KR100434162B1 - Apparatus and Method for Objective Measurement of Video Quality - Google Patents

Abstract

The present invention relates to an objective evaluation of video quality using wavelet transform. The present invention relates to an objective video evaluation score by wavelet transforming a video, generating a difference vector, combining the difference vectors, and weighting each element of the difference vector. Output

According to the present invention, the wavelet transform allows the video to be optimally evaluated according to the characteristics of the human visual system that changes in space-time frequency. In addition, the wavelet transform analysis of the present invention outputs a plurality of parameters that can be used for the calculation of the objective score, the parameters are expressed as a parameter vector, multiplied by the weight, and calculated as a single number, and then the number can be used as the objective score. have.

Description

Apparatus and Method for Objective Measurement of Video Quality}

The present invention relates to objective video quality evaluation, and more specifically, it is possible to calculate a difference in video quality by calculating a wavelet transform of a video and compare coefficients, and to calculate an optimal linear combination of various parameters. The present invention relates to an objective video quality estimation apparatus and method.

In the past, the evaluation of video quality was subjective by a number of evaluators who wanted to evaluate video quality. Such evaluation can be calculated in the presence of a reference video or in the absence of a reference video. In the presence of a reference video, the evaluator sees two videos of the original (reference) video and the processed video to be compared with the reference video. The evaluators then compare the two videos and assign subjective scores to the videos. Therefore, this method is called subjective evaluation of video quality. Such subjective evaluation reflects human perception and therefore has some limitations. First, such an evaluation method requires multiple evaluators. Therefore, such an evaluation method is time-consuming and expensive. Moreover, such an evaluation method has a problem that cannot be achieved in real time.

Accordingly, the present invention seeks to provide a method for objectively evaluating video quality using wavelet transform.

In addition, the present invention is to provide an optimization method capable of outputting an optimal linear combination of various parameters obtained for the objective evaluation of video quality.

1 is a view schematically showing the configuration of an objective video quality evaluation apparatus according to an embodiment of the present invention;

2A illustrates an example of a reference image according to an embodiment of the present invention;

FIG. 2B is a diagram illustrating an example of a video obtained by three-level wavelet transforming the reference image shown in FIG. 2A;

3 is a diagram illustrating an index of a subband block in a three-level wavelet transformed reference image or a processed image according to an embodiment of the present invention;

4 is a diagram illustrating an example of calculating a difference (Squared Error) of coefficients of an i-th index block of a subband of a reference image and a processed image according to an embodiment of the present invention;

5A illustrates an example of a transformed three-dimensional wavelet transform according to an embodiment of the present invention;

5B illustrates an example of calculating a difference vector according to an embodiment of the present invention.

6 is a flowchart schematically illustrating a method for evaluating objective video quality according to an embodiment of the present invention.

<Explanation of symbols for the main parts of the drawings>

310: reference video storage means 320: processed video storage means

330 wavelet transform means 340 difference vector generating means

350: difference vector combining means 360: image quality score output means

An objective video quality estimation apparatus according to the present invention is an apparatus for evaluating spatial frequency quality of a processed video based on a reference video, wherein each frame of the reference video and each frame of the processed video are two-dimensional wavelet transformed. Wavelet transform means for generating reference video wavelet coefficients for each frame of the reference video and generating processed video wavelet coefficients for each frame of the processed video; The difference between the reference video wavelet coefficients generated from the wavelet transform means and the processed video wavelet coefficients is calculated for every subband block for each frame, and the result is generated as one difference vector. Difference vector generating means for outputting a frame difference vector for all frames of the processed video; Difference vector combining means for combining a frame difference vector generated from the difference vector generating means for each subband block to generate a single video difference vector; And weighting means for setting a weight on each element of the video difference vector output from the difference vector combining means, summing all the elements of the video difference vector according to the weight, and outputting a single objective video quality score value. do.

