Coding parameter calculation device, coding parameter calculation method, and program

By dividing videos into scenes and predicting user viewing discontinuation rates, the bitrate ladder is optimized for adaptive bitrate streaming, improving user experience and reducing network usage.

JP2024543480AActive Publication Date: 2024-11-21NIPPON TELEGRAPH & TELEPHONE CORP
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
JP2024528465
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2021-11-17
Publication Date
2024-11-21
Estimated Expiration
2041-11-17

AI Technical Summary

Technical Problem

Existing methods for defining bitrate ladders in adaptive bitrate video streaming fail to consider user behavior, leading to suboptimal video quality and increased network usage, without addressing the likelihood of users discontinuing the service.

Method used

A device and method that divides videos into scenes, applies multiple bit rate and resolution combinations, evaluates quality using VMAF, and predicts user viewing discontinuation rates to optimize the bitrate ladder based on content encoding complexity and network throughput.

Benefits of technology

The solution reduces the likelihood of user service discontinuation by optimizing bitrate ladders, balancing quality and network usage, considering both video and audio quality, and historical throughput data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024543480000001_ABST
    Figure 2024543480000001_ABST
Patent Text Reader

Abstract

1. An encoding parameter calculation device comprising: a processor; and a memory, the memory including instructions for causing the processor to execute the steps of: dividing an original video into a plurality of scenes and selecting one of the plurality of scenes; applying a plurality of combinations of bit rate and resolution to the selected one of the scenes to generate a plurality of encoded scenes; evaluating quality of each of the encoded scenes; estimating a relationship between a viewing abandonment rate and a bit rate for the one of the scenes; applying the relationship to throughput history data for the one of the scenes to obtain a plurality of values ​​of the viewing abandonment rate; applying a cluster algorithm to the plurality of values ​​of the viewing abandonment rate to calculate N discrete quantization values ​​of the viewing abandonment rate; and identifying a plurality of combinations of bit rate, resolution, and quality corresponding to the N discrete quantization values.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to an encoding parameter calculation device, an encoding parameter calculation method, and a program. [Background technology]

[0002] The increasing use of video streaming services such as YouTube, Netflix, and Twitch presents new challenges. As the amount of data transmitted increases, video encoding needs to be optimized to conserve storage capacity and reduce network usage. Since network conditions vary depending on the load on the network and the location of the user, adaptive bitrate (ABR) video streaming should be used to ensure optimal viewing quality for the viewer. To use ABR video streaming, multiple parameters such as a combination of resolution, frame rate, and bitrate must be defined. Each set of parameters defines a different throughput requirement, and these multiple parameter sets allow streaming in various network conditions. Therefore, the video is encoded at multiple different bitrate values. The encoding of the video at multiple different bitrate values ​​is called a bitrate ladder.

[0003] There are several different ways to define a bitrate ladder. First, there are pre-defined coding recepts. These are tables published by video streaming platforms that provide recommendations on encoding settings, such as bitrate values ​​for various resolutions. A clear limitation of this approach is that the bitrate requirements to achieve high quality are not the same for all content. Thus, using the same bitrate values ​​for different content would result in an over-allocation of bits for footage that is easy to code, wasting space on the server and using the network unnecessarily. On the other hand, footage that is not easy to code may show coding artifacts and result in poor quality.

[0004] To address this issue, it is necessary to define a bitrate ladder based on the content encoding complexity. Defining a bitrate ladder based on the content encoding complexity is called "per-title video encoding". A common way to achieve this is to encode video at multiple resolutions and bitrate values ​​and evaluate the video using a quality metric such as Video Multi-method Assessment Fusion (VMAF). VMAF is a computer program that uses machine learning to predict the quality of encoded and / or rescaled video. The algorithm inputs the encoded video and the source content before encoding, and predicts the quality of the encoded video by measuring the change in pixel values ​​before and after encoding. Such an algorithm is used to measure the rate-distortion curve for each resolution, taking into account the encoding complexity of the video content and specifying the resolution to be used at a given bitrate. For the bitrate value, it is common to set the value so that the video quality is separated by one Just Noticeable Difference (JND). Scene-by-scene analysis, which is an extension of this method, has also been proposed.

