CN102984540A

CN102984540A - Video quality assessment method estimated on basis of macroblock domain distortion degree

Info

Publication number: CN102984540A
Application number: CN2012105325655A
Authority: CN
Inventors: 陈耀武; 林翔宇; 田翔
Original assignee: Zhejiang University ZJU
Current assignee: Zhejiang University ZJU
Priority date: 2012-12-07
Filing date: 2012-12-07
Publication date: 2013-03-20

Abstract

The invention discloses a video quality evaluation method based on macroblock domain distortion degree estimation, which comprises the following steps: (1) dividing a video image into several macroblocks; (2) calculating block effect distortion degree; (3) calculating blur effect Distortion; (4) Calculate brightness contrast; (5) Calculate texture complexity; (6) Calculate motion intensity contrast; (7) Calculate motion direction consistency; (8) Calculate visual perception; (9) Calculate video quality Evaluation value. In the evaluation of video quality, the model of the present invention is simple, and the objective quality can be obtained only by the video to be evaluated, which has high flexibility. At the same time, it can obtain relatively accurate evaluation results for various video scenes, and has a better universality.

Description

A Video Quality Assessment Method Based on Distortion Estimation in Macroblock Domain

技术领域 technical field

本发明属于视频质量评价技术领域，具体涉及一种基于宏块域失真度估计的视频质量评价方法。The invention belongs to the technical field of video quality evaluation, and in particular relates to a video quality evaluation method based on macroblock domain distortion degree estimation.

背景技术 Background technique

随着计算机与网络通信技术的飞速发展，人们对获取多媒体信息的需求日益旺盛。近年来，与视频相关的应用涵盖各个领域，如视频会议、视频监控和移动电视等。在这些应用中，视频信息在到达接收者之前都需要经过压缩和传输，而这些过程往往会造成视频质量损失。为了获得更好的主观效果，有必要对视频质量进行评价，根据结果调整编码器和传输信道的参数。视频的最终受体是人类的眼睛，人眼观察被认为是最精确的评价视频质量的方法。然而，由于视频的信息量非常大，依靠人工观察的主观方法对视频质量进行评价需要消耗大量的人力和时间，不适合大规模实际应用。因此，如何根据人眼视觉系统(HVS)特性建立视频质量评价模型，在此基础上由计算机自动完成视频的质量评价，成为一个非常有意义的课题。With the rapid development of computer and network communication technology, people's demand for multimedia information is increasingly strong. In recent years, video-related applications cover various fields, such as video conferencing, video surveillance, and mobile TV. In these applications, video information needs to be compressed and transmitted before reaching the receiver, and these processes often cause video quality loss. In order to obtain a better subjective effect, it is necessary to evaluate the video quality and adjust the parameters of the encoder and transmission channel according to the results. The final receptor of the video is the human eye, and human eye observation is considered to be the most accurate method for evaluating video quality. However, due to the large amount of information in the video, the subjective method of relying on manual observation to evaluate the video quality needs to consume a lot of manpower and time, which is not suitable for large-scale practical applications. Therefore, how to establish a video quality evaluation model based on the characteristics of the human visual system (HVS), and then automatically complete the video quality evaluation by computer on this basis, has become a very meaningful topic.

视频客观质量评价方法(Video Objective Quality Assessment)是指通过设计数学模型对视频进行智能化分析，并按设定的尺度对视频进行自动评分的客观评价方法。根据对原始视频的依赖程度，视频客观质量评价方法可以分为全参考型、部分参考型和无参考型三类。由于全参考型和部分参考型评价方法都需要额外的带宽来传输原始视频及相关信息，其实用价值非常有限。相比之下，无参考质量评价方法不需要依赖任何与原始视频相关的信息，直接根据待评价视频的信息计算视频质量，具有更好的灵活性和适应性，以及更广泛的应用价值。特别是在与网络多媒体相关的视频应用中，无参考视频客观质量评价在服务器质量(Quality of Service，QoS)检测和终端质量体验(Quality of Experience，QoE)上面起到重要作用，根据视频质量评价反馈信息，视频服务器可以动态调整视频编码器参数和传输信道参数，以保证传输稳定性，提高接收端视频质量。另外，无参考视频客观质量评价可以取代人眼，公正地比较不同视频编解码器输出的视频质量，为视频接收端提供参考，做出最优选择。The video objective quality assessment method (Video Objective Quality Assessment) refers to the objective evaluation method of intelligently analyzing the video by designing a mathematical model, and automatically scoring the video according to the set scale. According to the degree of dependence on the original video, video objective quality evaluation methods can be divided into three categories: full-reference type, partial-reference type and no-reference type. Since both full-reference and partial-reference evaluation methods require additional bandwidth to transmit raw video and related information, their practical value is very limited. In contrast, the no-reference quality evaluation method does not need to rely on any information related to the original video, and directly calculates the video quality based on the information of the video to be evaluated, which has better flexibility and adaptability, and wider application value. Especially in video applications related to network multimedia, no-reference video objective quality evaluation plays an important role in server quality (Quality of Service, QoS) detection and terminal quality of experience (Quality of Experience, QoE). Feedback information, the video server can dynamically adjust video encoder parameters and transmission channel parameters to ensure transmission stability and improve video quality at the receiving end. In addition, the objective quality evaluation of no-reference video can replace the human eye, fairly compare the video quality output by different video codecs, and provide reference for the video receiver to make the optimal choice.

现有的视频质量评价方法虽然取得了一定的效果，形成了一些比较成熟的模型；如传统基于PSNR(峰值信噪比)的视频质量评价方法以及Wang Zhou在标题为Image quality assessment：From error visibility to structural similarity(IEEE Transactios on Image Processing，2004，13(4))的文献中提出了一种基于SSIM(结构相似度)的视频质量评价方法；但这些方法没有考虑HVS在视频质量评价中的作用，忽视了视频内容特征对视频质量的影响，准确度还有待提高；且很难适用于不同场景的视频，普适性不高。Although the existing video quality assessment methods have achieved certain results, some relatively mature models have been formed; such as the traditional video quality assessment method based on PSNR (peak signal-to-noise ratio) and Wang Zhou's titled Image quality assessment: From error visibility To structural similarity (IEEE Transactios on Image Processing, 2004, 13(4)) proposed a video quality evaluation method based on SSIM (structural similarity); but these methods did not consider the role of HVS in video quality evaluation , ignoring the impact of video content features on video quality, the accuracy needs to be improved; and it is difficult to apply to videos in different scenes, and the universality is not high.

发明内容 Contents of the invention

针对现有技术所存在的上述技术缺陷，本发明提供了一种基于宏块域失真度估计的视频质量评价方法，其得到的指标具有较高的准确度，且能够满足各种不同场景视频的需要。Aiming at the above-mentioned technical defects existing in the prior art, the present invention provides a video quality evaluation method based on macroblock domain distortion degree estimation. need.

一种基于宏块域失真度估计的视频质量评价方法，包括如下步骤：A method for evaluating video quality based on macroblock domain distortion estimation, comprising the steps of:

(1)将待评价视频每帧图像分割成若干个宏块；(1) segment each frame of the video to be evaluated into several macroblocks;

(2)计算出宏块的块效应失真度；(2) Calculate the blockiness distortion degree of the macroblock;

(3)计算出宏块的模糊效应失真度；(3) Calculate the blur effect distortion degree of the macroblock;

(4)计算出宏块的亮度对比度；(4) Calculate the brightness contrast of the macroblock;

(5)计算出宏块的纹理复杂度；(5) Calculate the texture complexity of the macroblock;

(6)计算出宏块的运动强度对比度；(6) Calculate the motion intensity contrast of the macroblock;

(7)计算出宏块的运动方向一致度；(7) Calculate the degree of consistency of the motion direction of the macroblock;

(8)根据所述的亮度对比度、纹理复杂度、运动强度对比度和运动方向一致度，计算出宏块的视觉感知度；(8) Calculate the visual perception of the macroblock according to the brightness contrast, texture complexity, motion intensity contrast, and motion direction consistency;

(9)根据所述的块效应失真度、模糊效应失真度和视觉感知度，计算出待评价视频每帧图像的质量评价值；对待评价视频所有图像的质量评价值求平均，得到的平均值即为待评价视频的质量评价值。(9) Calculate the quality evaluation value of each frame image of the video to be evaluated according to the degree of block effect distortion, blur effect degree of distortion and visual perception; the quality evaluation value of all images of the video to be evaluated is averaged, and the average value obtained is the quality evaluation value of the video to be evaluated.

