JP2016178641A - Method and apparatus for measurement of quality of video based on frame loss pattern - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and apparatus for measuring quality of video on the basis of a frame loss pattern.SOLUTION: The method comprises: generating a frame loss pattern of video by indicating whether each frame in the video is lost or successfully transmitted; and evaluating quality of the video as a function of the generated frame loss pattern.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a method and apparatus for measuring video quality based on frame loss patterns.

  This section is intended to introduce the reader to various aspects of the technology that may be associated with various aspects of the invention described and / or claimed below. This discussion is intended to help provide readers with background information to make the various aspects of the present invention easier to understand. Accordingly, it should be understood that such a description should be read in this regard and not as prior art approval.

  In the transmission of digitally compressed video, a fairly significant source of failure is due to the delivery of video streams over error-prone channels. Partial loss and corruption of information can have a dramatic impact on the user's perceived quality because local distortions in the frame propagate spatially and temporally through the frame. The visual impact of such frame loss depends on the video decoder's ability to deal with corrupted streams. In some examples, the decoder can determine to drop some frames spontaneously. For example, the decoder can drop or discard all frames with corrupted or missing information and instead repeat the previous video frame until the next valid decoded frame is available. The encoder can also drop the frame when the content movement increases rapidly if the target encoding bit rate is too low. In all the above examples, it can be said that frame loss occurs in the video.

  In many existing video quality monitoring products, the overall media video quality is analyzed based on three main coding artifacts: jerkiness, blockiness and blurring. Block noise and blur are the two main types of spatial coding artifacts, which behave as block boundary discontinuities and high frequency losses, respectively. On the other hand, jerkiness is the most significant time artifact.

  The time degradation of video quality caused by a set of group frame losses is called jerkiness, and group frame loss means the fact that one or more consecutive frames of a video sequence have been lost together.

  There are several studies on assessing the perceptual impact of video frame loss (periodic or non-periodic) on perceived video quality.

  In Non-Patent Document 1, it is pointed out that humans are usually quite tolerant of consistent frame loss, and its negative impact is greatly related to the consistency of frame loss, and is therefore used as a measure of jerkiness. It became so.

  Non-Patent Document 2 describes the relationship between the perceptual impact of jerkiness and the length and frequency of group frame loss.

K. C. Yang, C. C. Guest, K. EI-Maleh and P. K. Das, “Perceptual Temporal Quality Metric for Compressed Video”, IEEETransaction on Multimedia, vol.9, no.7, Nov. 2007, pp.1528-1535. R. R. Pastrama-Vidal and J. C. Gicquel, “Automatic Quality Assessment of Video Fluidity Impairments Using a No-Reference Metric”, the 2nd International Workshop on Video Processing and Quality Metric for Consumer Electronics, Scottsdale, USA 22-24, Jan. 2006. L. Y. Duan, J. Q. Wamg et al, “Shot-LevelCamera Motion Estimation based on a Parametric Model”

  The inventors of the present invention have found that video frame loss patterns have a significant impact on the perceived impact of jerkiness and eventually impact on the overall video quality. “Frame loss pattern” means a pattern generated by sequentially recording the state of whether each frame of a video sequence was successfully transmitted or lost when transmitted using a different representation.

  Thus, the present invention takes advantage of this discovery by providing a method for measuring the quality of video and corresponding devices.

  In one embodiment, a method for measuring video quality is provided. The method generates a video frame loss pattern by indicating whether each frame of the video was lost or transmitted successfully, and evaluates the video quality as a function of the generated frame loss pattern Prepare for that.

  In one embodiment, an apparatus for measuring video quality is provided. The apparatus comprises means for receiving an input video and generating a frame loss pattern for the received video, and means for evaluating the quality of the video as a function of the generated frame loss pattern.

  Advantageous embodiments of the invention are disclosed in the dependent claims, the following description and the drawings.

Exemplary embodiments of the invention will be described with reference to the accompanying drawings.
FIG. 6 is an exemplary diagram illustrating some examples of video frame loss patterns. 3 is a flow diagram illustrating a method for measuring video quality based on a frame loss pattern, in accordance with an embodiment of the present invention. FIG. 6 is an exemplary diagram illustrating that a frame loss pattern is divided into two subsections. FIG. 2 is an exemplary diagram illustrating a framework for a method for measuring video quality. FIG. 2 is a block diagram illustrating an apparatus for measuring video quality based on a frame loss pattern according to an embodiment of the present invention. FIG. 2 shows an interface of a software tool designed to perform a subjective test of video quality.