[0005] However, there are still some challenges. First, the method requires a huge amount of computation and depends on multiple encoding trials. Second, the method reduces the dimension of the search space of the encoding conditions to one dimension (bit rate), but does not define the optimal value to be used. To address the computational complexity, a support vector regression method has been proposed that uses both spatial and temporal texture features. This method predicts the range of bit rates at which a given resolution should be used while significantly reducing the number of encodings. Additionally, original solutions based on machine learning have also been proposed that directly predict the encoding complexity or bit rate ladder. However, the details of the algorithms are not public, making it difficult to compare or evaluate these methods. To address the problem of network-optimized bit rate ladders, researchers have used real throughput history data to define bit rate ladders based on probabilistic models with the aim of ensuring minimum quality while minimizing network usage. An analytical evaluation method for bit rate ladders based on nonlinear constrained optimization is known. A simulation model of a player is defined, and a bit rate ladder is defined to maximize session quality while considering throughput probabilities based on historical data. This method has been extended to support multiple video codecs in the same ladder. The problem of identifying a bitrate ladder is solved using integer linear programming while maximizing the Quality of Experience (QoE) measured using the National Telecommunications and Information Administration Video Quality Metric (NTIA VQM).

[0006] However, these works share certain limitations in common: while the above mentioned methods address the problem of optimizing the ladder for the network, they are based on a model of a video player whose behavior is simple (always selecting the highest available quality based on known future throughput values). A player that always selects the highest available quality is limited in this case, since it would result in undesirable frequent quality changes. Therefore, a player-independent method for selecting a content- and network-optimized bitrate ladder is proposed, which corresponds to a more advanced video player model that utilizes more advanced algorithms. [Prior art documents] [Non-patent literature]

[0007] [Non-Patent Document 1] P. Lebreton and K. Yamagishi, "Network and content-dependent bitrate ladder estimation for adaptive bitrate video streaming," IEEE International Conference on Acoustics, Speech and Signal Processing, 2021, pp.1-5. [Non-Patent Document 2] S. Chen, X. Nie, D. Fan, D. Zhang, V. Bhat, R. Hamid, "Shot Conservative Self-Supervised Learning for Scene Boundary Detection," Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 1-10. Summary of the Invention [Problem to be solved by the invention]

[0008] Streaming video to a large audience poses many challenges. Each user accesses the Internet with different performance in terms of available throughput, which may vary as a function of time. Therefore, streaming video to all users requires the use of different representations (video encoded at various resolutions, bit rates, and frame rate values) to be able to adapt the video bit rate requirements to match the user's available throughput. Such a method of adapting the quality to the available throughput is called adaptive bit rate video streaming, and the bit rate values ​​corresponding to each possible representation are called the bit rate ladder.

[0009] When defining the bitrate ladder, it is necessary to set sufficient granularity in terms of bitrate values ​​so that each user can enjoy the best quality allowed for their Internet access. However, from the operator's point of view, it is also necessary to avoid unnecessary use of the network and unnecessary use of storage space on the servers. Therefore, to optimize the user experience while reducing operational costs, it is necessary to optimally encode the video.

[0010] Finding an optimal bitrate ladder has already been addressed. However, some challenges remain. There is a need to define a bitrate ladder that improves the user experience and reduces the likelihood that users will abandon the service. [Means for solving the problem]

[0011] According to one aspect of the present invention, there is provided an encoding parameter calculation device comprising a processor; and a memory, the memory including instructions for causing the processor to execute the steps of: dividing an original video into a plurality of scenes and selecting one of the plurality of scenes; applying a plurality of combinations of bit rate and resolution to the selected one scene to generate a plurality of encoded scenes; evaluating quality of each of the encoded scenes; estimating a relationship between a viewing abandonment rate and a bit rate for the one scene; applying the relationship to throughput history data for the one scene to obtain a plurality of values ​​of the viewing abandonment rate; applying a cluster algorithm to the plurality of values ​​of the viewing abandonment rate to calculate N discrete quantization values ​​of the viewing abandonment rate; and identifying a plurality of combinations of bit rate, resolution, and quality corresponding to the N discrete quantization values. Effect of the Invention