所述的步骤(2)中，计算宏块的块效应失真度的方法如下：In described step (2), the method for calculating the block effect distortion degree of macroblock is as follows:

a.根据以下算式计算当前宏块的水平块效应失真度：a. Calculate the horizontal blockiness distortion degree of the current macroblock according to the following formula:

S_BH(i)＝|A_(16，i)-B_(1，i)|S _BH(i) = |A _{(16, i)} -B _{(1, i)} |

S_IH(1，i)＝|A_(14，i)-A_(15，i)|S _{IH(1, i)} = |A _{(14, i)} -A _{(15, i)} |

S_IH(2，i)＝|A_(15，i)-A_(16，i)|S _{IH (2, i)} = |A _{(15, i)} -A _{(16, i)} |

S_IH(3，i)＝|B_(1，i)-B_(2，i)|S _{IH (3, i)} = |B _{(1, i)} -B _{(2, i)} |

S_IH(4，i)＝|B_(2，i)-B_(3，i)|S _{IH (4, i)} = |B _{(2, i)} -B _{(3, i)} |

${S S}_{IH IH__AVG AVG ((i i))} = = \frac{{S S}_{IH IH ((11,, i i))} + + {S S}_{IH IH ((22,, i i))} + + {S S}_{IH IH ((33,, i i))} + + {S S}_{IH IH ((44,, i i))}}{44}$

${J J}_{H h ((i i))} = = \{\begin{matrix} {S S}_{BH BH ((i i))} - - {S S}_{IH IH__AVG AVG ((i i))} & if if (({S S}_{BH BH ((i i))} > > {S S}_{IH IH__AVG AVG ((i i))})) \\ 00 & otherwise otherwise \end{matrix}$

${D D.}_{BLOCK BLOCK__H h} = = \frac{11}{1616} {Σ Σ}_{i i = = 11}^{1616} {J J}_{H h ((i i))}$

其中，D_{BLOCK_H}为当前宏块的水平块效应失真度，A(16，i)为当前宏块第16行第i列像素的亮度值，A(15，i)为当前宏块第15行第i列像素的亮度值，A(14，i)为当前宏块第14行第i列像素的亮度值，B(1，i)为与当前宏块下边相邻的宏块第1行第i列像素的亮度值，B(2，i)为与当前宏块下边相邻的宏块第2行第i列像素的亮度值，B(3，i)为与当前宏块下边相邻的宏块第3行第i列像素的亮度值，i为自然数且1≤i≤16；Among them, D _{BLOCK_H} is the horizontal block effect distortion degree of the current macroblock, A(16, i) is the brightness value of the pixel in the 16th row and column i of the current macroblock, and A(15, i) is the 15th row of the current macroblock The brightness value of the i-column pixel, A(14, i) is the brightness value of the i-column pixel in the 14th row of the current macroblock, and B(1, i) is the first row i of the macroblock adjacent to the bottom of the current macroblock The luminance value of the column pixel, B(2, i) is the luminance value of the i-column pixel in the second row of the macroblock adjacent to the bottom of the current macroblock, and B(3, i) is the macroblock adjacent to the bottom of the current macroblock The brightness value of the pixel in the third row and column i of the block, i is a natural number and 1≤i≤16;

b.根据以下算式计算当前宏块的垂直块效应失真度：b. Calculate the vertical blockiness distortion degree of the current macroblock according to the following formula:

S_BV(i)＝|A_(i，16)-C_(i，1)|S _BV(i) = |A _{(i, 16)} -C _{(i, 1)} |

S_IV(i，1)＝|A_(i，14)-A_(i，15)|S _{IV(i, 1)} = |A _{(i, 14)} -A _{(i, 15)} |

S_IV(i，2)＝|A_(i，15)-A_(i，16)|S _{IV(i, 2)} = |A _{(i, 15)} -A _{(i, 16)} |

S_IV(i，3)＝|C_(i，1)-C_(i，2)|S _{IV(i, 3)} = |C _{(i, 1)} -C _{(i, 2)} |

S_IV(i，4)＝|C_(i，2)-C_(i，3)|S _{IV(i, 4)} = |C _{(i, 2)} -C _{(i, 3)} |

${S S}_{IV IV__AVG AVG ((i i))} = = \frac{{S S}_{IV IV ((i i,, 11))} + + {S S}_{IV IV ((i i,, 22))} + + {S S}_{IV IV ((i i,, 33))} + + {S S}_{IV IV ((i i,, 44))}}{44}$

${J J}_{V V ((i i))} = = \{\begin{matrix} {S S}_{BV BV ((i i))} - - {S S}_{IV IV__AVG AVG ((i i))} & if if (({S S}_{BV BV ((i i))} > > {S S}_{IV IV__AVG AVG ((i i))})) \\ 00 & otherwise otherwise \end{matrix}$

${D D.}_{BLOCK BLOCK__V V} = = \frac{11}{1616} {Σ Σ}_{i i = = 11}^{1616} {J J}_{V V ((i i))}$

其中，D_{BLOCK_v}为当前宏块的垂直块效应失真度，A(i，16)为当前宏块第i行第16列像素的亮度值，A(i，15)为当前宏块第i行第15列像素的亮度值，A(i，14)为当前宏块第i行第14列像素的亮度值，C(i，1)为与当前宏块右边相邻的宏块第i行第1列像素的亮度值，C(i，2)为与当前宏块右边相邻的宏块第i行第2列像素的亮度值，C(i，3)为与当前宏块右边相邻的宏块第i行第3列像素的亮度值；Among them, D _{BLOCK_v} is the vertical block effect distortion degree of the current macroblock, A(i, 16) is the brightness value of the pixel in the 16th column of the i-th row of the current macroblock, and A(i, 15) is the i-th row of the current macroblock The luminance value of the 15th column pixel, A(i, 14) is the luminance value of the pixel in the 14th column of the i-th row of the current macroblock, and C(i, 1) is the 1st row of the i-th row of the macroblock adjacent to the right of the current macroblock The brightness value of the pixel in the column, C(i, 2) is the brightness value of the second column pixel in row i of the macroblock adjacent to the right of the current macroblock, and C(i, 3) is the macroblock adjacent to the right of the current macroblock. The brightness value of the pixel in the third column of the ith row of the block;

c.对所述的水平块效应失真度和垂直块效应失真度求平均，得到当前宏块的块效应失真度。c. Calculate the average of the horizontal block-effect distortion degree and the vertical block-effect distortion degree to obtain the block-effect distortion degree of the current macroblock.

所述的步骤(3)中，计算宏块的模糊效应失真度的方法如下：In described step (3), the method for calculating the blur effect degree of distortion of macroblock is as follows:

a.利用Sobel算子对宏块进行边缘检测，确定宏块中的边缘像素及其梯度方向；a. Use the Sobel operator to perform edge detection on the macroblock, and determine the edge pixels and their gradient directions in the macroblock;

b.将边缘像素的梯度方向根据就近原则归类于{0°、45°、90°、135°、180°、225°、270°、315°}这八种方向当中的一种，并作为边缘像素的近似梯度方向；b. Classify the gradient direction of the edge pixels into one of the eight directions {0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°} according to the nearest principle, and use it as Approximate gradient directions for edge pixels;

c.根据以下算式计算边缘像素的边缘锐利度：c. Calculate the edge sharpness of the edge pixels according to the following formula:

$S S = = \frac{11}{44} {Σ Σ}_{l l = = 11}^{44} \frac{| | {I I}_{((θ θ,, l l))} - - I I | |}{l l}$

其中：S为边缘像素的边缘锐利度，I为边缘像素的亮度值，I(θ，l)为与边缘像素沿近似梯度方向θ距离为l的像素的亮度值，l为自然数且1≤l≤4；Among them: S is the edge sharpness of the edge pixel, I is the brightness value of the edge pixel, I(θ, l) is the brightness value of the pixel whose distance from the edge pixel is l along the approximate gradient direction θ, l is a natural number and 1≤l ≤4;

d.对宏块中所有边缘像素的边缘锐利度求平均，得到宏块的模糊效应失真度。d. Average the edge sharpness of all edge pixels in the macroblock to obtain the blur effect distortion degree of the macroblock.

