraction step of Claim 8 or 9,
Based on the edge amount and texture amount extracted for each video frame in the edge / texture amount extraction step, the total value of the edge amount for all video frames of the estimation target video, and similarly for all video frames of the estimation target video A video frame average feature quantity derivation step is performed in which a total value of the texture quantities is calculated, and a video frame average feature quantity is derived by dividing the total edge quantity by the total texture quantity.

Further, the present invention (Claim 12) is characterized in that, in the first feature amount extraction step according to Claim 8, 9, or 11,
A video frame feature that derives a video frame feature amount by dividing the total edge amount by the total texture amount for each video frame based on the edge amount and the texture amount extracted for each video frame in the edge / texture amount extraction step. A quantity derivation step;
Deriving an average feature value by summing up the video frame feature values derived from the video frame feature value deriving step for a plurality of video frames in a certain section of the video to be estimated, and deriving an average feature value divided by the number of video frames Steps,
The maximum feature quantity indicating the maximum value of the plurality of average feature quantities of all the estimation target sections is calculated with respect to the average feature quantity of a plurality of video frames in a certain section of the estimation target video derived from the average feature quantity deriving step. A maximum feature amount derivation step to be derived;
The standard deviation feature value obtained by taking the standard deviation of the multiple average feature values of all the estimation target sections from the average feature value of multiple video frames in a certain section of the estimation target video derived from the average feature value deriving step. And a standard deviation feature quantity derivation step for deriving.

Further, the present invention (Claim 13) is the first feature amount extraction step according to Claim 8, 9, 11 or 12,
A motion amount deriving step for deriving a motion amount indicating the motion of the video from the input encoded video before transcoding is further performed.

The present invention (Claim 14) is characterized in that, in the second feature amount extraction step of Claim 8 or 9,
A PSNR deriving step of deriving a peak-signal-to-noise ratio (PSNR) from the input encoded video before transcoding and transcoded video after transcoding is performed.

  The present invention (Claim 15) is a transcoded video quality objective evaluation program for causing a computer to function as each means constituting the transcoded video quality objective evaluation apparatus according to any one of claims 1 to 7. .

  Conventionally, block noise, blur noise, mosquito noise, flicker noise, etc., which are extracted from the original video and transcoded video, the encoded original video before transcoding is input. Noise, blur (blurred) noise, mosquito noise, flicker noise, etc. cannot be extracted properly, and the video quality estimation accuracy has been significantly reduced. In addition, in the technology that estimates the video quality using only the received video and the received packet, the encoded video before transcoding cannot be used, so the deterioration of the quality of the encoded video before transcoding cannot be captured. The quality estimation accuracy was significantly reduced.

  On the other hand, according to the present invention, the quality of the encoded video before transcoding is derived as the first video quality evaluation value, the feature quantity is derived from both the encoded video and the transcoded video, A quality evaluation value is appropriately estimated for the transcoded video by deriving the second video quality evaluation value from the video quality evaluation value of 1, the feature amount derived from the encoded video and the transcoded video. be able to.

  Accordingly, the video communication service provider can monitor the video quality value of the video of the video communication service actually viewed by the user according to the present invention. It is possible to easily determine whether or not it is maintained.

  Therefore, the video communication service provider can grasp and manage the actual quality of the service being provided in more detail than before.

It is a principle block diagram of this invention. It is a figure for demonstrating the principle of this invention. It is a block diagram of the video quality objective evaluation apparatus in the 1st Embodiment of this invention. It is a figure which shows the area | region division method in the 1st Embodiment of this invention. It is a figure for demonstrating notionally the derivation process of the difference clustering image in the 1st Embodiment of this invention. It is a flowchart of operation | movement of the video quality objective evaluation apparatus in the 1st Embodiment of this invention. It is a block diagram of the video quality objective evaluation apparatus in the 2nd Embodiment of this invention. It is a flowchart of operation | movement of the video quality objective evaluation apparatus in the 2nd Embodiment of this invention.

  Embodiments of the present invention will be described below with reference to the drawings.