[0012] According to an embodiment, it is possible to define a bit rate ladder that improves the user experience and reduces the likelihood that the user will abandon the service. [Brief description of the drawings]

[0013] [Figure 1] 1 is a diagram illustrating an example of a hardware configuration of an encoding parameter calculation device 10 according to an embodiment. [Diagram 2] 1 is a diagram illustrating an example of a functional configuration of an encoding parameter calculation device 10 according to an embodiment of the present invention. [Diagram 3] 11 is a flowchart illustrating an example of a processing procedure executed by the encoding parameter calculation device 10. [Figure 4] FIG. 1 is a diagram showing an example of dividing a video into scenes. [Diagram 5] FIG. 1 shows an example of content encoding complexity analysis with rate-distortion curves at various resolutions. [Figure 6]FIG. 1 illustrates an example of identifying the highest achievable quality among multiple resolutions. [Figure 7] FIG. 13 is a diagram illustrating an example of the relationship between bit rate and viewing abandonment rate. [Figure 8] FIG. 13 is a diagram showing an example of a time series of throughput for a viewing session. [Figure 9] FIG. 1 shows an example time series of maximum achievable quality taking into account available throughput. [Figure 10] FIG. 13 is a diagram showing an example of a time series of viewing abandonment rates. [Figure 11] FIG. 1 illustrates an example distribution function of abandonment rates across multiple streaming sessions. [Figure 12] FIG. 13 illustrates an example of identifying steps in a bit rate ladder by optimal quantization. [Figure 13] FIG. 13 illustrates an example of back-projection of abandonment rates to bitrate values ​​based on abandonment rate analysis. [Figure 14] FIG. 13 illustrates an example of using content encoding complexity analysis to identify resolutions to be used corresponding to abandonment rates. [Figure 15] FIG. 1 illustrates an example of a content and network-based optimized bitrate ladder. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] As described in the background section, finding the optimal bitrate ladder has already been addressed. However, some challenges remain. In fact, previous studies have shown that it is possible to optimize the bitrate ladder based on quality, but previous studies have not addressed the behavior of stopping viewing when defining the bitrate ladder. It is difficult to address the behavior of stopping viewing because the relationship between the quality of the service controlled by the bitrate value in the ladder and whether the user wants to continue using the service is not clear. This type of ladder decision mechanism is important because service providers are interested in whether the user wants to use the service or not, not the quality itself. Therefore, it is necessary to develop a bitrate ladder optimization strategy that considers the impact of quality on whether the user will continue using the service or stop using it. Since the relationship between quality and acceptability is time-dependent, it is also a new challenge to consider the duration of the stimulus when defining the ladder. This point has also not been addressed in previous studies and will be considered in the following examples.

[0015] Considering that the definition of the encoding conditions is not clear, in the following example a solution is described for optimizing the bitrate ladder with respect to the content characteristics and network statistics.

[0016] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a diagram showing an example of a hardware configuration of an encoding parameter calculation device 10 in an embodiment of the present invention. The encoding parameter calculation device 10 in Fig. 1 includes a drive device 100, an auxiliary storage device 102, a memory device 103, a central processing unit (CPU) 104, and an interface device 105, which are all connected to each other via a bus B.

[0017] A program for realizing the processing in the encoding parameter calculation device 10 is provided by a recording medium 101 such as a CD-ROM. When the recording medium 101 storing the program is set in the drive device 100, the program is installed from the recording medium 101 to the auxiliary storage device 102 via the drive device 100. However, the program does not necessarily have to be installed from the recording medium 101, but may be downloaded from another computer via a network. The auxiliary storage device 102 stores the installed program as well as necessary files, data, etc.

[0018] When an instruction to start the program is received, the memory device 103 reads out the program from the auxiliary storage device 102 and stores it. The CPU 104 executes functions related to the coding parameter calculation device 10 in accordance with the program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

[0019] Fig. 2 is a diagram showing an example of a functional configuration of an encoding parameter calculation device 10 in an embodiment of the present invention. In Fig. 2, the encoding parameter calculation device 10 has a division unit 11, an encoding unit 12, a viewing abandonment rate calculation unit 13, a throughput input unit 14, a conversion unit 15, and an analysis unit 16. Each of these units is realized by a process executed by the CPU 104 of one or more programs installed in the encoding parameter calculation device 10. In other words, each of these units is realized by cooperation between the hardware resources of the encoding parameter calculation device 10 and the programs (software) installed in the encoding parameter calculation device 10.