  In the following description, various aspects of embodiments of the present invention will be described. For purposes of explanation, specific configurations and details are set forth in order to provide a thorough understanding. However, it will be apparent to those skilled in the art that the present invention may be practiced without the specific details presented herein.

  FIG. 1 is an exemplary diagram illustrating some examples of video frame loss patterns. As described above, the frame loss pattern is generated by sequentially recording the state of each frame of the video using different representations. For example, the frame loss pattern can be represented by a sequence of 0-1 such as “1, 1, 1, 1, 0, 0, 1, 1, 1, 0, 1”, where “0” “1” indicates that the frame was successfully transmitted. Frame loss patterns can also be in the form of diagrams using high / low levels, representing successful frame transmission and frame loss, respectively. FIG. 1 (a) shows the frame loss pattern of such a video, where each low level (valley) in the figure represents a group frame loss that includes one or more consecutive frame losses and transmitted successfully. The rendered frame is represented by the high level (top) of the figure. FIG. 1 (b) shows two different frame loss patterns, “Pattern 1” and “Pattern 2,” both consisting of 10 group frame losses, each group frame loss having the same length. ing. According to the research results of Non-Patent Documents 1 and 2, which do not regard the frame loss pattern as one factor affecting the perceptual jerkiness of video, the video produced by the above two “pattern 1” and “pattern 2” The quality degradation should be very similar. However, the inventor's work of the present invention arrives at a very different conclusion from the above case.

According to the inventor's discovery, the perceptual impact of jerkiness is greatly influenced by the frame loss pattern. The following case (1) to case (3), which have the same frame loss rate, are taken as examples.
Case (1): One frame is lost every two frames.
Case (2): Four consecutive frames are lost every 8 frames.
Case (3): The second half of the video sequence is completely lost.

  The overall frame loss rate in all three cases above is 50 percent. However, their perceptual impact is quite different. In Case 1, the viewer will perceive a clear dithering but will feel unwell after a long period of browsing. In Case 3, the viewer does not perceive such kind of phenomenon, but will face freezing over time. That is, completely different perceptions result from different frame loss patterns having the same frame loss rate.

  In accordance with an embodiment of the present invention, a method is provided for measuring video quality based on the above findings.

  FIG. 2 is a flowchart illustrating a method for measuring video quality based on a frame loss pattern according to an embodiment of the present invention.

  As shown in FIG. 2, the method comprises the following steps.

  Step 201: A frame loss pattern is generated by indicating the state (lost or transmitted successfully) of each frame of the video sequence.

  Step 202: Group one or more consecutive lost frames of a video sequence from the first lost frame into a group frame loss.

  Step 203: Dividing the frame loss pattern from the first group frame loss into multiple sections having one or more consecutive group frame losses, and each group frame loss in the section Have the same number of successfully transmitted frames and the same number of lost frames between the loss and the previous group frame loss.

  Step 204: Calculate quality degradation values generated by each section of group frame loss.

  Step 205: Evaluate the quality of the video sequence by combining the values of all sections.

  A detailed description will now be given with reference to the accompanying drawings.

  In the method according to an embodiment of the invention, firstly a frame loss pattern of the video sequence is generated. This generation can be accomplished by indicating the status of all frames of the video sequence in an appropriate manner. One skilled in the art will recognize that frame loss can be detected by well-known methods. Details on this point will not be described further.

  Next, starting from the first of all the lost frames in the video sequence, one or more consecutive lost frames are grouped into a group called group frame loss.

  A possible frame loss pattern is shown.

This pattern is given by a time stamp, where gdi represents the i-th group frame loss. The frame loss pattern is then divided (or segmented) into a plurality of subsections from the first group frame loss. Each subsection comprises one or more consecutive group frame losses, each group frame loss having a similar perceptual impact on the quality degradation of the video sequence.