[First Embodiment]
The transcoded video quality objective evaluation device according to this embodiment of the present invention performs clustering processing on an encoded video, creates a clustered image, derives a feature amount from the clustered image, and encodes the encoded video before transcoding. The first video quality evaluation value that quantitatively represents the video quality of the video is derived, then the PSNR (Peak Signal to Noise Ratio) is derived using the encoded video and the transcoded video, and the first video quality evaluation The second video quality evaluation value (transcoded video quality evaluation value) indicating the quality of the transcoded video from the value and the PSNR is realized by objective evaluation.

  For example, in this embodiment, in order to realize objective video quality evaluation in video communication such as IPTV service and video distribution service performed via an IP network such as the Internet, encoded video and transcode before transcoding are used. The later transcoded video is analyzed, and a transcoded video quality value that quantitatively represents a feature quantity that affects the video quality related to the video communication is derived.

  FIG. 3 shows a configuration of the video quality objective evaluation apparatus according to the embodiment of the present invention.

  As shown in the figure, the video quality objective evaluation apparatus 1 includes an edge / texture amount extraction unit 100, a first feature amount extraction unit 200, a first video quality estimation unit 300, and a second feature amount extraction. Unit 400 and a second video quality estimation unit 500.

  The edge / texture amount extraction unit 100 includes a clustering unit 10, a filtering unit 11, a difference clustering image deriving unit 12, and a counting unit 13.

  The first feature quantity extraction unit 200 includes a video frame average feature quantity derivation section 14, a video frame feature quantity derivation section 15, an average feature quantity derivation section 16, a maximum feature quantity derivation section 17, and a standard deviation feature quantity derivation section. 18 and a motion amount deriving unit 19.

  The clustering unit 10 applies Sobel filters Fx and Fy to the input encoded video I (x, y, f) before transcoding based on the method of Non-Patent Document 6 described above, and performs the vertical direction (Y direction). Edge images, edge images Sx (x, y, f) and Sy (x, y, f) in the horizontal direction (X direction) are derived. Specifically, an edge image is created based on the following formula (1) using information on eight pixels adjacent to the target pixel.

各画素に対し導出された垂直及び水平方向のエッジ値を、図４に示す領域分割方法で１〜４の４つのクラスに分類されたクラスタリング画像（クラスタリング画像はＣ１〜Ｃ４に対応するクラスタリング値１〜４の値を持つ）を作成する。

Clustering images in which the vertical and horizontal edge values derived for each pixel are classified into four classes 1 to 4 by the region dividing method shown in FIG. 4 (the clustering image is a clustering value 1 corresponding to C1 to C4). With a value of ~ 4).

  The filtering unit 11 creates a filtered image by applying an erosion filter and a dilatation filter to the clustered image output from the clustering unit 10 as shown in FIG. FIG. 5 shows an example of no padding (A) and padding (B) for images clustered into four classes (8 × 8 pixel images).

  Specifically, Erosion filter is applied to clustering image C (x, y) (x is horizontal pixel position, y is vertical pixel position) based on the following formula (2) to create Erosion image E (x, y) To do. Next, a dilatation filter is applied to the Erosion image based on the following formula (3) to create a dilatation image D (x, y) (that is, a filtered image). By applying two types of filtering, a filtered image D (x, y) obtained by smoothing the clustering image C (x, y) can be obtained, and C3 and C4 can be appropriately extracted from the difference between the two.

ただし、画面の端の画素に対しては、画面外に画素がないため、画面の端の画素値をパディングするか、もしくは画面の端の画素を処理しないことで対応する。

However, since there is no pixel outside the screen, the pixel at the edge of the screen is dealt with by padding the pixel value at the edge of the screen or not processing the pixel at the edge of the screen.

  The difference clustering image derivation unit 12 is a difference value between the clustering image C (x, y) output from the clustering unit 10 and the filtering image D (x, y) output from the filtering unit 11 (difference clustering image: C (x, y) -D (x, y) +1, and if C (x, y) -D (x, y) +1 is less than 1, then 1) is derived.