[0020] The following describes an example of a processing procedure executed by the coding parameter calculation device 10. FIG.

[0021] In step S101, the division unit 11 analyzes the input original video and divides the original video into a plurality of scenes. In general, a scene refers to a period of action at a certain location. A scene in a video is a collection of a plurality of series of shots at a certain location.

[0022] FIG. 4 is a diagram showing an example of dividing an original video into multiple scenes. There are two main purposes for dividing an original video into multiple scenes: Since the coding complexity in the original video differs from scene to scene, it is possible to define an optimal bit rate ladder for each scene. Also, by dividing the original video into multiple scenes, the duration of each scene can be obtained. Since the acceptability of a given quality model has a temporal aspect, it is important to specify the duration of a video scene.

[0023] A known method can be used to divide an original video into a plurality of scenes. For example, the method described in Non-Patent Document 2 can be used. The method in Non-Patent Document 2 is just one example, and the method of dividing an original video into a plurality of scenes is not limited to the method in Non-Patent Document 2.

[0024] In the following process, as described above, the original video is divided into multiple scenes, and a bit rate ladder is defined for each scene. The process for each scene obtained by dividing the original video into multiple scenes is described below.

[0025] In step S102, the encoding unit 12 encodes one scene to be evaluated using a plurality of combinations of a plurality of resolutions and a plurality of bit rates.

[0026] Since the coding complexity and bitrate requirements vary from scene to scene, the coding unit 12 codes the scenes using a constant rate factor (CRF) that is in the range of 12 to 48 for AVC and HEVC and is separated by 4, and in the range of 12 to 60 for VP8, VP9 and AV1. This results in 10 to 14 different coded scenes for each resolution. Coding with a constant rate factor allows coding the scenes with a constant quantization that depends on the motion within the frame. For frames with large motion, a higher value of the quantization parameter (QP) is used, which has the effect of masking the coding compression distortion due to motion, and allows achieving a higher coding efficiency. Furthermore, by using a constant rate factor coded scene in a wide range, an almost uniform sampling can be obtained with respect to the quality of the possible coding conditions.

[0027] In step S103, the encoding unit 12 evaluates the quality of the multiple encoded scenes obtained using different quantization parameters using a content-dependent quality assessment method such as Video Multimethod Assessment Fusion (VMAF). For example, the encoding unit 12 uses VMAF to calculate an index value that indicates an estimate of the viewer's quality of experience (QoE) for each of the multiple encoded scenes.

[0028] VMAF is a full-reference method that calculates the quality difference by comparing the original video signal with the received video signal. That is, VMAF compares the encoded video with the original video and evaluates the video quality on a scale of 0 to 100, where 0 is the lowest quality and 100 is the highest quality. When using a quality evaluation method other than VMAF, the quality evaluation value may be different from the range of 0 to 100. For example, the output quality value may range from 1 to 5.

[0029] By performing the above process, the characteristics (relationship between the coding parameter value and QoE) shown in Fig. 5 are obtained. Furthermore, by drawing a curve so as to obtain the maximum QoE for the bit rate, the dashed curve in Fig. 5 is obtained. This dashed line is generally called the Covex hull. The encoding unit 12 derives the Covex hull shown by the dashed curve in Fig. 5.

[0030] The convex hull is represented by the dashed curve in Figure 5. The convex hull shows the highest quality that can be obtained when changing the resolution among all resolutions (all finite resolutions) for one bit rate.

[0031] Fig. 6 is a diagram showing an example of a process for deriving a convex hull. In the example of Fig. 6, the encoding unit 12 identifies the highest quality value obtained when the resolution is changed among all resolutions (all finite number of resolutions) for each bit rate value.