In addition, the objective video quality evaluation apparatus of the present invention is a device for evaluating temporal frequency quality of a processed video based on a reference video, by converting each frame of the reference video and each frame of the processed video by two-dimensional wavelet conversion Wavelet transform means for generating reference video wavelet coefficients for each frame of the reference video and generating processed video wavelet coefficients for each frame of the processed video; The difference between the reference video wavelet coefficients generated from the wavelet transform means and the processed video wavelet coefficients is calculated for every subband block for each frame, and the result is generated as one difference vector. Difference vector generating means for outputting a frame difference vector for all frames of the processed video; Window wavelet transform means for aligning the difference vectors generated by the difference vector generating means in chronological order, applying a predetermined window to each row on the time axis, converting the one-dimensional wavelet to each window, and then outputting the window difference vector. ; Difference vector combining means for combining the window difference vector outputted by the window wavelet converting means for each subband and outputting one video difference vector; And weighting means for setting a weight on each element of the video difference vector output from the difference vector combining means, summing all the elements of the video difference vector according to the weight, and outputting a single objective video quality score value. do.

In addition, the objective video quality evaluation apparatus of the present invention, in the apparatus for evaluating the spatial and temporal frequency image quality of the processed video based on the reference video, calculates the spatial and temporal frequency difference between the reference video and the processed video, Frequency difference generating means for generating a difference for each frequency band for the reference video and the processed video as a vector; And

And weighting means for adding up an objective score by adding the weights to the elements of the vector of frequency differences output from the frequency difference generating means.

The difference generating means for each frequency band wavelet transforms the reference video and the processed video, and outputs a difference vector based on the difference of coefficients of the transformed wavelet.

The objective video quality estimation method of the present invention generates a reference video wavelet coefficient for each frame of the reference video by two-dimensional wavelet transforming each frame of the reference video and each frame of the processed video. A wavelet transform step of generating processed video wavelet coefficients for each frame of video; The difference between the reference video wavelet coefficient generated from the wavelet transform step and the processed video wavelet coefficient is calculated for every subband block for each frame, and the result is generated as one difference vector, and then the reference video is generated. A difference vector generation step of outputting a frame difference vector for all frames of the processed video; A difference vector combining step of generating a single video difference vector by combining the frame difference vectors generated from the difference vector generation step for each subband block; And a weight calculation step of setting weights for each element of the video difference vector output from the difference vector combining step, summing all elements of the video difference vector according to the weight, and outputting a single objective video quality score value. do.

Furthermore, the objective video quality estimation method of the present invention generates a reference video wavelet coefficient for each frame of the reference video by two-dimensional wavelet transforming each frame of the reference video and each frame of the processed video. A wavelet transform step of generating processed video wavelet coefficients for each frame of video; The difference between the reference video wavelet coefficient generated from the wavelet transform step and the processed video wavelet coefficient is calculated for every subband block for each frame, and the result is generated as one difference vector, and then the reference video is generated. A difference vector generation step of outputting a frame difference vector for all frames of the processed video; Window wavelet transformation step of aligning the difference vectors generated from the difference vector generation step in chronological order, applying a predetermined window to each row on the time axis, converting the one-dimensional wavelet to each window, and then outputting the window difference vector. ; Difference vector combining means for combining the window difference vectors outputted by the window wavelet transforming step for each subband and outputting one video difference vector; And a weight calculation step of setting weights to respective elements of the video difference vector outputted from the difference vector combining means, summing all the elements of the video difference vector according to the weights, and outputting a single objective video quality score value. do.

In addition, the objective video quality estimation method of the present invention calculates the spatial and temporal frequency difference between the reference video and the processed video, and generates a frequency band difference for each frequency band for the reference video and the processed video as a vector And a weight calculation step of adding an objective score by summing according to weights given to elements of the vector of frequency differences output from the frequency difference generating step.

Finally, the objective video quality evaluation method of the present invention comprises the steps of setting the score of the processed video; Setting a vector of a desired target; Generating an eigenvector using a covariance matrix of the vector of the target set in the step of setting the target vector; An optimal weight selection step of selecting and outputting an eigenvector corresponding to the maximum eigenvalue as an optimal weight vector.

Hereinafter, embodiments of the present invention will be described in detail with reference to FIGS. 1 to 6.

In general, the effectiveness of an objective assessment is measured by its correlation with the subjective assessment score. In other words, an objective evaluation is best when it provides an evaluation score that is closest to the subjective score.

An object of the present invention is to provide a method for objectively evaluating video quality using wavelet transform. In particular, it aims to provide an objective evaluation method that takes into account the characteristics of the human visual system with different cognitive characteristics at space-time frequencies. The wavelet transform is used to calculate the space-time frequency, and the modified 3-D wavelet transform is used to consider the time frequency. The difference in space-time frequency can be obtained by summing the squared error of the wavelet coefficients in each subband. The difference in space-time frequency is represented by a vector. Each element of this mean vector represents a difference in any space-time frequency band. One number from the vector is calculated as the weighted sum of the elements of the vector, which number is used for objective picture quality evaluation. This process can be optimized to provide the highest correlation with the subjective score.