  The counting unit 13 counts the number of pixels NC3 in which the clustering value is 3 (edge amount) and the number of pixels NC4 in which the clustering value is 4 (texture amount) in the difference clustering image.

  As shown in the following equation (4), the video frame average feature amount deriving unit 14 adds the NC3 for each video frame output from the counting unit 13 over all video frames and divides the value by the number of video frames. An image frame average feature amount P obtained by adding NC4 for each frame over all the image frames and dividing by the value obtained by dividing by the number of image frames is derived.

ここで、fは映像フレーム番号、Ｆは総映像フレーム数（例えば、推定対象の映像が３０ｆｐｓ、１０秒分の場合、３００フレームとなる）を示す。

Here, f indicates the video frame number, and F indicates the total number of video frames (for example, if the estimation target video is 30 fps and 10 seconds, it is 300 frames).

  The video frame feature quantity deriving unit 15, as shown in the following equation (5), for each video frame, the video frame feature obtained by dividing NC3 for each video frame output from the counting unit 13 by NC4 for each video frame. The quantity P (f) is derived based on the following equation (5).

ここで、fは映像フレーム番号を示す。

Here, f indicates a video frame number.

  The average feature amount deriving unit 16 sets a certain section (Z frame, for example, 10 frames) as one section, and averages the average value of the video frame feature amounts P (f) in the section by the following equation (6). The feature amount is Pt (k).

ここで、fは映像フレーム番号、ｋは平均区間の番号（例えば、推定対象の映像が３００フレームから構成され、Ｚが１０フレームの場合、ｋは０〜２９の値となる）を示す。

Here, f represents a video frame number, and k represents an average interval number (for example, when the estimation target video is composed of 300 frames and Z is 10 frames, k is a value from 0 to 29).

  The maximum feature amount deriving unit 17 derives the maximum value among the average feature amounts Pt (k) output from the average feature amount deriving unit 16 as the maximum feature amount MaxPt (for example, Pt (0) to Pt (29 ), If Pt (10) is the maximum, Pt (10) becomes MaxPt).

標準偏差特徴量導出部１８は、平均特徴量導出部１６より出力された平均特徴量Pt（k）の標準偏差を標準偏差特徴量StdPtとして導出する。

The standard deviation feature quantity derivation unit 18 derives the standard deviation of the average feature quantity Pt (k) output from the average feature quantity derivation unit 16 as the standard deviation feature quantity StdPt.

ここで、Stdevは、標準偏差を計算する演算子を表す。

Here, Stdev represents an operator for calculating a standard deviation.

  The motion amount deriving unit 19 derives a motion amount TI indicating the motion between video frames. P, MaxPt, and StdPt found above are statistical values of feature values derived for each video frame, and are feature values that are unrelated to video motion. Is possible.

ここで、式（７）のI（x，y，f）は水平方向ｘ、垂直方向ｙ、映像フレーム番号ｆの画素を示す。また、式（７）、（８）のM（f）はI（x，y，f）とI（x，y，f−1）との差分画像、式（８）、（９）のTI（f）は映像フレーム毎のフレーム関差分値の標準偏差、式（９）のＦは総映像フレーム数（例えば、30fpsの映像１０秒分の場合、３００フレームとなる）を示す。

Here, I (x, y, f) in Expression (7) indicates pixels in the horizontal direction x, the vertical direction y, and the video frame number f. Further, M (f) in equations (7) and (8) is a difference image between I (x, y, f) and I (x, y, f−1), and TI in equations (8) and (9). (F) indicates the standard deviation of the frame-related difference value for each video frame, and F in Expression (9) indicates the total number of video frames (for example, 300 frames for 10 seconds of 30 fps video).

  The first video quality estimation unit 300 includes a video frame average feature P output from the video frame average feature deriving unit 14, a maximum feature MaxPt output from the maximum feature deriving unit 17, and a standard deviation feature deriving unit. The first video quality evaluation value (Vq) is derived from the standard deviation feature amount StdPt output from 18 and the motion amount TI output from the motion amount deriving unit 19 based on the following equation (10).