所述的步骤(4)中，计算宏块的亮度对比度的方法为：根据以下算式计算宏块中每个像素的亮度对比度，取宏块中所有像素亮度对比度的最大值作为宏块的亮度对比度；In the described step (4), the method for calculating the brightness contrast of the macroblock is: calculate the brightness contrast of each pixel in the macroblock according to the following formula, get the maximum value of the brightness contrast of all pixels in the macroblock as the brightness contrast of the macroblock ;

${C C}_{Luma Luma__P P} ((θ θ,, l l)) = = \frac{| | {I I}_{((θ θ,, l l))} - - I I | |}{{l l}^{η η}}$

${C C}_{Luma Luma} = = {Σ Σ}_{θ θ &Element; &Element; M m}^{M m} {Σ Σ}_{l l = = 11}^{L L} {C C}_{Luma Luma__P P} ((l l,, N N))$

其中，C_Luma为当前像素的亮度对比度，I为当前像素的亮度值，I(θ，l)为与当前像素沿方向θ距离为l的像素的亮度值，l为自然数且1≤l≤L，L为预设的最大距离，η为像素距离衰退系数，M为0°、90°、180°和270°这四种方向的集合。Among them, C _Luma is the brightness contrast of the current pixel, I is the brightness value of the current pixel, I(θ, l) is the brightness value of the pixel whose distance from the current pixel is l along the direction θ, l is a natural number and 1≤l≤L , L is the preset maximum distance, η is the pixel distance attenuation coefficient, M is a set of four directions of 0°, 90°, 180° and 270°.

所述的步骤(5)中，计算宏块的纹理复杂度的方法如下：In the described step (5), the method for calculating the texture complexity of the macroblock is as follows:

a.利用Sobel算子对宏块进行边缘检测，确定宏块中边缘像素的总个数以及每个像素的梯度方向；a. utilize Sobel operator to carry out edge detection to macroblock, determine the total number of edge pixels in macroblock and the gradient direction of each pixel;

b.将像素的梯度方向根据就近原则归类于{0°、45°、90°、135°、180°、225°、270°、315°}这八种方向当中的一种，并作为像素的近似梯度方向；b. Classify the gradient direction of the pixel into one of the eight directions {0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°} according to the principle of proximity, and use it as a pixel The approximate gradient direction of ;

c.将像素的近似梯度方向分成四类：0°和180°归为一类、45°和225°归为一类、90°和270°归为一类、135°和315°归为一类；c. Divide the approximate gradient direction of the pixel into four categories: 0° and 180° are classified into one category, 45° and 225° are classified into one category, 90° and 270° are classified into one category, and 135° and 315° are classified into one category kind;

d.根据以下算式计算宏块的纹理复杂度：d. Calculate the texture complexity of the macroblock according to the following formula:

${T T}_{block block} = = \{\begin{matrix} 0.5 0.5 & if if (({k k}_{θ θ} = = 11)) \\ 11 & if if (({k k}_{θ θ} = = 22)) \\ ((22 - - {c c}_{e e})) / / 22 & if if (({k k}_{θ θ} = = 33)) \\ ((11 - - {c c}_{e e})) / / 22 & if if (({k k}_{θ θ} = = 44)) \end{matrix}$

${c c}_{e e} = = \{\begin{matrix} 11 & if if (({n no}_{edge edge} > > {T T}_{edge edge})) \\ 00 & otherwise otherwise \end{matrix}$

其中，T_block为宏块的纹理复杂度，k_θ为宏块中所有像素近似梯度方向的种类数，n_edge为宏块中边缘像素的总个数，T_edge为给定的边缘像素个数阈值。Among them, T _block is the texture complexity of the macro block, k _θ is the number of types of approximate gradient directions of all pixels in the macro block, n _edge is the total number of edge pixels in the macro block, and T _edge is the given number of edge pixels threshold.

所述的步骤(6)中，计算宏块的运动强度对比度的方法如下：In the described step (6), the method for calculating the motion intensity contrast of the macroblock is as follows:

a.通过帧间预测计算出宏块的水平运动矢量和垂直运动矢量；对所述的水平运动矢量和垂直运动矢量均方根得到宏块的运动强度；a. Calculate the horizontal motion vector and the vertical motion vector of the macroblock by inter-frame prediction; obtain the motion intensity of the macroblock to the root mean square of the horizontal motion vector and the vertical motion vector;

b.以当前宏块为中心，建立由7×7个宏块组成的参考窗口；b. Taking the current macroblock as the center, establish a reference window composed of 7×7 macroblocks;

c.根据以下算式计算当前宏块与参考窗口中每个宏块的运动强度差异距离比M_{I_diff}，并取其中最大值为M_{Id_max}：c. Calculate the motion intensity difference distance ratio _{MI_diff} between the current macroblock and each macroblock in the reference window according to the following formula, and take the maximum value as _{MId_max} :

${M m}_{I I__diff diff} = = \frac{| | {M m}_{I I} - - {M m}_{I I} ((D D.)) | |}{{d d}^{δ δ}}$

其中：M_I为当前宏块的运动强度，M_I(D)为宏块D的运动强度，宏块D为参考窗口中的任一宏块，d为当前宏块与宏块D的距离，δ为宏块距离衰退系数；Wherein: _MI is the motion intensity of the current macroblock, _MI (D) is the motion intensity of the macroblock D, the macroblock D is any macroblock in the reference window, and d is the distance between the current macroblock and the macroblock D, δ is the macroblock distance decay coefficient;

d.根据以下关系式确定当前宏块的运动强度对比度：d. Determine the motion intensity contrast of the current macroblock according to the following relationship:

${C C}_{Motion motion__block block} = = \{\begin{matrix} \frac{{M m}_{Id ID__max max}}{{M m}_{I I__max max}} & if if (({M m}_{I I__max max} &NotEqual; &NotEqual; 00)) \\ 00 & otherwise otherwise \end{matrix}$

其中：C_{Motion_block}为当前宏块的运动强度对比度，M_{I_max}为参考窗口中所有宏块运动强度的最大值。Where: C _{Motion_block} is the motion intensity contrast of the current macroblock, M _{I_max} is the maximum value of the motion intensity of all macroblocks in the reference window.

所述的步骤(7)中，计算宏块的运动方向一致度的方法如下：In the described step (7), the method for calculating the degree of motion direction consistency of the macroblock is as follows:

a.通过帧间预测计算出宏块的水平运动矢量mv_x和垂直运动矢量mv_y；对所述的水平运动矢量和垂直运动矢量均方根得到宏块的运动强度，并根据公式θ_mv＝arctan(mv_y/mv_x)算出宏块的运动方向θ_mv；a. Calculate the horizontal motion vector mv _x and the vertical motion vector mv _y of the macroblock by inter-frame prediction; obtain the motion intensity of the macroblock according to the root mean square of the horizontal motion vector and the vertical motion vector, and according to the formula θ _mv = arctan(mv _y /mv _x ) calculates the motion direction θ _mv of the macroblock;

b.将0°到360°的圆周等分成12块扇区，使宏块的运动方向映射至对应的扇区；b. Divide the circle from 0° to 360° into 12 sectors, so that the motion direction of the macroblock is mapped to the corresponding sectors;

c.以当前宏块为中心，建立由21×21个宏块组成的参考窗口；c. With the current macroblock as the center, establish a reference window composed of 21×21 macroblocks;

d.根据以下算式计算当前宏块的运动方向一致度：d. Calculate the motion direction consistency degree of the current macroblock according to the following formula:

${M m}_{Con Con__block block} = = - - {Σ Σ}_{j j = = 11}^{1212} p p ((j j)) log log [[p p ((j j))]] \frac{{M m}_{I I__avg avg} ((j j))}{{M m}_{I I__max max}}$ $p p ((j j)) = = \frac{k k ((j j))}{{21 twenty one}^{22}}$

其中：M_{Con_block}为当前宏块的运动方向一致度，k(j)为参考窗口中归属于第j扇区的宏块总个数，M_{I_max}为参考窗口中所有宏块运动强度的最大值，M_{I_avg}(j)为参考窗口中归属于第j扇区的所有宏块运动强度的平均值，j为自然数且1≤j≤12。Among them: M _{Con_block} is the motion direction consistency of the current macroblock, k(j) is the total number of macroblocks belonging to the jth sector in the reference window, M _{I_max} is the maximum value of the motion intensity of all macroblocks in the reference window, M _{I_avg} (j) is the average value of the motion intensity of all macroblocks belonging to the jth sector in the reference window, j is a natural number and 1≤j≤12.