1 is a diagram schematically showing the configuration of an objective video quality evaluation apparatus according to an embodiment of the present invention. In the figure, reference numeral 310 denotes a reference video storage means, 320 denotes a processed video storage means, 330 denotes a wavelet converting means, 340 denotes a difference vector generating means, 350 denotes a differential vector combining means, and 360 denotes an image quality score output means.

The reference video storage unit 310 stores data of a moving picture or a still image used for image quality evaluation. Similarly, the processed video storage means 320 stores data after calculating a predetermined process of the reference video stored in the reference video storage means. The processed data stored in the processed video storage means 320 is subject to video quality evaluation.

In the drawing, the wavelet converting means 330 converts the video data stored in the reference video storing means 310 and the processed video storing means 320, respectively, and then outputs the wavelet coefficients as wavelet coefficients. The wavelet coefficients output from the wavelet converting means 330 are compared by the subvector blocks by the difference vector generating means 340 and then output as a difference vector.

Then, the difference vector combining means 350 combines the difference vectors output for each frame for each subband block and outputs one difference vector. Finally, the image quality score output means 360 calculates and outputs an image quality score by applying an optimal weight applied to each element of the difference vector output from the difference vector combining means 350.

2A is a diagram illustrating an example of a reference image according to an embodiment of the present invention. 2B is a diagram illustrating an example of an image obtained by three-level wavelet transforming the reference image illustrated in FIG. 2A.

<Example 1>

The present invention shows an example of objective video quality measurement using a reference video. This embodiment assumes that a reference video is provided.

In general, video consists of consecutive frames. One of the simplest methods of measuring the quality of the processed video is to calculate a mean squared error (e _mse ) between the reference video and the processed video, as shown in the following equation.

Where U is the reference video and V is the processed video. M denotes the number of pixels included in one row, N denotes the number of pixels included in one column, and L denotes the number of frames. However, the sensitivity of the human visual system varies with space-time frequency. In other words, the human eye perceives the difference of various spatio-temporal frequency components differently, and it is possible to develop an objective measurement method of video quality by using the characteristics of the human visual system.

Recently, instead of calculating a mean square error between the reference video and the processed video, a weighted difference between various frequency components between the reference video and the processed video is used. Frequency components of a video signal are classified into two types, spatial frequency components and temporal frequency components. The spatial high frequency indicates that the pixel value of the frame is greatly changed. On the other hand, the time high frequency indicates that there is a sudden movement along the frame row. In the case of color video, frequency components may be calculated for each color. Methods such as Fourier Transform and Wavelet Transform are used to calculate frequency components.

The present invention used a wavelet transform. However, even if the Fourier transform is used, the effect of the present invention can be obtained.

3 illustrates an example of three-level wavelet transform of the reference image shown in FIG. 2A. As shown in FIG. 3, the three-level wavelet transform includes ten subband blocks. Each block represents various spatial frequency components. In the figure, block 120 on the top left represents the lowest spatial frequency component of the frame, and block 121 on the bottom right represents the highest spatial frequency component.

If there are two level wavelet transforms, there are seven subband blocks, and in the case of four level wavelet transforms, there are 13 blocks.

Wavelet transform is performed on each frame of the reference video and the processed video to calculate the spatial frequency components. As shown in FIG. 4, the difference (square error) of wavelet coefficients in each subband block is calculated and summed. That is, the difference in coefficients in the i-th subband block can be calculated by the following equation.

[Equation 1]

here Is the j th wavelet coefficient of the i th block of the reference video, Is the j-th wavelet coefficient of the i-th block of video processed corresponding to the reference video. Therefore, in the embodiment of the present invention, since three-level wavelet transform is assumed to be applied, ten values are generated, which can be represented by one vector. Each element of the vector represents a difference of the corresponding subband block. When the above process is performed for every frame, a continuous vector sequence is generated. That is, the difference vector of the l-th frame may be represented by the following equation.

[Equation 2]

here Represents the sum of square errors of the i th block of the l th frame, Is the j th wavelet coefficient of the i th block of the l th frame of the reference video, K is the number of blocks included in the two-dimensional wavelet transform, Denotes the j th wavelet coefficient of the i th block of the l th frame of the processed video. According to such an embodiment of the present invention, a method of obtaining a difference value can be various.