ここで、a〜iは、映像フォーマット（例えば、QCIF，VGA，HDなど）に固有の係数とする。

Here, a to i are coefficients specific to the video format (for example, QCIF, VGA, HD, etc.).

  However, Vq is expressed by a cubic function, but may be expressed by other mathematical expressions as shown below.

ここで、j〜lは、映像フォーマット（例えば、QCIF，VGA，HDなど）に固有の係数とする。

Here, j to l are coefficients specific to the video format (for example, QCIF, VGA, HD, etc.).

  The second feature amount extraction unit 400 includes the PSNR derivation unit 20.

  PSNR is derived using the following equations (12) to (14).

ここで、ｙ_transcode（x,y,f））は水平方向ｘ、垂直方向ｙ、フレーム番号ｆのトランスコード映像の画素を示し、ｙ_code（x,y,f）は水平方向ｘ、垂直方向ｙ、フレーム番号ｆの符号化映像の画素を示し、Ｗは水平方向画素の総数、Ｈは垂直方向画素の総数、Ｎは総映像フレーム数を示す。

Here, y _transcode (x, y, f)) represents the pixel of the transcoded video image in the horizontal direction x, vertical direction y, and frame number f, and y _code (x, y, f) represents the horizontal direction x, vertical direction. y represents the pixel of the encoded video of frame number f, W represents the total number of horizontal pixels, H represents the total number of vertical pixels, and N represents the total number of video frames.

  The second video quality estimation unit 500 uses the first video quality evaluation value Vq and the feature amount PSNR to derive a second video quality evaluation value based on the following equation (15).

ここで、m，n，oは映像フォーマット（例えば、QCIF，VGA，HDなど）に固有の係数とする。

Here, m, n, and o are coefficients specific to the video format (for example, QCIF, VGA, HD, etc.).

However, in this embodiment, the quality evaluation value of the transcoded video is derived using only the PNSR as the feature quantity extracted from both the coded video and the transcoded video. It is also possible to evaluate the quality evaluation value of the transcoded video by weighting and adding feature quantities such as Min _HV shown in .247.

  Next, the operation of the video quality objective evaluation device 1 according to the present embodiment will be described.

  FIG. 6 is a flowchart of the operation of the video quality objective evaluation device according to the first embodiment of the present invention.

When the deteriorated video is input to the clustering unit 10 of the video quality objective evaluation apparatus 1 (S101), the clustering unit 10 inputs the encoded video y _code (x, y, f) (x is the horizontal direction, y Is the vertical direction, f is the frame number), and Sobel filters Fx and Fy are applied to derive edge images Sx (x, y, f) and Sy (x, y, f), and edge image Sx (x, y, f ) And Sy (x, y, f) and clustering into classes 1 to 4 to derive a clustered image C (x, y, f) (S102), to the filtering unit 11 and the differential clustering image deriving unit 12. Output.

  The filtering unit 11 receives the clustering image C (x, y, f) derived by the clustering unit 10 and applies an erosion filter and a dilatation filter. The filtering image D (x , y, f) is derived (S103) and output to the difference clustering image deriving unit 12.

  The difference clustering image derivation unit 12 takes the difference value of the filtering image D (x, y, f) output from the filtering unit 11 from the clustering image C (x, y, f) output from the clustering unit 10. And a difference clustering image having values of 1 to 4 is derived (S104) and output to the counting unit 13.

  The counting unit 13 counts the number of pixels of C3 (edge amount) and C4 (texture amount) of the difference clustering image output from the difference clustering image deriving unit 12 for each video frame (S105), and derives the video frame average feature amount. Output to the unit 14 and the video frame feature quantity deriving unit 15.

  The video frame average feature quantity deriving unit 14 divides the total number of C3 by the total number of C4 by the above-described equation (4) with respect to the total number of C3 and C4 pixels output from the counting unit 13 for all video frames. The frame average feature amount P is derived (S106) and output to the video quality estimation unit 20.