所述的步骤(8)中，根据以下算式计算宏块的视觉感知度：In described step (8), calculate the visual perception of macroblock according to following formula:

VP_S＝log(α₁+C_{Luma_block})×(α₂+T_block)² VP _S ＝log(α ₁ +C _{Luma_block} )×(α ₂ +T _block ) ²

VP_T＝α₃C_{Motion_block}+α₄M_{con_block} VP _T = α ₃ C _{Motion_block} + α ₄ M _{con_block}

VM＝λ×VP_S×VP_T VM＝λ×VP _S ×VP _T

其中：VM为宏块的视觉感知度，M_{Con_block}为宏块的运动方向一致度，C_{Motion_block}为宏块的运动强度对比度，T_block为宏块的纹理复杂度，C_{Luma_block}为宏块的亮度对比度，λ、α₁、α₂、α₃和α₄均为给定的权重系数。Among them: VM is the visual perception of the macro block, M _{Con_block} is the consistency of the motion direction of the macro block, C _{Motion_block} is the motion intensity contrast of the macro block, T _block is the texture complexity of the macro block, and C _{Luma_block} is the brightness contrast of the macro block , λ, α ₁ , α ₂ , α ₃ and α ₄ are given weight coefficients.

所述的步骤(9)中，根据以下算式计算待评价视频每帧图像的质量评价值：In described step (9), calculate the quality evaluation value of every frame image of video to be evaluated according to following formula:

$Q Q = = \frac{{Σ Σ}_{n no = = 11}^{N N} [[{Q Q}_{MB MB} ((n no)) \times \times VM VM ((n no))]]}{{Σ Σ}_{n no = = 11}^{N N} VM VM ((n no))}$

Q_MB(n)＝γ₁×D_{BLOCK_MB}(n)+γ₂×D_{BLUR_MB}(n)Q _MB (n)＝γ ₁ ×D _{BLOCK_MB} (n)+γ ₂ ×D _{BLUR_MB} (n)

其中：Q为待评价视频任一帧图像的质量评价值，D_{BLOCK_MB}(n)为待评价视频任一帧图像中第n宏块的块效应失真度，D_{BLUR_MB}(n)为待评价视频任一帧图像中第n宏块的模糊效应失真度，VM(n)为待评价视频任一帧图像中第n宏块的视觉感知度，γ₁和γ₂均为给定的权重系数，n为自然数且1≤n≤N，N为待评价视频任一帧图像中宏块的总个数。Among them: Q is the quality evaluation value of any frame image of the video to be evaluated, D _{BLOCK_MB} (n) is the block effect distortion degree of the nth macroblock in any frame image of the video to be evaluated, D _{BLUR_MB} (n) is the value of any frame of the video to be evaluated The blur effect distortion degree of the nth macroblock in a frame image, VM(n) is the visual perception degree of the nth macroblock in any frame image of the video to be evaluated, γ ₁ and γ ₂ are given weight coefficients, n is a natural number and 1≤n≤N, and N is the total number of macroblocks in any frame of video to be evaluated.

本发明的有益技术效果如下：Beneficial technical effect of the present invention is as follows:

(1)本发明只需要待评价视频就可以得到视频质量评价结果，无需参考信息，具有很好的灵活性和适应性。(1) The present invention only needs the video to be evaluated to obtain the video quality evaluation result without reference information, and has good flexibility and adaptability.

(2)本发明评价结果准确度较高，符合人眼对视频的主观感知。(2) The accuracy of the evaluation result of the present invention is relatively high, which conforms to the subjective perception of the video by the human eye.

(3)本发明对各种不同的视频场景都能得到比较准确的评价结果，具有较好的普适性。(3) The present invention can obtain relatively accurate evaluation results for various video scenes, and has good universality.

附图说明 Description of drawings

图1为本发明方法的流程示意图。Fig. 1 is a schematic flow chart of the method of the present invention.

图2为宏块边界像素的位置示意图。FIG. 2 is a schematic diagram of the positions of macroblock boundary pixels.

图3为运动方向的分类示意图。Fig. 3 is a schematic diagram of classification of motion directions.

具体实施方式 Detailed ways

为了更为具体地描述本发明，下面结合附图及具体实施方式对本发明的技术方案进行详细说明。In order to describe the present invention more specifically, the technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

我们首先将待评价视频每帧图像分割成若干个宏块；如图1所示，一种基于宏块域失真度估计的视频质量评价方法，包括如下步骤：We first divide each frame of the video to be evaluated into several macroblocks; as shown in Figure 1, a video quality evaluation method based on macroblock domain distortion estimation includes the following steps:

(1)计算块效应失真度。(1) Calculate block effect distortion.

块效应一般出现在基于离散余弦变换(DCT)的压缩算法中，粗糙的量化过程也会加剧块效应。不同宏块受DCT和量化的影响不一样，细节损失也有差异，因此，在宏块边界往往会有明显的不连续，形成块效应失真。本实施方式分别从水平和垂直方向计算宏块边缘的块效应失真程度，具体过程如下：Blocking artifacts generally appear in discrete cosine transform (DCT)-based compression algorithms, and rough quantization processes can also aggravate blocking artifacts. Different macroblocks are affected by DCT and quantization differently, and the loss of details is also different. Therefore, there are often obvious discontinuities at the boundaries of macroblocks, resulting in block effect distortion. In this embodiment, the degree of blockiness distortion at the edge of the macroblock is calculated from the horizontal and vertical directions respectively, and the specific process is as follows:

a.根据以下算式计算当前宏块的水平块效应失真度；如图2所示，黑色实线上方和下方的方块分别代表当前宏块A和其下边相邻的宏块B。a. Calculate the horizontal blockiness distortion degree of the current macroblock according to the following formula; as shown in FIG. 2 , the squares above and below the solid black line represent the current macroblock A and the adjacent macroblock B below it respectively.

S_BH(i)＝|A_(16，i)-B_(1，i)|S _BH(i) = |A _{(16, i)} -B _{(1, i)} |

S_IH(3，i)＝|B_(1，i)-B_(2，i)|S _{IH (3, i)} = |B _{(1, i)} -B _{(2, i)} |

S_IH(4，i)＝|B_(2，i)-B_(3，i)|S _{IH (4, i)} = |B _{(2, i)} -B _{(3, i)} |

S_BV(i)＝|A_(i，16)-C_(i，1)|S _BV(i) = |A _{(i, 16)} -C _{(i, 1)} |

S_IV(i，3)＝|C_(i，1)-C_(i，2)|S _{IV(i, 3)} = |C _{(i, 1)} -C _{(i, 2)} |

S_IV(i，4)＝|C_(i，2)-C_(i，3)|S _{IV(i, 4)} = |C _{(i, 2)} -C _{(i, 3)} |

c.对水平块效应失真度和垂直块效应失真度求平均，得到当前宏块的块效应失真度D_{BLOCK_MB}。c. Calculate the average of the horizontal block-effect distortion degree and the vertical block-effect distortion degree to obtain the block-effect distortion degree D _{BLOCK_MB} of the current macroblock.

由于量化过程是以宏块为单位进行的，相邻的宏块可以采用不同的量化参数。因此，宏块边缘像素间的差异往往比内部像素间的差异更大，而随着差异的增大，宏块边缘之间的不连续跳变就越明显。宏块的块效应失真度D_{BLOCK_MB}即为水平和垂直失真度的平均值；D_{BLOCK_MB}越大，代表块效应越严重。Since the quantization process is performed in units of macroblocks, adjacent macroblocks may use different quantization parameters. Therefore, the difference between pixels at the edge of a macroblock is often greater than the difference between pixels inside the macroblock, and as the difference increases, the discontinuous jump between the edges of the macroblock becomes more obvious. The block effect distortion D _{BLOCK_MB} of the macroblock is the average value of the horizontal and vertical distortion; the larger D _{BLOCK_MB} is, the more serious the block effect is.

(2)计算模糊效应失真度。(2) Calculate the degree of blur effect distortion.