[0032] In Fig. 6, each encoded scene is indicated by a star. To derive the convex hull, the encoding unit 12 compares, for each encoded scene, the quality assessment result of a scene having the same bit rate as the encoded scene but a different resolution.

[0033] Figure 6 shows an example of this procedure. A scene encoded at 720p with a bit rate value b is encoded at a resolution of 720p with a quality value of Q 720 indicates the quality value Q at 1080p resolution 1080 Shows. Q 1080 >Q 720 And Q 1080 is the highest quality achievable for bit rate b, so the convex hull is 1080 ] is calculated based on.

[0034] Here, there is no encoded video with resolution 1080p and bit rate b, so Q 1080As shown in FIG. 6, the encoding unit 12 estimates Q by using linear regression on two quality values ​​corresponding to bit rate values ​​before and after bit rate value b of the 1080p encoded image. 1080 Estimate the following.

[0035] By applying this procedure, the encoding unit 12 obtains, for each bit rate, the highest quality value that can be obtained when the resolution is changed among all resolutions (all finite number of resolutions).

[0036] A model (e.g., a convex hull) that associates each bit rate with the highest quality value obtained when changing the resolution among all resolutions (all finite resolutions) can be formulated by the sigmoid function shown in the following equation 1.

[0037]

number

[0038] When VMAF is used, the quality value is determined in the range of [0, 100]. However, in a user abandonment prediction model that predicts the possibility of abandoning viewing, video quality is evaluated in the range of [1, 5]. Therefore, the encoding unit 12 maps the VMAF quality value evaluated using the convex hull to the range [1, 5] using a linear function defined by Equation 2.

[0039]

number

[0040] When quality assessment methods other than VMAF are used, this linear function may be maintained to align the quality prediction with the baseline data, so that the assessment method used gives a uniform quality value.

[0041] After identifying the relationship between bitrate and content encoding quality, the abandonment rate calculation unit 13 models the probability that a user will abandon watching at the end of a scene due to poor quality using Equation 3.

[0042]

number

[0043] The viewing stop rate calculation unit 13 obtains MOSv, which is the video quality, using Equation 2. The viewing stop rate calculation unit 13 obtains MOSa, which is the audio quality, using Equation 4.

[0044]

number

[0045] In Equation 4, B1, B2, and B3 are model parameters obtained using regression analysis on a database for learning speech quality evaluation. The model parameters B1, B2, and B3 are independent of the content and are constant for all scenes. Since the coding efficiency differs for each speech codec, the model parameters B1, B2, and B3 depend only on the speech codec used to code the speech.

[0046] A table of model parameters B1, B2, B3 for audio codecs is obtained using regression analysis on different databases. The viewing abandonment rate calculation unit 13 selects appropriate model parameters B1, B2, B3 based on the codec used for audio encoding.

[0047] In Formula 4, b a is the audio bitrate. For the audio bitrate there are different options possible:

[0048] The first option is to set the audio bitrate to a high quality level, for example 384 kbps, so that dropout depends only on the video quality.

[0049] However, the user abandonment prediction model also takes into account the quality of the audio. It is therefore possible to define a relationship between the video bitrate and the audio bitrate. For example, a linear relationship could be specified, such as setting the audio bitrate at 1 / 42 of the video bitrate, limited to a range between 32 kbps and 384 kbps. Alternatively, a more general logistic function between the video and audio bitrates could be specified.

[0050] In step S104, the viewing abandonment rate calculation unit 13 uses the video quality evaluation and the audio quality evaluation to evaluate the viewing abandonment rate after the viewing session of duration d.

[0051] The duration of a scene is obtained using the scene segmentation analysis shown in Figure 4. Based on this model, the stop rate calculation unit 13 obtains the relationship between the stop rate and the bit rate for each scene in step S105. Figure 7 shows an example of the relationship between the stop rate and the bit rate for the scene under evaluation.

[0052] In step S106, the throughput input unit 14 inputs network throughput history data for the scene to be evaluated. If multiple viewings (distributions) are performed within a certain period of time, multiple time series data may be input. The time axis of each time series data corresponds to the video viewing time. From such time series data, the throughput input unit 14 can obtain characteristics (relationship between viewing time and throughput) as shown in FIG. 8 for each past viewing (distribution).