  The video frame feature quantity deriving unit 15 divides the number of C3 and C4 pixels output from the counting unit 13 for each video frame by the number of C4 pixels by the above-described equation (5) to obtain the video. A video frame feature amount P (f) for each frame is derived (S107) and output to the average feature amount deriving unit 16.

  The average feature amount deriving unit 16 averages the video frame feature amount P (f) output from the video frame feature amount deriving unit 15 by the above-described equation (6) for each specific number of video frames, and calculates the average feature amount Pt ( k) is derived (S108) and output to the maximum feature quantity deriving unit 17 and the standard deviation feature quantity deriving unit 18.

  The maximum feature amount deriving unit 17 derives the maximum feature amount MaxPt that is the maximum among the average feature amounts Pt (k) (S109), and outputs the maximum feature amount MaxPt to the video quality estimation unit 20.

  The standard deviation feature quantity deriving unit 18 derives a standard deviation feature quantity StdPt indicating the standard deviation of the average feature quantity Pt (k) (S110) and outputs it to the video quality estimation unit 20.

  The motion amount deriving unit 19 derives a motion amount TI indicating the motion of the video (S111) and outputs it to the video quality estimating unit 20.

  The first video quality estimation unit 300 calculates the video quality evaluation value Vq from the video frame average feature amount P, the maximum feature amount MaxPt, the standard deviation feature amount StdPt, and the motion amount TI according to the above equation (10) or equation (11). Deriving (S112), the first video quality evaluation value Vq is output.

The second feature quantity extraction unit 400 calculates the feature quantity PSNR from the above-described equations (12) to (14) from the encoded video y _code (x, y, f) and the transcoded video y _transcode (x, y, f). Is derived (S113) and output to the second video quality estimation unit 500.

  The second video quality estimation unit 500 derives the second video quality evaluation value Vq_transcode from the first video quality evaluation value Vq and the feature amount PSNR by the above equation (15) (S114), and ends the process.

Thus, according to the present embodiment, the first video quality evaluation value Vq before transcoding is derived from the encoded video, and then the encoded video y _code (x, y, f) and the transcode are derived. By deriving the feature amount PSNR from both the video y _transcode (x, y, f) and using the first video quality evaluation value Vq and the second feature amount PSNR, Since the quality evaluation value Vq_transcode of the transcoded video can be calculated, it is possible to estimate the video quality by the transcoded video quality objective evaluation method which is more accurate than before.

  Therefore, the video communication service provider can easily determine whether the service being provided maintains a certain level of quality to the user, and can grasp the actual quality of the service being provided in real time.・ It becomes possible to manage.

[Second Embodiment]
In the first embodiment described above, after obtaining the first video quality evaluation value Vq, the feature amount PSNR is weighted and added to obtain the second video quality evaluation value (video quality evaluation value of the transcoded video). However, in the present embodiment, an example is shown in which the feature amount constituting the first video quality evaluation value and the feature amount PNSR are directly weighted and added.

  FIG. 7 shows the configuration of the video quality objective evaluation apparatus in the second embodiment of the present invention.

  In the figure, the same components as those in FIG. 3 are denoted by the same reference numerals, and the description thereof is omitted. In the present embodiment, the first video quality estimation unit 300 is not used, and a video quality estimation unit 600 is provided instead of the second video quality estimation unit 500.

  In the video quality objective evaluation device 2 of the present embodiment, the video frame average feature value deriving unit 14, the maximum feature value deriving unit 17, the standard deviation feature value deriving unit 18, and the motion amount deriving unit 19 This is output to the video quality estimation unit 600 instead of the video quality estimation unit 300.