模糊效应是一种常见的视频失真效应，从主观视觉上看，模糊效应主要体现在视频中物体边缘附近细节退化，锐利程度下降。物体边缘上的像素点，如果受到模糊效应的影响，一般都会有最大的局部梯度值。本实施方式用Sobel算子对视频进行边缘检测，对每一个位于物体边缘的像素点，检测其梯度方向的边缘锐利度S，以此来评估视频受到模糊效应影响的失真程度；具体过程如下：The blur effect is a common video distortion effect. From the subjective perspective, the blur effect is mainly reflected in the degradation of details near the edge of the object in the video, and the sharpness decreases. Pixels on the edge of the object, if affected by the blur effect, generally have the largest local gradient value. This embodiment uses the Sobel operator to perform edge detection on the video, and detects the edge sharpness S in the gradient direction for each pixel point located at the edge of the object, so as to evaluate the degree of distortion of the video affected by the blur effect; the specific process is as follows:

b.将边缘像素的梯度方向根据就近原则归类于{0°、45°、90°、135°、180°、225°、270°、315°}这八种方向当中的一种，并作为边缘像素的近似梯度方向；b. Classify the gradient direction of edge pixels into one of the eight directions {0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°} according to the nearest principle, and use it as Approximate gradient directions for edge pixels;

d.对宏块中所有边缘像素的边缘锐利度求平均，得到宏块的模糊效应失真度D_{BLUR_MB}。d. Average the edge sharpness of all edge pixels in the macroblock to obtain the blur effect distortion D _{BLUR_MB} of the macroblock.

本实施方式在梯度方向上选取四个点，用它们的平均渐变速度来计算S，S越小，代表边缘锐利度越小，模糊效应越严重；最后，统计整个宏块中每个边缘像素点对应的S值，取其平均值作为该宏块的模糊效应失真度D_{BLUR_MB}，D_{BLUR_MB}越小，代表模糊效应越严重。In this embodiment, four points are selected in the gradient direction, and their average gradient speed is used to calculate S. The smaller the S, the smaller the edge sharpness and the more serious the blur effect; finally, count each edge pixel in the entire macroblock The average value of the corresponding S value is taken as the blur distortion degree D _{BLUR_MB} of the macroblock, and the smaller D _{BLUR_MB} means the more serious the blur effect.

(3)计算亮度对比度。(3) Calculate brightness contrast.

对HVS特性研究表明，人类在观看视频时，一般对视频内物体的绝对亮度并不敏感，而对局部区域与周围区域的亮度的相对大小更加敏感，因此，亮度对比度可以简单描述为像素间亮度值之差。研究同时表明，人眼对亮度差值的敏感度随着像素间的距离增大而减小。结合以上两点，本实施方式用像素亮度差值与距离来定义两个像素间的亮度对比度，具体方法如下：The study of HVS characteristics shows that when humans watch a video, they are generally not sensitive to the absolute brightness of objects in the video, but are more sensitive to the relative brightness of the local area and the surrounding area. Therefore, the brightness contrast can be simply described as the brightness between pixels value difference. Research also shows that the sensitivity of the human eye to brightness differences decreases as the distance between pixels increases. Combining the above two points, this embodiment uses the pixel brightness difference and the distance to define the brightness contrast between two pixels. The specific method is as follows:

根据以下算式计算宏块中每个像素的亮度对比度，取宏块中所有像素亮度对比度的最大值作为宏块的亮度对比度C_{Luma_block}；Calculate the brightness contrast of each pixel in the macroblock according to the following formula, and take the maximum value of the brightness contrast of all pixels in the macroblock as the brightness contrast C _{Luma_block} of the macroblock;

其中，C_Luma为当前像素的亮度对比度，I为当前像素的亮度值，I(θ，l)为与当前像素沿方向θ距离为l的像素的亮度值，l为自然数且1≤l≤L，本实施方式预设L＝5，η为像素距离衰退系数(本实施方式中该系数取2)，M为0°、90°、180°和270°这四种方向的集合。Among them, C _Luma is the brightness contrast of the current pixel, I is the brightness value of the current pixel, I(θ, l) is the brightness value of the pixel whose distance from the current pixel is l along the direction θ, l is a natural number and 1≤l≤L , this embodiment presets L=5, n is the pixel distance attenuation coefficient (the coefficient is 2 in this embodiment), and M is a set of four directions of 0°, 90°, 180° and 270°.

亮度对比度具有一定的区域性，人眼对局部区域内对比度最大的某些点最为敏感。因此，本实施方式将整帧视频图像划分为16×16像素的互不重叠的宏块，以每个块中的像素点C_Luma最大值作为该宏块的亮度对比度C_{Luma_block}。C_{Luma_block}的值越大表示该区域的亮度对比度越高，即该区域与周围区域之间的亮度差值越大，越容易引起人眼的注意。The brightness contrast has a certain regionality, and the human eye is most sensitive to certain points with the highest contrast in a local area. Therefore, in this embodiment, the entire frame of video image is divided into non-overlapping macroblocks of 16×16 pixels, and the maximum value of pixel C _Luma in each block is used as the brightness contrast C _{Luma_block} of the macroblock. The larger the value of C _{Luma_block} , the higher the brightness contrast of this area, that is, the larger the brightness difference between this area and the surrounding area, the easier it is to attract the attention of human eyes.

(4)计算纹理复杂度。(4) Calculate texture complexity.

按照纹理的不同特性，可以将视频图像分为结构纹理区域和随机纹理区域。结构纹理区域的纹理较为简单，与周围图像关联性较低，而随机纹理区域的纹理较为丰富，空间对比度低，与周围图像关联性较高。HVS特性研究表明，结构纹理区域内的图像失真更容易吸引人类的注意。According to the different characteristics of the texture, the video image can be divided into structural texture area and random texture area. Structural texture regions have simpler textures and less correlation with surrounding images, while random texture regions have richer textures, low spatial contrast, and high correlation with surrounding images. The study of HVS properties shows that image distortions in structurally textured regions are more likely to attract human attention.

本实施方式首先用Sobel算子对整幅图像进行卷积，提取边缘像素点，根据以下算式分别计算每个像素点的水平和垂直梯度(G_hor，G_ver)，进而确定宏块中边缘像素的总个数n_edge以及每个像素的梯度方向θ(i，j)；In this embodiment, first, the Sobel operator is used to convolute the entire image to extract edge pixels, and the horizontal and vertical gradients (G _hor , G _ver ) of each pixel are respectively calculated according to the following formula, and then the edge pixels in the macroblock are determined The total number n _edge and the gradient direction θ(i, j) of each pixel;

G_hor(i，j)＝I(i，j)*S_hor G _hor (i,j)=I(i,j)*S _hor

G_ver(i，j)＝I(i，j)*S_ver G _ver (i, j)=I(i, j)*S _ver

$θ θ ((i i,, j j)) = = arctan arctan \frac{{G G}_{ver ver} ((i i,, j j))}{{G G}_{hor hor} ((i i,, j j))}$

然后，将像素的梯度方向根据就近原则归类于{0°、45°、90°、135°、180°、225°、270°、315°}这八种方向当中的一种，并作为像素的近似梯度方向；Then, the gradient direction of the pixel is classified into one of the eight directions {0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°} according to the nearest principle, and used as the pixel The approximate gradient direction of ;

将像素的近似梯度方向分成四类：0°和180°归为一类、45°和225°归为一类、90°和270°归为一类、135°和315°归为一类；The approximate gradient direction of the pixel is divided into four categories: 0° and 180° are classified into one category, 45° and 225° are classified into one category, 90° and 270° are classified into one category, and 135° and 315° are classified into one category;

最后，根据以下算式计算宏块的纹理复杂度：Finally, the texture complexity of the macroblock is calculated according to the following formula:

其中，T_block为宏块的纹理复杂度，k_θ为宏块中所有像素近似梯度方向的种类数，n_edge为宏块中边缘像素的总个数，T_edge为边缘像素个数阈值(本实施方式中该阈值取16)。Among them, T _block is the texture complexity of the macro block, k _θ is the number of types of approximate gradient directions of all pixels in the macro block, n _edge is the total number of edge pixels in the macro block, and T _edge is the threshold of the number of edge pixels (this In the embodiment, the threshold is set to 16).

T_block的取值范围为[0，1]，当T_block趋向0时，代表该区域纹理丰富，属于随机纹理区域，其失真效应不容易被人眼察觉；当T_block趋向1时，代表该区域纹理简单，属于随机纹理区域，人眼对其中的失真效应比较敏感。The value range of T _block is [0, 1]. When T _block tends to 0, it means that the area is rich in texture and belongs to random texture area, and its distortion effect is not easy to be noticed by human eyes; when T _block tends to 1, it means that the area The area texture is simple and belongs to the random texture area, and the human eye is more sensitive to the distortion effect in it.