Finally, the average of the vector over the entire frame can be expressed by the following equation.

[Equation 3]

In the present invention, one number may be calculated as the weighted sum of the elements of the mean vector, and the resultant value may be used as an objective measure of the processed video. That is, the above value can be obtained by the following equation.

here Is the weight vector, And K represent the magnitude of the vector.

<Example 2>

The difference in the i-th block of Equation 1 can be obtained by summing the difference of wavelet coefficients of each pixel. However, the human eye cannot recognize the difference between pixels that show a difference smaller than a predetermined threshold. Therefore, the difference of the i-th block can be represented by the following equation in consideration of the characteristics of the human visual system.

here Denotes a threshold.

<Example 3>

The difference vector in Equation 3 represents only a spatial frequency difference. In order to consider the time frequency difference, three-dimensional wavelet transform may be applied. However, applying three-dimensional wavelet transform to video requires a lot of computation and memory. That is, to perform such a task requires a large memory and a long work time. In the embodiment of the present invention, the modified 3D wavelet transform is applied to take into account the time-frequency characteristic of the video. However, if the cost and time can be put enough, the conventional three-dimensional wavelet transform may be applied.

By calculating the difference vector of Equation 2 for the entire frame, a continuous difference vector sequence can be obtained. The columns of the difference vectors may be made into a two-dimensional array such that the difference vectors form respective columns of the two-dimensional array as shown in FIG. 5A. Each row of the two-dimensional array then represents a difference in temporal change of each subband block. In order to calculate the temporal frequency characteristic, one-dimensional wavelet transformation is performed on each row of a two-dimensional array in which each column is a difference vector.

First, as shown in the drawing, a window 220 is applied to each row of a two-dimensional array to generate a segment of a row, and a one-dimensional wavelet transformation is performed on the segment in the time direction. The sum of squares of the respective subbands of the one-dimensional wavelet transform of the j-th row of the l-th window may be calculated by the following equation.

Where l is the l-th window, j is the j-th row, and i is the i-th subband.

By doing this, a result as shown in FIG. 5B can be obtained.

The above process is repeatedly performed for all rows, and all the result values can be expressed as vectors as follows.

In this case, it is assumed that the level of the one-dimensional wavelet transform is three. When performed as above, the size of the result vector becomes larger than the original vector. For example, if the level of the one-dimensional wavelet transform is 3 and the magnitude of the original vector is K, the magnitude of the result vector is 4K.

The window 221 is then moved by a predetermined size and the above process is repeated. In this way, when the processing is completed for all the vector columns, a new vector larger than the original vector can be obtained. The new vector column includes information on temporal frequency characteristics as well as spatial frequency characteristics.

In such a vector, an average vector is calculated as follows.

Here, L 'represents the number of vectors including the information of the spatial and temporal frequency characteristics.

The above embodiment uses a three-dimensional wavelet transform that is modified to process space-time frequency characteristics, but other methods may be used to include the difference between spatial and temporal frequencies. For example, three-dimensional wavelet transforms or three-dimensional Fourier transforms can generally be used to generate a number of parameters representing space-time frequency components. This difference in spatial and temporal frequency is expressed as a vector, and can be used to generate a number representing an objective score by applying the optimum combination of linear differences by the optimization method shown in the following examples. Transforms used to calculate such spatial and temporal frequencies include a Har Transform and a Discrete Cosine Transform.

<Example 4>

Using a two-dimensional wavelet transform, using a modified three-dimensional wavelet transform or using a traditional three-dimensional wavelet transform results in a single vector representing the difference between the reference video and the processed video. It is necessary to calculate a single value as the sum of the weights of the elements of the vector from the vector, and the single value thus calculated is used as the objective score. Such a value can be obtained as follows.

[Equation 4]

Where T stands for Transpose, , , K represents the magnitude of the vector.

First, x is defined as a subjective score such as a Difference Mean Opinion Score (DMOS) of the processed video. Then x and y can be thought of as random variables. In this embodiment, the optimal weight vector W is selected to maximize the correlation coefficient between the variables x and y. The absolute value is important in the correlation coefficient. In other words, two objective test methods with correlation coefficients of 0.9 and -0.9 provide the same performance.