  FIG. 8 is a flowchart of the operation of the video quality objective evaluation device according to the second embodiment of the present invention. S201 to S211 are the same as S101 to S111 of FIG. 6 of the first embodiment, but in S212, the PSNR deriving unit 20 derives the PSNR based on the input encoded video and transcoded video. After that, in S213, the video quality estimation unit 600 obtains the video frame average feature quantity P acquired from the video frame average feature quantity derivation unit 14 and the maximum feature quantity obtained from the maximum feature quantity derivation unit 17 as shown in Expression (16). The feature amount MaxPt, the standard deviation feature amount StdPt acquired from the standard deviation feature amount deriving unit 18, the motion amount TI acquired from the motion amount deriving unit 19 and the feature amount PSNR acquired from the PNSR deriving unit 20 are directly weighted and added. The video quality evaluation value Vq_transcod of the code video is directly derived.

ここで、u，v，w，x，y，zは映像フォーマット（例えば、QCIF，VGA，HDなど）に固有の係数とする。

Here, u, v, w, x, y, and z are coefficients specific to the video format (for example, QCIF, VGA, HD, etc.).

However, in this embodiment, the quality evaluation value of the transcoded video is derived using only the PSNR as the feature quantity extracted from both the coded video and the transcoded video. It is also possible to evaluate the quality evaluation value of the transcoded video by weighting and adding feature quantities such as Min _HV shown in .247.

  The video quality objective evaluation apparatuses 1 and 2 in the first and second embodiments described above are realized by installing a computer program in a computer including a CPU (Central Processing Unit), a memory, and an interface. Various functions of the video quality objective evaluation apparatus 1 are realized by cooperation of various hardware resources of the computer and the computer program (software).

  The operations of the constituent elements of the video quality objective evaluation devices 1 and 2 shown in FIGS. 3 and 7 are constructed as a program and installed and executed on a computer used as the video quality objective evaluation device, or via a network. It can be distributed.

  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 embodiment, and various modifications and applications can be made within the scope of the claims.

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

1, 2 Video quality objective evaluation device 10 Clustering unit 11 Filtering unit 12 Difference clustering image deriving unit 13 Counting unit 14 Video frame average feature deriving unit 15 Video frame feature deriving unit 16 Average feature deriving unit 17 Maximum feature deriving unit 18 standard deviation feature quantity deriving section 19 motion quantity deriving section 20 PSNR deriving section 100 edge / texture quantity extraction means, edge / texture quantity extraction section 200 first feature quantity extraction means, first feature quantity extraction section 300 first Video quality estimation means, first video quality estimation section 400 second feature quantity extraction means, second feature quantity extraction section 500 Second video quality estimation means 600 Video quality estimation section

Claims (15)