(5)计算运动强度对比度。(5) Calculate the exercise intensity contrast.

在视频质量评价领域，运动信息作为最重要的初级视觉信息之一，被广泛的用于反映HVS对视觉信号的感知。运动强度对比度和运动方向一致性是最能体现视频运动特性的特征值。HVS研究表明，人眼对物体相对运动的敏感度要高于单个物体的绝对运动，而运动强度对比度正好反映了视频内不同物体的运动速度差异，反映了HVS对视频的整体感知。本实施方式计算宏块的运动强度对比度的过程如下：In the field of video quality evaluation, motion information, as one of the most important primary visual information, is widely used to reflect the perception of visual signals by HVS. The motion intensity contrast and motion direction consistency are the characteristic values that can best reflect the video motion characteristics. HVS research shows that the human eye is more sensitive to the relative motion of an object than the absolute motion of a single object, and the contrast of motion intensity just reflects the difference in motion speed of different objects in the video, reflecting the overall perception of the video by HVS. In this embodiment, the process of calculating the motion intensity contrast of a macroblock is as follows:

a.通过帧间预测计算出宏块的水平运动矢量和垂直运动矢量；对水平运动矢量和垂直运动矢量均方根得到宏块的运动强度；a. Calculate the horizontal motion vector and the vertical motion vector of the macroblock by inter-frame prediction; obtain the motion intensity of the macroblock to the root mean square of the horizontal motion vector and the vertical motion vector;

c.根据以下算式计算当前宏块与参考窗口中每个宏块的运动强度差异距离比MI_diff，并取其中最大值为M_{Id_max}：c. Calculate the motion intensity difference distance ratio MI_diff between the current macroblock and each macroblock in the reference window according to the following formula, and take the maximum value as M _{Id_max} :

其中：M_I为当前宏块的运动强度，M_I(D)为宏块D的运动强度，宏块D为参考窗口中的任一宏块，d为当前宏块与宏块D的距离，δ为宏块距离衰退系数(本实施方式中该系数取2)；Wherein: _MI is the motion intensity of the current macroblock, _MI (D) is the motion intensity of the macroblock D, the macroblock D is any macroblock in the reference window, and d is the distance between the current macroblock and the macroblock D, δ is the macroblock distance decay coefficient (this coefficient gets 2 in the present embodiment);

考虑到局部区域里的运动强度最大值也会影响到HVS对运动强度差异的感知，最后根据以下关系式确定当前宏块的运动强度对比度：Considering that the maximum value of the motion intensity in the local area will also affect the perception of the difference in motion intensity by the HVS, finally determine the motion intensity contrast of the current macroblock according to the following relationship:

其中：C_{Motion_block}为当前宏块的运动强度对比度，M_{I_max}为参考窗口中所有宏块运动强度的最大值。C_{Motion_block}值越大，代表宏块的运动强度对比度越大。Where: C _{Motion_block} is the motion intensity contrast of the current macroblock, M _{I_max} is the maximum value of the motion intensity of all macroblocks in the reference window. The larger the C _{Motion_block} value, the larger the motion intensity contrast of the macroblock.

(6)计算运动方向一致度。(6) Calculate the consistency of the motion direction.

在视频场景中，各物体的运动行为是多种多样的，运动一致性反映了这些运动行为的相似性，其中又以运动方向最容易引起HVS关注。在本实施方式中，通过分析局部区域内各种运动方向的概率分布，结合运动强度，来表示运动方向一致性。本实施方式计算宏块的运动方向一致度的方法如下：In a video scene, the motion behavior of each object is diverse, and the motion consistency reflects the similarity of these motion behaviors, among which the motion direction is the most likely to attract the attention of HVS. In this embodiment, the consistency of the motion direction is represented by analyzing the probability distribution of various motion directions in the local area and combining the motion intensity. In this embodiment, the method for calculating the consistency degree of motion direction of a macroblock is as follows:

a.通过帧间预测计算出宏块的水平运动矢量mv_x和垂直运动矢量mv_y；对水平运动矢量mv_x和垂直运动矢量mv_y均方根得到宏块的运动强度，并根据公式θ_mv＝arctan(mv_y/mv_x)算出宏块的运动方向θ_mv；a. Calculate the horizontal motion vector mv _x and vertical motion vector mv _y of the macroblock through inter-frame prediction; get the motion intensity of the macroblock by the root mean square of the horizontal motion vector mv _x and vertical motion vector mv _y , and according to the formula θ _mv =arctan(mv _y /mv _x ) calculates the motion direction θ _mv of the macroblock;

b.将0°到360°的圆周等分成12块扇区，使宏块的运动方向映射至对应的扇区；如图3所示，将运动方向大致分为12种，即将[0，360°)等分为12种不同的方向Angle区域，采用直方图方法把每个宏块的运动方向θ_mv映射到对应的Angle区域；b. Divide the circle from 0° to 360° into 12 sectors, so that the motion direction of the macroblock is mapped to the corresponding sector; as shown in Figure 3, the motion direction is roughly divided into 12 types, namely [0, 360 °) is equally divided into 12 different direction Angle regions, and the motion direction θ _mv of each macroblock is mapped to the corresponding Angle region by using the histogram method;

其中：M_{Con_block}为当前宏块的运动方向一致度，k(j)为参考窗口中归属于第j扇区的宏块总个数，M_{I_max}为参考窗口中所有宏块运动强度的最大值，M_{I_avg}(j)为参考窗口中归属于第j扇区的所有宏块运动强度的平均值，j为自然数且1≤j≤12；M_{Con_block}值越大，代表宏块的运动一致性越高。Among them: M _{Con_block} is the motion direction consistency of the current macroblock, k(j) is the total number of macroblocks belonging to the jth sector in the reference window, M _{I_max} is the maximum value of the motion intensity of all macroblocks in the reference window, M _{I_avg} (j) is the average value of the motion intensity of all macroblocks belonging to the jth sector in the reference window, j is a natural number and 1≤j≤12; the larger the value of M _{Con_block} , the higher the motion consistency of the macroblock .

(7)计算视觉感知度。(7) Calculate visual perception.

根据亮度对比度、纹理复杂度、运动强度对比度和运动方向一致度，根据以下算式计算待评价视频每帧图像每个宏块的视觉感知度：According to the brightness contrast, texture complexity, motion intensity contrast, and motion direction consistency, the visual perception of each macroblock of each frame of the video to be evaluated is calculated according to the following formula:

VM＝λ×VP_S×VP_T VM＝λ×VP _S ×VP _T

其中：VM为宏块的视觉感知度，M_{Con_block}为宏块的运动方向一致度，C_{Motion_block}为宏块的运动强度对比度，T_block为宏块的纹理复杂度，C_{Luma_block}为宏块的亮度对比度，λ、α₁、α₂、α₃和α₄均为权重系数，本实施方式中λ取1.2，α₁、α₂和α₄均取1，α₃取2。Among them: VM is the visual perception of the macro block, M _{Con_block} is the consistency of the motion direction of the macro block, C _{Motion_block} is the motion intensity contrast of the macro block, T _block is the texture complexity of the macro block, and C _{Luma_block} is the brightness contrast of the macro block , λ, α ₁ , α ₂ , α ₃ and α ₄ are all weight coefficients. In this embodiment, λ is 1.2, α ₁ , α ₂ and α ₄ are all 1, and α ₃ is 2.

(8)计算视频的质量评价值。(8) Calculate the quality evaluation value of the video.