The correlation coefficient between two random variables can be expressed by the following equation.

here If you replace with

Becomes Cov represents covariance and Var represents dispersion.

here Denotes a covariance matrix of D shown in equation (4), and E (·) denotes an expected value operator.

The expected value for the random variable x can be calculated by the following equation.

here Denotes the probability density function of x.

In general, by applying normalization and translation, ego Can be Such normalization and transition do not affect the correlation coefficient with other random variables. After this process, the correlation coefficient can be expressed by the following equation.

here to be.

In this embodiment, the correlation coefficient Find W that maximizes. To simplify the equation, To Instead, by maximizing, the optimal weight value W can be found. And Can be obtained from the following equation.

here to be.

In this example, to find W To maximize Find the gradient of. remind The gradient of can be obtained by the following equation.

As indicated above, W is a weight vector Eigenvector of, Is The eigenvalue of (weight vector). therefore Eigenvectors of is computed first The eigenvector corresponding to is used as the optimal weight vector (W). = Therefore, the correlation coefficient becomes maximum when the eigenvector corresponding to the maximum eigenvalue is used as the optimal weight vector (W).

In Equation 4, the vector D may be any vector. For example, if each element of the vector D represents an arbitrary measure of video quality, then the optimization process can be used to find the optimal weight vector W, thus representing the subjective score and the best correlation coefficient. That is, even if the wavelet transform is not used to calculate the difference between the spatial and temporal frequency components, another method may be used to measure the video quality, and the optimal linear sum may be obtained using the optimization method described above. In this case, the final objective score provides the maximum correlation coefficient of the subjective score.

6 is a flowchart schematically illustrating a method for evaluating objective video quality according to an embodiment of the present invention.

As shown in the figure, first, wavelet transforms the reference video and the processed video (S310). In this case, the result of the wavelet transform is output as a wavelet transform coefficient value for each pixel of each subband block. Subsequently, a difference vector is generated for each subband block based on the output wavelet transform coefficient value (S320).

When the difference vector is generated as described above, one difference vector is generated by combining the difference vectors of the subband blocks (S330). Finally, a weight is applied to all the elements of the generated one difference vector and summed to output a single final evaluation result value (S340).

Therefore, according to the present invention, by using the modified wavelet transform of the video to be evaluated, it is possible to obtain an excellent video quality evaluation result with high processing speed.

In addition, according to the present invention, the weighting coefficient can be calculated and applied so that the human visual system can be evaluated in an approximate manner, and thus can be used as an objective image quality evaluation method.

Claims (14)