A transcoded video quality objective evaluation device that objectively evaluates the quality of transcoded video,
Edge / texture amount extraction means for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistic derived from the edge amount and the texture amount derived from the edge / texture amount extraction means , and an average value of all video frames with respect to a ratio of the edge amount and the texture amount per video frame; A feature amount that is a maximum value and a standard deviation value of a set of average values for every certain interval, or at least one first feature amount of a motion amount that is a feature amount derived from an encoded video before the transcoding. First feature amount extraction means for deriving;
Second feature quantity extraction means for deriving a second feature quantity from difference information using the number of pixels in the horizontal and vertical directions of the input encoded video before transcoding and transcoded video after transcoding;
First video quality for deriving a video quality evaluation value of the input encoded video before transcoding as a first video quality evaluation value from the first feature quantity output from the first feature quantity extraction means Means,
A second video quality evaluation value indicating a quality evaluation value of a transcoded video after transcoding is derived from the first video quality evaluation value and the second feature quantity derived from the second feature quantity extraction means. Second video quality estimating means for
A transcoded video quality objective evaluation apparatus comprising: A transcoded video quality objective evaluation device that objectively evaluates the quality of transcoded video,
Edge / texture amount extraction means for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistic derived from the edge amount and the texture amount derived from the edge / texture amount extraction means , and an average value of all video frames with respect to a ratio of the edge amount and the texture amount per video frame; A feature amount that is a maximum value and a standard deviation value of a set of average values for every certain interval, or at least one first feature amount of a motion amount that is a feature amount derived from an encoded video before the transcoding. First feature amount extraction means for deriving;
Second feature quantity extraction means for deriving a second feature quantity from difference information using the number of pixels in the horizontal and vertical directions of the input encoded video before transcoding and transcoded video after transcoding;
The quality evaluation value of the transcoded video transcoded from the first feature extraction and the second feature quantity derived and the derived first feature quantity from the second feature extraction means than means Second video quality estimation means for deriving a second video quality evaluation value to be indicated;
A transcoded video quality objective evaluation apparatus comprising: The edge / texture amount extraction means includes:
Clustering means for capturing the input encoded video before transcoding and creating an edge image, performing clustering based on the value of each pixel of the edge image, and deriving the clustered image;
Filtering the clustering image derived by the clustering unit to create a filtered image;
A difference clustering image derivation unit that creates a difference clustering image from a difference value between the clustering image derived by the clustering and the filtering image derived from the filtering unit;
Counting means for counting the edge amount and the texture amount of the difference clustering image derived by the difference clustering image deriving means in units of video frames;
The transcoded video quality objective evaluation apparatus according to claim 1 or 2, characterized by comprising: The first feature amount extraction means includes:
Based on the edge amount and the texture amount extracted in units of video frames by the edge / texture amount extraction means, the total value of the edge amounts for all video frames of the estimation target video, as well as all video frames of the estimation target video A video frame average feature amount derivation unit is provided for calculating a total value of texture amounts with respect to, and deriving an average video frame feature amount obtained by dividing the total value of edge amounts by the total value of texture amounts. The transcoded video quality objective evaluation apparatus according to 1 or 2. The first feature amount extraction means includes:
A video frame for deriving a video frame feature value obtained by dividing a total value of edge amounts by a total value of texture amounts in units of video frames based on the edge amount and texture amount extracted in units of video frames by the edge / texture amount extraction unit A feature derivation means;
Average feature value for deriving an average feature value obtained by summing up the video frame feature values derived from the video frame feature value deriving means for a plurality of video frames in a certain section of the video to be estimated and dividing by the number of video frames Deriving means;
A maximum feature that indicates a maximum value of a plurality of average feature values of all sections to be estimated with respect to the average feature quantity of a plurality of video frames in a certain section of the estimation target video derived by the average feature quantity deriving unit. Maximum feature amount deriving means for deriving the amount;
A standard deviation obtained by taking a standard deviation of a plurality of average feature values of all sections to be estimated with respect to the average feature quantity of a plurality of video frames in a certain section of the estimation target video derived by the average feature quantity deriving unit. Standard deviation feature quantity deriving means for deriving the feature quantity;
The transcoded video quality objective evaluation apparatus according to claim 1, 2, or 4, further comprising: The first feature amount extraction means includes:
6. The transcoded video quality according to claim 1, 2, 4 or 5, further comprising motion amount deriving means for deriving a motion amount indicating the motion of the video from the input encoded video before transcoding. Objective evaluation device. The second feature amount extraction unit includes:
3. The PSNR deriving means for deriving a peak-signal-to-noise ratio (PSNR) from the input encoded video before transcoding and transcoded video after transcoding. Transcode video quality objective evaluation device. A transcoded video quality objective evaluation method for objectively evaluating the quality of transcoded video,
Transcode video quality objective evaluation device