  For the above purpose by dividing successive group frame losses into similar perceptual impacts, in this method each group frame loss of the video sequence can be specified by two parameters gd = (gpgd, lengthd). . The first parameter gappgd is the number of frames successfully transmitted between the current group frame loss and the previous group frame loss. The second parameter lengthd is the number of lost frames of the current group frame loss. Both gapgd and lengthd values are limited to integers from 1 to 10. If all group frame losses of a segmented subsection have the same parameters gappgd and lengthd, those parameters have a similar perceptual impact on the quality degradation of the video sequence.

  Distance function

Can be used as the magnitude of the difference in perceptual impact between two group frame losses in a subsection. In the above distance function, the function f (x, y) is used for evaluating the perceptual quality of the video, which will be described later.

  The frame loss pattern is then segmented into subsections based on the distance function definition.

  The following is pseudo code for the distance function.

Segmentation procedure.
1, nCount = 0, pool = {}; // defined as numbers / subsection set 2, (i = 0, i <= n, i ++) {2, 1 if each gd∈pool is ( If d (gd, gdi) <c),
Insert gdi into the pool otherwise {SubSectionnCount = pool; pool = {}; nCount ++};}
The above pseudo code c is a constant. This procedure sets the frame loss pattern FLP to a set of subsections.

Divide into

  FIG. 3 shows an example in which the above frame loss pattern is divided into two subsections. As shown in FIG. 3, group frame loss within one subsection is considered to consist of similar perceptual impacts, and two adjacent subsections are linked to several successfully transmitted frames. Yes.

  Next, a perceptual evaluation of each subsection is performed.

  It will be appreciated that sub-sections can typically be treated as regularly spaced frame losses since group frame losses within the same sub-section are considered to consist of similar perceptual impacts.

  Therefore, each subsection is also specified by two parameters SubSection = (gapSS, lenSS). The first parameter gapSS is the average number of frames successfully transmitted between two adjacent group frame losses, and the second parameter lenSS is the average number of lost frames of all group frame losses in the subsection. is there.

  Put simply, the feature values of the gapSS and lenSS subsections are exactly the average values of gapSS and lenSS for all group frame losses in that subsection.

  Therefore, it is conceivable that the degradation of the perceptual quality of the subsection is determined by the feature values of gapSS and lenSS. It is defined as follows.

  By subjective examination, perceptual quality evaluation can be manually marked for some discrete values of (gapSS, lenSS). For this reason, we express the discrete function f (x, y) as x, yε {1, 2,. . . , 10}.

  As an example, the discrete function f (x, y) can be defined as follows:

  Here, cstill is a constant used as a threshold, and CameraMotion is the size of the camera motion level of the subsection. And f1 (x, y) f2 (x, y) is given in the following table.

  Since camera motion is another important factor that affects perceived quality, it is also necessary to estimate the level of camera motion.

  The camera motion can be estimated by a known method. The most important global motion estimation model is an 8-parameter perspective motion model described in Non-Patent Document 3.

  Non-Patent Document 3 becomes clear by the following equation.

  Where (aD,..., A7) are global motion parameters, (xi, yi) indicates the spatial coordinates of the i th pixel of the current frame, and (xi ′, yi ′) is Indicates the spatial coordinates of the corresponding pixel in the previous frame. A relationship between motion model parameters and symbol level interpretation is established.

The algorithm introduced in Non-Patent Document 3 is applied to extract an 8-parameter GME model in the method of the embodiment of the present invention. The level of camera motion is finally defined as follows.

So fp (x, y) = f (x, y) is x, yε {1,2,. . , 10}. There is also a problem with how to generalize the function fp (x, y) into a function with non-integer variables, and typical training is a problem. Thus, using a learning machine (eg, an artificial neural network (ANN), well known to those skilled in the art), while the machine learns with f (x, y)

Can be assigned.

  So far, the perceptual quality degradation value Jp generated by each subsection of the frame loss pattern has been obtained.

  Finally, the quality of the video sequence is evaluated by combining the values of all sections of the frame loss pattern.

  In this way, pooling strategies can be used to integrate those values into the overall video sequence quality assessment. It should be pointed out that such a temporal quality pooling strategy is quite different from the pooling strategy when considering spatial artifacts such as block noise and blur. Due to the nature of the human visual system (HVS), people are “easy to feel bad and unforgivable”. A successfully transmitted frame between two subsections of higher temporal quality is usually ignored when considering the overall temporal quality.

  In the above segmentation step, the video sequence is lost at regular intervals.