An apparatus for evaluating image quality using spatial frequencies of processed video based on reference video, 2D wavelet transform each frame of the reference video and each frame of the processed video to generate a reference video wavelet coefficient for each frame of the reference video, and process the processed video wave for each frame of the processed video. Wavelet transform means for generating a lett coefficient; The difference between the reference video wavelet coefficients generated from the wavelet transform means and the processed video wavelet coefficients is calculated for every subband block for each frame, and the result is generated as one difference vector. Difference vector generating means for outputting a frame difference vector for all frames of the processed video; Difference vector combining means for combining a frame difference vector generated from the difference vector generating means for each subband block to generate a single video difference vector; And Weighting means for setting weights of the elements of the video difference vectors outputted from the difference vector combining means, summing all the elements of the video difference vectors according to the weights, and outputting a single objective video quality score value. An objective video quality evaluation apparatus. The method of claim 1, And the difference vector generating means calculates when the difference between the reference video wavelet coefficient and the processed video wavelet coefficient is equal to or greater than a predetermined threshold value and outputs the difference vector. The method of claim 1, And the weight calculating means calculates an eigenvector corresponding to the maximum eigenvalue of the weight matrix and sets the weight as a weight. An apparatus for evaluating image quality using a spatiotemporal frequency of a processed video based on a reference video, 2D wavelet transform each frame of the reference video and each frame of the processed video to generate a reference video wavelet coefficient for each frame of the reference video, and process the processed video wave for each frame of the processed video. Wavelet transform means for generating a lett coefficient; The difference between the reference video wavelet coefficients generated from the wavelet transform means and the processed video wavelet coefficients is calculated for every subband block for each frame, and the result is generated as one difference vector. Difference vector generating means for outputting a frame difference vector for all frames of the processed video; Window wavelet transform means for aligning the difference vectors generated by the difference vector generating means in chronological order, applying a predetermined window to each row on the time axis, converting the one-dimensional wavelet to each window, and then outputting the window difference vector. ; Difference vector combining means for combining the window difference vector outputted by the window wavelet converting means for each subband and outputting one video difference vector; And Weighting means for setting weights of the elements of the video difference vectors outputted from the difference vector combining means, summing all the elements of the video difference vectors according to the weights, and outputting a single objective video quality score value. An objective video quality evaluation apparatus. The method of claim 4, wherein And the difference vector generating means calculates when the difference between the reference video wavelet coefficient and the processed video wavelet coefficient is equal to or greater than a predetermined threshold value and outputs the difference vector. The method of claim 4, wherein And the weight calculating means calculates an eigenvector corresponding to the maximum eigenvalue of the weight matrix and sets the weight as a weight. An apparatus for evaluating spatial and temporal frequency image quality of a processed video based on a reference video, Frequency band difference generating means for calculating a spatial and temporal frequency difference between the reference video and the processed video, and generating a vector for each frequency band for the reference video and the processed video as a vector; And And weight calculation means for adding up an objective score by adding the weights to the elements of the vector of frequency differences output from the frequency difference generating means. The method of claim 7, wherein And the difference generation means for each frequency band converts the reference video and the processed video into a wavelet, and outputs a difference vector based on a difference of coefficients of the transformed wavelet. The method of claim 8, And the difference generating means for each frequency band calculates and outputs a difference vector when the difference between the reference video wavelet coefficient and the processed video wavelet coefficient is greater than or equal to a predetermined threshold value. The method of claim 7, wherein And the weight calculating means calculates an eigenvector corresponding to the maximum eigenvalue of the weight matrix and sets the weight as a weight. In the method of evaluating the image quality using the spatial frequency of the processed video based on the reference video, 2D wavelet transform each frame of the reference video and each frame of the processed video to generate a reference video wavelet coefficient for each frame of the reference video, and process the processed video wave for each frame of the processed video. Generating a wavelet coefficient; The difference between the reference video wavelet coefficient generated from the wavelet transform step and the processed video wavelet coefficient is calculated for every subband block for each frame, and the result is generated as one difference vector, and then the reference video is generated. A difference vector generation step of outputting a frame difference vector for all frames of the processed video; A difference vector combining step of generating a single video difference vector by combining the frame difference vectors generated from the difference vector generation step for each subband block; And A weight calculation step of setting weights to respective elements of the video difference vector output from the difference vector combining step, summing all the elements of the video difference vector according to the weights, and outputting a single objective video quality score value. Characterized by the objective video quality evaluation method. In the method for evaluating temporal frequency image quality of processed video based on reference video, 2D wavelet transform each frame of the reference video and each frame of the processed video to generate a reference video wavelet coefficient for each frame of the reference video, and process the processed video wave for each frame of the processed video. Generating a wavelet coefficient; The difference between the reference video wavelet coefficient generated from the wavelet transform step and the processed video wavelet coefficient is calculated for every subband block for each frame, and the result is generated as one difference vector, and then the reference video is generated. A difference vector generation step of outputting a frame difference vector for all frames of the processed video; Window wavelet transformation step of aligning the difference vectors generated from the difference vector generation step in chronological order, applying a predetermined window to each row on the time axis, converting the one-dimensional wavelet to each window, and then outputting the window difference vector. ; Difference vector combining means for combining the window difference vectors outputted by the window wavelet transforming step for each subband and outputting one video difference vector; And A weighting step of setting weights of the elements of the video difference vectors output from the difference vector combining means, summing all the elements of the video difference vectors according to the weights, and outputting a single objective video quality score value Characterized by the objective video quality evaluation method. In the method for evaluating the image quality using the space-time frequency of the processed video based on the reference video, A step of generating a frequency band difference for calculating a spatial and temporal frequency difference between the reference video and the processed video, and generating a difference of each frequency band for the reference video and the processed video as a vector; And And a weight calculation step of outputting an objective score by summing according to weights given to elements of the vector of frequency differences output from the frequency difference generating step. In the method for evaluating the temporal frequency image quality of the processed video based on the reference video, a method for generating coefficients for combining the parameter vector, Setting a score x of the processed video; Setting a parameter vector D indicative of the objective quality; Covariance matrix of the parameter vector set in the step of setting the parameter vector Wow Weight matrix Generating a; The weight matrix And an optimal weight selection step of selecting and outputting an eigenvector corresponding to the maximum eigenvalue as an optimal weight vector.