根据块效应失真度、模糊效应失真度和视觉感知度，根据以下算式计算待评价视频每帧图像的质量评价值；According to the degree of block effect distortion, blur effect distortion and visual perception, the quality evaluation value of each frame of the video to be evaluated is calculated according to the following formula;

其中：Q为待评价视频任一帧图像的质量评价值，D_{BLOCK_MB}(n)为待评价视频任一帧图像中第n宏块的块效应失真度，D_{BLUR_MB}(n)为待评价视频任一帧图像中第n宏块的模糊效应失真度，VM(n)为待评价视频任一帧图像中第n宏块的视觉感知度，γ₁和γ₂均为权重系数；n为自然数且1≤n≤N，N为待评价视频任一帧图像中宏块的总个数。Q_MB越大代表该块失真越严重，视频质量越差。本实施方式引入基于视觉感知的视觉注意模型，用该模型来调整视频各个区域在质量评价体系中的权重，本实施方式中γ₁取1，γ₂取-0.5。Among them: Q is the quality evaluation value of any frame image of the video to be evaluated, D _{BLOCK_MB} (n) is the block effect distortion degree of the nth macroblock in any frame image of the video to be evaluated, D _{BLUR_MB} (n) is the value of any frame of the video to be evaluated The blur effect distortion degree of the nth macroblock in a frame image, VM(n) is the visual perception degree of the nth macroblock in any frame image of the video to be evaluated, γ ₁ and γ ₂ are weight coefficients; n is a natural number and 1≤n≤N, N is the total number of macroblocks in any frame of video to be evaluated. The larger the Q _MB , the more serious the distortion of the block and the worse the video quality. This embodiment introduces a visual attention model based on visual perception, and uses this model to adjust the weight of each region of the video in the quality evaluation system. In this embodiment, _γ1 is 1, and _γ2 is -0.5.

最后，对待评价视频所有图像的质量评价值求平均，得到的平均值Q_AVG即为待评价视频的质量评价值；Q_AVG的取值范围为[0，100]，0代表视频质量最好，100代表视频质量最差。Finally, the quality evaluation values of all images of the video to be evaluated are averaged, and the obtained average value _QAVG is the quality evaluation value of the video to be evaluated; the value range of _QAVG is [0, 100], and 0 represents the best video quality, 100 represents the worst video quality.

以下我们通过100个JM压缩视频来验证算法的效果，分别用Spearman相关系数和Pearson相关系数来衡量本实施方式的单调性和准确性，并与传统基于PSNR和基于SSIM方法作对比(Spearman系数和Pearson系数越大，说明方法的单调性和准确性越好，即方法的准确度越高)，如表1所示：Below we verify the effect of the algorithm through 100 JM compressed videos, use the Spearman correlation coefficient and the Pearson correlation coefficient to measure the monotonicity and accuracy of this embodiment, and compare it with the traditional PSNR-based and SSIM-based methods (Spearman coefficient and The larger the Pearson coefficient, the better the monotonicity and accuracy of the method, that is, the higher the accuracy of the method), as shown in Table 1:

表1Table 1

质量评价方法 Quality Evaluation Method Spearman相关系数 Spearman correlation coefficient Pearson相关系数 Pearson correlation coefficient PSNR PSNR 0.5389 0.5389 0.6135 0.6135 SSIM SSIM 0.7023 0.7023 0.7321 0.7321 本实施方式 This implementation 0.7754 0.7754 0.7912 0.7912

从表1结果可以看到，本实施方式的结果准确率要高于其他2种现有的视频质量评价方法。It can be seen from the results in Table 1 that the result accuracy rate of this embodiment is higher than that of the other two existing video quality evaluation methods.

Claims

1. A video quality evaluation method based on macroblock domain distortion estimation, comprising the steps:

(1) segment each frame of the video to be evaluated into several macroblocks;

(2) Calculate the blockiness distortion degree of the macroblock;

(3) Calculate the blur effect distortion degree of the macroblock;

(4) Calculate the brightness contrast of the macroblock;

(5) Calculate the texture complexity of the macroblock;

(6) Calculate the motion intensity contrast of the macroblock;

(7) Calculate the degree of consistency of the motion direction of the macroblock;

(8) Calculate the visual perception of the macroblock according to the brightness contrast, texture complexity, motion intensity contrast, and motion direction consistency;

(9) Calculate the quality evaluation value of each frame image of the video to be evaluated according to the degree of block effect distortion, blur effect degree of distortion and visual perception; the quality evaluation value of all images of the video to be evaluated is averaged, and the average value obtained is the quality evaluation value of the video to be evaluated.

2. video quality evaluation method according to claim 1, is characterized in that: in described step (2), the method for calculating the blocking effect distortion degree of macroblock is as follows:

a. Calculate the horizontal blockiness distortion degree of the current macroblock according to the following formula:

S _BH(i) = |A _{(16, i)} -B _{(1, i)} |

S _{IH(1, i)} = |A _{(14, i)} -A _{(15, i)} |

S _{IH (2, i)} = |A _{(15, i)} -A _{(16, i)} |

S _{IH (3, i)} = |B _{(1, i)} -B _{(2, i)} |

S _{IH (4, i)} = |B _{(2, i)} -B _{(3, i)} |

{S S}_{IH IH__AVG AVG ((i i))} = = \frac{{S S}_{IH IH ((11,, i i))} + + {S S}_{IH IH ((22,, i i))} + + {S S}_{IH IH ((33,, i i))} + + {S S}_{IH IH ((44,, i i))}}{44}

{J J}_{H h ((i i))} = = \{\begin{matrix} {S S}_{BH BH ((i i))} - - {S S}_{IH IH__AVG AVG ((i i))} & if if (({S S}_{BH BH ((i i))} > > {S S}_{IH IH__AVG AVG ((i i))})) \\ 00 & otherwise otherwise \end{matrix}

{D D.}_{BLOCK BLOCK__H h} = = \frac{11}{1616} {Σ Σ}_{i i = = 11}^{1616} {J J}_{H h} ((i i))

Among them, D _{BLOCK_H} is the horizontal block effect distortion degree of the current macroblock, A(16, i) is the brightness value of the pixel in the 16th row and column i of the current macroblock, and A(15, i) is the 15th row of the current macroblock The brightness value of the i-column pixel, A(14, i) is the brightness value of the i-column pixel in the 14th row of the current macroblock, and B(1, i) is the first row i of the macroblock adjacent to the bottom of the current macroblock The luminance value of the column pixel, B(2, i) is the luminance value of the i-column pixel in the second row of the macroblock adjacent to the bottom of the current macroblock, and B(3, i) is the macroblock adjacent to the bottom of the current macroblock The brightness value of the pixel in the third row and column i of the block, i is a natural number and 1≤i≤16;

b. Calculate the vertical blockiness distortion degree of the current macroblock according to the following formula:

S _BV(i) = |A _{(i, 16)} -C _{(i, 1)} |

S _{IV(i, 1)} = |A _{(i, 14)} -A _{(i, 15)} |

S _{IV(i, 2)} = |A _{(i, 15)} -A _{(i, 16)} |

S _{IV(i, 3)} = |C _{(i, 1)} -C _{(i, 2)} |

S _{IV(i, 4)} = |C _{(i, 2)} -C _{(i, 3)} |

{S S}_{IV IV__AVG AVG ((i i))} = = \frac{{S S}_{IV IV ((i i,, 11))} + + {S S}_{IV IV ((i i,, 22))} + + {S S}_{IV IV ((i i,, 33))} + + {S S}_{IV IV ((i i,, 44))}}{44}

{J J}_{V V ((i i))} = = \{\begin{matrix} {S S}_{BV BV ((i i))} - - {S S}_{IV IV__AVG AVG ((i i))} & if if (({S S}_{BV BV ((i i))} > > {S S}_{IV IV__AVG AVG ((i i))})) \\ 00 & otherwise otherwise \end{matrix}

{D D.}_{BLOCK BLOCK__V V} = = \frac{11}{1616} {Σ Σ}_{i i = = 11}^{1616} {J J}_{V V ((i i))}

Among them, D _{BLOCK_v} is the vertical block effect distortion degree of the current macroblock, A(i, 16) is the brightness value of the pixel in the 16th column of the i-th row of the current macroblock, and A(i, 15) is the i-th row of the current macroblock The luminance value of the 15th column pixel, A(i, 14) is the luminance value of the pixel in the 14th column of the i-th row of the current macroblock, and C(i, 1) is the 1st row of the i-th row of the macroblock adjacent to the right of the current macroblock The brightness value of the pixel in the column, C(i, 2) is the brightness value of the second column pixel in row i of the macroblock adjacent to the right of the current macroblock, and C(i, 3) is the macroblock adjacent to the right of the current macroblock. The brightness value of the pixel in the third column of the ith row of the block;

c. Calculate the average of the horizontal block-effect distortion degree and the vertical block-effect distortion degree to obtain the block-effect distortion degree of the current macroblock.