Each two adjacent subsections are segmented into a set of subsections and separated by a number of successfully transmitted frames indicating NoLossi, a successfully transmitted frame between Subsectioni and Subsectioni + 1. For clarity, NoLossi is treated as a special type of regularly spaced frame loss with the least quality degradation value of 1. In other words, we

And set. All these NoLossi were then inserted into the set FLP.

  Therefore, the overall quality degradation is defined as follows.

Here, w (flpi) is a weight function of the elements of the FLP and is defined as follows.

In this equation, length (flpi) is the number of frames of flpi. dist (flpi) is the distance from the center of flpi to the last frame. Jp (flpi) is the perceptual time degradation derived by flpi, as defined above.

  fT is a function describing the human “remember and forget” property. The viewer is thought to make his / her overall assessment when he / she has finished viewing the last frame. Subsections far from the last frame may be forgotten by the viewer. The farther away, the more likely you are to forget.

  fD is a function that describes a visual property that is “prone to feel bad and unacceptable”. Humans have a strong impact on fairly distorted subsections, while almost ignoring undistorted subsections.

  FIG. 4 is an exemplary diagram illustrating a framework for a method for measuring video quality. The framework input is the received video sequence, indicating the lost / successful reception of each frame, along with a frame loss pattern (or time stamp). The output of the framework is a value J that indicates the level of video volume (jerkness) for the input video sequence.

  As shown in FIG. 4, the main part of the framework consists of (1) segmentation of the frame loss pattern of the input video, (2) perceptual evaluation of each section that is considered as regularly spaced frame loss, and (3) It consists of three actions: pooling.

  In the segmentation step, the frame loss pattern of the input video sequence is divided into sets of sections as described above. These sections can be classified into two types. One of them (SubSectioni) is composed of similar group frame loss and is considered as frame loss at regular intervals within the segment. The other type (NoLossi) does not include any frame loss.

  In the perceptual evaluation step, Nolossi's perceptual evaluation is set to a constant 1. The SubSectioni perceptual evaluation is estimated based on Equation (1) as described above.

  In the pooling step, the overall jerkiness rating is estimated based on the perceptual rating of all sections according to equation (3) as described above.

  Another embodiment of the present invention provides an apparatus for measuring video quality based on frame loss patterns.

  FIG. 5 is a block diagram illustrating an apparatus for measuring video quality based on a frame loss pattern according to an embodiment of the present invention.

  As shown in FIG. 5, an apparatus 500 receives an input video sequence and generates a frame loss pattern generation unit 501 for generating a frame loss pattern of the received video sequence, and a frame loss from the frame loss pattern generation unit 501. A grouping unit 502 for receiving the pattern and grouping one or more consecutive loss frames of the frame loss pattern from the first loss frame into a group frame loss; Dividing the frame loss pattern from the first group frame loss into a plurality of sections having one or more consecutive group frame losses, wherein each group frame loss of the section is The group A division unit 503 for having the same number of successfully transmitted frames and the same number of lost frames between a frame loss and a previous group frame loss, and the group frame loss output by the division unit 503 A calculation unit 504 for calculating the quality degradation value generated by each section, and evaluating the quality of the video sequence by combining the values of all sections output by the calculation unit 504 and outputting the value The evaluation unit 505 is provided.

  An experiment was performed to estimate the evaluation accuracy of the present invention in comparison with Non-Patent Documents 1 and 2. For this reason, a software tool was designed to perform a subjective test of video quality. FIG. 6 is a diagram showing an interface of a software tool designed to perform the subjective test. The input video sequence can be selected in the “YUV Sequence” group box on the right, and the frame loss pattern can be selected in the “Frame Loss Pattern” group box on the right center. The video sequence affected by the frame loss pattern is then displayed in a small window on the left.

The viewer is then required to mark the perception of jerkiness as follows:
1-Do not perceive quality degradation at all 2-If you look carefully and notice that there is something unnatural on the time axis, it will not affect the enjoyment of the video 3-It is clear quality degradation, 4-usually feels uncomfortable-very uncomfortable with quality degradation 5-severe degradation, can't stand video at all

  In the subjective test, a 10 CIF (352 × 288 pixel video resolution) sequence was selected and 20 frame loss patterns were selected. Three viewers are asked to score and their average value is taken into account in the subjective score, shown as JS. All sequences with marked scores are assembled into the data set DS.