An edge / texture amount extraction step for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistical amount derived from the edge amount and the texture amount derived from the edge / texture amount extraction step , and an average value of all video frames with respect to a ratio of the edge amount and the texture amount per video frame; A feature amount that is a maximum value and a standard deviation value of a set of average values for every certain interval, or at least one first feature amount of a motion amount that is a feature amount derived from an encoded video before the transcoding. A first feature amount extraction step to be derived;
A second feature amount extraction step of deriving a second feature amount from difference information using the number of pixels in the horizontal direction and the vertical direction of the input encoded video before transcoding and transcoded video after transcoding;
First video quality for deriving a video quality evaluation value of an input encoded video before transcoding as a first video quality evaluation value from the first feature value output from the first feature value extraction step An estimation step;
A second video quality evaluation value indicating a quality evaluation value of the transcoded video after transcoding is derived from the first video quality evaluation value and the second feature quantity derived from the second feature quantity extraction step. A second video quality estimation step,
Transcode video quality objective evaluation method characterized by A transcoded video quality objective evaluation method for objectively evaluating the quality of transcoded video,
Transcode video quality objective evaluation device
An edge / texture amount extraction step for deriving an edge amount and a texture amount from the input encoded video before transcoding;
A statistical amount derived from the edge amount and the texture amount derived from the edge / texture amount extraction step , and an average value of all video frames with respect to a ratio of the edge amount and the texture amount per video frame; A feature amount that is a maximum value and a standard deviation value of a set of average values for every certain interval, or at least one first feature amount of a motion amount that is a feature amount derived from an encoded video before the transcoding. A first feature amount extraction step to be derived;
A second feature amount extraction step of deriving a second feature amount from difference information using the number of pixels in the horizontal direction and the vertical direction of the input encoded video before transcoding and transcoded video after transcoding;
A quality evaluation value of a transcoded video after transcoding is calculated from the first feature quantity derived from the first feature quantity extraction step and the second feature quantity derived from the second feature quantity extraction step. A second video quality estimation step for deriving a second video quality evaluation value to be indicated;
Transcode video quality objective evaluation method characterized by In the edge / texture amount extraction step,
A clustering step of capturing an input encoded video before transcoding to create an edge image, performing clustering based on the value of each pixel of the edge image, and deriving a clustered image;
Filtering the clustering image derived by the clustering step to create a filtered image;
A difference clustering image derivation step for creating a difference clustering image from a difference value between the clustering image derived by the clustering and the filtering image derived from the filtering step;
A counting step of counting the edge amount and the texture amount of the difference clustering image derived by the difference clustering image derivation step in units of video frames;
10. The transcoded video quality objective evaluation method according to claim 8 or 9, wherein: In the first feature amount extraction step,
Based on the edge amount and texture amount extracted in units of video frames in the edge / texture amount extraction step, the total value of edge amounts for all video frames of the estimation target video, as well as all video frames of the estimation target video A video frame average feature amount derivation step is performed, wherein a total value of texture amounts is calculated and a video frame average feature amount is derived by dividing the total amount of edge amounts by the total value of texture amounts. The transcoded video quality objective evaluation method according to 8 or 9. In the first feature amount extraction step,
A video frame for deriving a video frame feature amount obtained by dividing the total value of edge amounts by the total value of texture amounts in units of video frames based on the edge amount and texture amount extracted in units of video frames in the edge / texture amount extraction step A feature amount derivation step;
Average feature value for deriving an average feature value obtained by summing the video frame feature values derived from the video frame feature value derivation step for a plurality of video frames in a certain section of the estimation target video and dividing by the number of video frames A derivation step;
The maximum feature that indicates the maximum value of the plurality of average feature values of all the sections to be estimated with respect to the average feature amount of the plurality of video frames in a certain section of the estimation target image that is derived from the average feature amount derivation step. A maximum feature amount deriving step for deriving an amount;
A standard deviation obtained by taking a standard deviation of a plurality of average feature values of all sections to be estimated with respect to the average feature quantity of a plurality of video frames in a certain section of the estimation target video derived from the average feature quantity derivation step. A standard deviation feature amount deriving step for deriving the feature amount;
The transcoded video quality objective evaluation method according to claim 8, 9 or 11, further comprising: In the first feature amount extraction step,
13. The transcoded video quality according to claim 8, 9, 11, or 12, further comprising a motion amount deriving step of deriving a motion amount indicating a motion of the video from the input encoded video before transcoding. Objective evaluation method. In the second feature amount extraction step,
10. The PSNR deriving step of deriving a peak-signal-to-noise ratio (PSNR) from the inputted encoded video before transcoding and transcoded video after transcoding. Transcode video quality objective evaluation method.   A transcoded video quality objective evaluation program for causing a computer to function as each means constituting the transcoded video quality objective evaluation device according to any one of claims 1 to 7.

Images