3. video quality evaluation method according to claim 1, is characterized in that: in described step (3), the method for calculating the blur effect degree of distortion of macroblock is as follows:

a. Use the Sobel operator to perform edge detection on the macroblock, and determine the edge pixels and their gradient directions in the macroblock;

b. Classify the gradient direction of edge pixels into one of the eight directions {0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°} according to the nearest principle, and use it as Approximate gradient directions for edge pixels;

c. Calculate the edge sharpness of the edge pixels according to the following formula:

S S = = \frac{11}{44} {Σ Σ}_{l l = = 11}^{44} \frac{| | {I I}_{((θ θ,, l l))} - - I I | |}{l l}

Among them: S is the edge sharpness of the edge pixel, I is the brightness value of the edge pixel, I(θ, l) is the brightness value of the pixel whose distance from the edge pixel is l along the approximate gradient direction θ, l is a natural number and 1≤l ≤4;

d. Average the edge sharpness of all edge pixels in the macroblock to obtain the blur effect distortion degree of the macroblock.

4. video quality assessment method according to claim 1, is characterized in that: in described step (4), the method for calculating the brightness contrast of macroblock is: calculate the brightness contrast of each pixel in macroblock according to the following formula , take the maximum value of the brightness contrast of all pixels in the macroblock as the brightness contrast of the macroblock;

{C C}_{Luma Luma__P P} ((θ θ,, l l)) = = \frac{| | {I I}_{((θ θ,, l l))} - - I I | |}{{l l}^{η η}}

{C C}_{Luma Luma} = = {Σ Σ}_{θ θ &Element; &Element; M m}^{M m} {Σ Σ}_{l l = = 11}^{L L} {C C}_{Luma Luma__P P} ((l l,, N N))

Among them, C _Luma is the brightness contrast of the current pixel, I is the brightness value of the current pixel, I(θ, l) is the brightness value of the pixel whose distance from the current pixel is l along the direction θ, l is a natural number and 1≤l≤L , L is the preset maximum distance, η is the pixel distance attenuation coefficient, M is a set of four directions of 0°, 90°, 180° and 270°.

5. video quality evaluation method according to claim 1, is characterized in that: in described step (5), the method for calculating the texture complexity of macroblock is as follows:

a. utilize Sobel operator to carry out edge detection to macroblock, determine the total number of edge pixels in macroblock and the gradient direction of each pixel;

b. Classify the gradient direction of the pixel into one of the eight directions {0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°} according to the principle of proximity, and use it as a pixel The approximate gradient direction of ;

c. Divide the approximate gradient direction of the pixel into four categories: 0° and 180° are classified into one category, 45° and 225° are classified into one category, 90° and 270° are classified into one category, and 135° and 315° are classified into one category kind;

d. Calculate the texture complexity of the macroblock according to the following formula:

{T T}_{block block} = = \{\begin{matrix} 0.5 0.5 & if if (({k k}_{θ θ} = = 11)) \\ 11 & if if (({k k}_{θ θ} = = 22)) \\ ((22 - - {c c}_{e e})) / / 22 & if if (({k k}_{θ θ} = = 33)) \\ ((11 - - {c c}_{e e})) / / 22 & if if (({k k}_{θ θ} = = 44)) \end{matrix}

{c c}_{e e} = = \{\begin{matrix} 11 & if if (({n no}_{edge edge} > > {T T}_{edge edge})) \\ 00 & otherwise otherwise \end{matrix}

Among them, T _block is the texture complexity of the macro block, k _θ is the number of types of approximate gradient directions of all pixels in the macro block, n _edge is the total number of edge pixels in the macro block, and T _edge is the given number of edge pixels threshold.

6. video quality evaluation method according to claim 1, is characterized in that: in described step (6), the method for calculating the motion intensity contrast of macroblock is as follows:

a. Calculate the horizontal motion vector and the vertical motion vector of the macroblock by inter-frame prediction; obtain the motion intensity of the macroblock to the root mean square of the horizontal motion vector and the vertical motion vector;

b. Taking the current macroblock as the center, establish a reference window composed of 7×7 macroblocks;

c. Calculate the motion intensity difference distance ratio _{MI_diff} between the current macroblock and each macroblock in the reference window according to the following formula, and take the maximum value as _{MId_max} :

{M m}_{I I__diff diff} = = \frac{| | {M m}_{I I} - - {M m}_{I I} ((D D.)) | |}{{d d}^{δ δ}}

Wherein: _MI is the motion intensity of the current macroblock, _MI (D) is the motion intensity of the macroblock D, the macroblock D is any macroblock in the reference window, and d is the distance between the current macroblock and the macroblock D, δ is the macroblock distance decay coefficient;

d. Determine the motion intensity contrast of the current macroblock according to the following relationship:

{C C}_{Motion motion__block block} = = \{\begin{matrix} \frac{{M m}_{Id ID__max max}}{{M m}_{I I__max max}} & if if (({M m}_{I I__max max} &NotEqual; &NotEqual; 00)) \\ 00 & otherwise otherwise \end{matrix}

Where: C _{Motion_block} is the motion intensity contrast of the current macroblock, M _{I_max} is the maximum value of the motion intensity of all macroblocks in the reference window.

7. video quality evaluation method according to claim 1, is characterized in that: in described step (7), the method for calculating the motion direction consistency degree of macroblock is as follows:

a. Calculate the horizontal motion vector mv _x and the vertical motion vector mv _y of the macroblock by inter-frame prediction; obtain the motion intensity of the macroblock according to the root mean square of the horizontal motion vector and the vertical motion vector, and according to the formula θ _mv = arctan(mv _y /mv _x ) calculates the motion direction θ _mv of the macroblock;

b. Divide the circle from 0° to 360° into 12 sectors, so that the motion direction of the macroblock is mapped to the corresponding sectors;

c. With the current macroblock as the center, establish a reference window composed of 21×21 macroblocks;

d. Calculate the motion direction consistency degree of the current macroblock according to the following formula:

{M m}_{Con Con__block block} = = - - {Σ Σ}_{j j = = 11}^{1212} p p ((j j)) log log [[p p ((j j))]] \frac{{M m}_{I I__avg avg} ((j j))}{{M m}_{I I__max max}}

p p ((j j)) = = \frac{k k ((j j))}{{21 twenty one}^{22}}

Among them: M _{Con_block} is the motion direction consistency of the current macroblock, k(j) is the total number of macroblocks belonging to the jth sector in the reference window, M _{I_max} is the maximum value of the motion intensity of all macroblocks in the reference window, M _{I_avg} (j) is the average value of the motion intensity of all macroblocks belonging to the jth sector in the reference window, j is a natural number and 1≤j≤12.

8. video quality evaluation method according to claim 1, is characterized in that: in described step (8), calculate the visual perception degree of macroblock according to following formula:

VP _S ＝log(α ₁ +C _{Luma_block} )×(α ₂ +T _block ) ²

VP _T = α ₃ C _{Motion_block} + α ₄ M _{com_block}

VM＝λ×VP _S ×VP _T

Among them: VM is the visual perception of the macro block, M _{Con_block} is the consistency of the motion direction of the macro block, C _{Motion_block} is the motion intensity contrast of the macro block, T _block is the texture complexity of the macro block, and C _{Luma_block} is the brightness contrast of the macro block , λ, α ₁ , α ₂ , α ₃ and α ₄ are given weight coefficients.

9. video quality evaluation method according to claim 1, is characterized in that: in described step (9), calculate the quality evaluation value of every frame image of video to be evaluated according to following formula:

Q Q = = \frac{{Σ Σ}_{n no = = 11}^{N N} [[{Q Q}_{MB MB} ((n no)) \times \times VM VM ((n no))]]}{{Σ Σ}_{n no = = 11}^{N N} VM VM ((n no))}

Q _MB (n)＝γ1×D _{BLOCK_MB} (n)+γ ² ×D _{BLUR_MB} (n)

Among them: Q is the quality evaluation value of any frame image of the video to be evaluated, D _{BLOCK_MB} (n) is the block effect distortion degree of the nth macroblock in any frame image of the video to be evaluated, D _{BLUR_MB} (n) is the value of any frame of the video to be evaluated The blur effect distortion degree of the nth macroblock in a frame image, VM(n) is the visual perception degree of the nth macroblock in any frame image of the video to be evaluated, γ ₁ and γ ₂ are given weight coefficients, n is a natural number and 1≤n≤N, and N is the total number of macroblocks in any frame of video to be evaluated.