Parameter setting In implementation, constants are determined based on experience. The constants are β1 = β2 = 1, β3 = 2,
c = 1.5 and cstill = 0.23.

  For simplicity, a 300-frame window has been made available for storage while the viewer would forget about the quality of the previous frame of this window. In the window, fT = 1 is set and fD (d) = 6-d is set. f1 (x, y) and f2 (x, y) are determined by Table 1 and Table 2 above.

Experimental Result The evaluation accuracy of the present invention is estimated by comparing the subjective evaluation result J obtained according to the present invention and the subjective score JS. Pearson's correlation is used to measure prediction accuracy.

  The following table shows the Pearson correlation (prediction accuracy) of the present invention and the algorithm proposed in Non-Patent Documents 1 and 2.

  Although the basic novel features applied to the preferred embodiments of the present invention have been shown, described and pointed out, the apparatus and method described, the forms and details of the disclosed devices, and the variety of their operation It will be understood that any omissions, substitutions and changes may be made by those skilled in the art without departing from the spirit of the invention. It is expressly intended that all combinations of elements performing substantially the same function in substantially the same way to achieve the same result are within the scope of the invention. The substitution of elements from one described embodiment to another is all contemplated and contemplated.

  It will be understood that the present invention has been described by way of example only and modifications of detail can be made without departing from the scope of the invention. Each feature disclosed in the description, and (where appropriate) the claims and drawings may be provided alone or in any suitable combination. Suitable features may be implemented in hardware, software, or a combination of the two.

Claims (10)

A method for measuring video quality,
Generating a frame loss pattern for the video by indicating whether each frame of the video has been lost or transmitted successfully;
Evaluating the quality of the video as a function of the generated frame loss pattern;
Said method. The evaluation is
Grouping one or more consecutive lost frames of the video from a first lost frame into a group frame loss;
The frame loss pattern from the first group frame loss is divided into a plurality of sections having one or more consecutive group frame losses, and each group frame loss of the section is divided into the group frame loss and the previous group Having the same number of successfully transmitted frames and the same number of lost frames between frame loss;
Calculating the value of quality degradation generated by each section of group frame loss;
Assessing the quality of the video by combining the values of all sections;
The method of claim 1, comprising:   The value of the quality degradation generated by the section of group frame loss is calculated as the average number of frames successfully transmitted between two adjacent group frame losses of the section and the loss frame of all the group frame losses of the section. The method of claim 2, further comprising calculating as a function of an average number of.   The method according to claim 2 or 3, wherein the evaluation comprises matching the values of all sections by weighted pooling.   The weighting factor of one section of the weighted pooling depends on the number of frames in the section, the distance from the middle of the section to the last frame, and the perceptual time degradation induced by the section. the method of. An apparatus for measuring video quality, the means for receiving an input video and generating a frame loss pattern of the received video;
Means for evaluating the quality of the video as a function of the generated frame loss pattern;
Comprising the apparatus. The means for evaluating the quality of the video comprises:
A grouping unit for receiving the frame loss pattern from a frame loss pattern generation unit and grouping one or more consecutive loss frames of the frame loss pattern from the first loss frame into a group frame loss When,
Receiving the grouped frame loss pattern from the grouping unit and dividing the frame loss pattern from an initial group frame loss into a plurality of sections having one or more consecutive group frame losses; A division unit for causing each group frame loss of a section to have the same number of successfully transmitted frames and the previous same number of lost frames between the group frame loss and the previous group frame loss When,
A calculation unit for calculating a value of quality degradation generated by each section of the group frame loss output by the division unit;
7. An apparatus according to claim 6, comprising: an evaluation unit for combining the values of all sections output by the calculation unit to evaluate the quality of the video and outputting the values.   The calculation unit calculates the value of the quality degradation generated by a section of group frame loss as an average number of frames successfully transmitted between two adjacent group frame losses of the section and all the groups of the section. The apparatus of claim 7, wherein the apparatus calculates the frame loss as a function of the average number of lost frames.   The apparatus according to claim 7 or 8, wherein the evaluation unit matches the values of all sections by weighted pooling.   The weighting factor for one section of the weighted pooling depends on the number of frames in the section, the distance from the middle of the section to the last frame, and the perceptual time degradation induced by the section. The device described.

Images