[0053] 8 is a diagram showing an example of a time series of throughput values ​​for a scene to be evaluated. The throughput values ​​shown in FIG. 8 indicate a data transmission rate (bandwidth) that a client can use.

[0054] In step S107, the converter 15 converts the time series of throughput values ​​into achievable quality values. For example, the converter 15 converts the time series of throughput values ​​into achievable quality values ​​by applying the throughput values ​​as bit rates in the formulas (1)-(4). Figure 9 is a diagram showing an example of converting the time series of throughput values ​​into achievable quality values.

[0055] In step S108, the conversion unit 15 converts the time series of the throughput values ​​into a viewing abandonment rate. For example, the conversion unit 15 converts the time series of the throughput values ​​into a viewing abandonment rate time series based on the relationship between the viewing abandonment rate and the bit rate obtained in step S105. Fig. 10 is a diagram showing an example of converting the time series of the throughput values ​​into a viewing abandonment rate.

[0056] In step S109, the analysis unit 16 derives a distribution function of the viewing abandonment rate by analyzing the viewing abandonment rate for the scene to be evaluated based on the network throughput history data for the scene to be evaluated. For example, the analysis unit 16 converts a time series of throughput values ​​for the scene to be evaluated obtained for each past viewing (distribution) into a time series of viewing abandonment rates. From the multiple time series of viewing abandonment rates obtained in this way, multiple viewing abandonment rates may be collected, and the frequency of each viewing abandonment rate value may be analyzed to derive the distribution function of the viewing abandonment rate.

[0057] FIG. 11 is a diagram showing an example of a distribution function of the viewing abandonment rate for the scene to be evaluated.

[0058] Given all possible values ​​of quality and abandonment rates, identifying a bitrate ladder for the scene under evaluation can be expressed as an optimal quantization problem.

[0059] Finding the optimal bitrate ladder can be thought of as identifying the N discrete values ​​that better approximate the continuous values ​​of the abandonment rate. Let L in Equation 5 below be defined as the set of N discrete values ​​of the bitrate ladder.

[0060]

number

[0061] The problem of identifying an optimal bitrate ladder can be formulated by Equation 6 below.

[0062]

number

number

number

number

[0063] In formula 6

number

number

[0064] As described in Equation 6, the quantization error is expressed as the value of the viewing abandonment rate.

number

[0065] In step S110, the ladder calculation unit 17 calculates the number of pieces of data as the value of the viewing abandonment rate.

number

number

[0066] In addition, the ladder calculation unit 17 is

number

number

[0067] After the k-means cluster assignment has converged, the value of N,

number

[0068] FIG. 12 is a diagram showing an example of a result of classification using the k-means method.

[0069] Abandonment rate value

number

number

number

[0070] In step S110, the ladder calculation unit 17 calculates the value of the viewing abandonment rate.

number

number

[0071] In step S110, the ladder calculation unit obtains the resolution value to be used together with these bitrate values ​​from the coding complexity analysis performed when identifying the Covex hull. Furthermore, the ladder calculation unit 17 identifies the QoE corresponding to the resolution to be used using the Covex hull. Figure 14 is a diagram showing an example of identifying the resolution to be used and the corresponding QoE corresponding to the abandonment rate using the coding complexity analysis.

[0072] Therefore, the coding parameter calculation device 10 can obtain a set of a bit rate value, a resolution, and a QoE for each quantized representative value of the viewing abandonment rate value. As a result, a bit rate ladder in which the resolution, the bit rate (VBR), and the QoE are associated with each other is obtained as shown in FIG.

[0073] It has been shown that it is possible to perform a video coding complexity analysis in terms of user abandonment rates. Compared to traditional analysis methods using quality-based evaluation methods and quality-based rate-distortion curves, performing the analysis in terms of user abandonment rates allows taking into account the duration of the stimulus and the temporal impact of low quality on user viewing time.

[0074] Although quality is an important metric, knowing the impact of quality on viewing time and viewing behavior is important for network operators, so analyzing by user abandonment rate gives a high degree of interpretability of the impact of bitrate ladder selection on services.

[0075] According to this embodiment, it is possible to define a bitrate ladder to minimize user abandonment behavior. The optimization of the bitrate ladder is based on historical data of content encoding complexity and network throughput. It has been shown that it is possible to estimate the abandonment rate of users by combining the historical data of content encoding complexity and network throughput. It is possible to select a bitrate ladder to use when encoding video based on the combination of the historical data of content encoding complexity and network throughput.

[0076] Furthermore, according to this embodiment, it is possible to define a bitrate ladder that optimizes video encoding complexity as measured by user abandonment rate, without prior knowledge of the video player and the adaptive bitrate control scheme used by the video player.

[0077] Furthermore, according to this embodiment, by using a model of the viewing abandonment rate, it is possible to define a bit rate ladder to be used for video encoding, taking into account audio quality as well as video quality.

[0078] Although the embodiment of the present invention has been described in detail above, the present invention is not limited to such specific embodiment, and various modifications and variations are possible within the scope of the gist of the present invention described in the claims.

Claims

1. An encoding parameter calculation device, A processor; and Memory, Equipped with The memory may include: Dividing the original video into a plurality of scenes and selecting one of the plurality of scenes; applying multiple combinations of bit rates and resolutions to the selected scene to generate multiple encoded scenes; assessing the quality of each of the encoded scenes; estimating a relationship between abandonment rate and bit rate for the one scene; applying the relationship to historical throughput data for the scene to obtain a plurality of values ​​of the abandonment rate; applying a cluster algorithm to the plurality of values ​​of the abandonment rate to calculate N discrete quantized values ​​of the abandonment rate; and identifying a plurality of combinations of bitrate, resolution, and quality corresponding to the N discrete quantization values; containing instructions to execute Coding parameter calculation device.

2. In the step of calculating N discrete quantized values ​​of the abandonment rate, the processor applies a cluster classification algorithm to values ​​of the abandonment rate that are equal to or less than a predetermined threshold value among the plurality of values.

2. The encoding parameter calculation device according to claim 1.

3. In the step of estimating the relationship, the processor includes an audio bitrate and a video bitrate as the bitrates.

2. The encoding parameter calculation device according to claim 1.

4. the bit rate of the audio is a predetermined high bit rate; 4. The encoding parameter calculation device according to claim 3.

5. The audio bit rate is a bit rate obtained by multiplying the video bit rate by a specific coefficient.

4. The encoding parameter calculation device according to claim 3.

6. 1. A computer-implemented method for calculating an encoding parameter, comprising the steps of: Dividing the original video into a plurality of scenes and selecting one of the plurality of scenes; applying multiple combinations of bit rates and resolutions to the selected scene to generate multiple encoded scenes; assessing the quality of each of the encoded scenes; estimating a relationship between abandonment rate and bit rate for the one scene; applying the relationship to historical throughput data for the scene to obtain a plurality of values ​​of the abandonment rate; applying a cluster algorithm to the plurality of values ​​of the abandonment rate to calculate N discrete quantized values ​​of the abandonment rate; and identifying a plurality of combinations of bitrate, resolution, and quality corresponding to the N discrete quantization values; An encoding parameter calculation method comprising:

7. A non-transitory computer-readable storage medium storing an encoding parameter calculation program, wherein when the encoding parameter calculation program is executed by a computer, the computer: Dividing the original video into a plurality of scenes and selecting one of the plurality of scenes; applying multiple combinations of bit rates and resolutions to the selected scene to generate multiple encoded scenes; assessing the quality of each of the encoded scenes; estimating a relationship between abandonment rate and bit rate for the one scene; applying the relationship to historical throughput data for the scene to obtain a plurality of values ​​of the abandonment rate; applying a cluster algorithm to the plurality of values ​​of the abandonment rate to calculate N discrete quantized values ​​of the abandonment rate; and identifying a plurality of combinations of bitrate, resolution, and quality corresponding to the N discrete quantization values; A non-transitory computer-readable storage medium for executing the method of the present invention.

Citation Information

Patent Citations

  • Mathematical model derivation device, mathematical model derivation method, and program

    WO2021181681A1

  • Mathematical model deriving device, mathematical model deriving method and program

    WO2021181724A1