A virtual view image quality evaluation method and system based on dual-tree complex wavelet transform

Abstract

The application discloses a virtual viewpoint image quality evaluation method and system based on double-tree complex wavelet transform. The method comprises the following steps: using an algorithm to perform feature matching on a reference image; performing multi-scale double-tree complex wavelet transform on the matched reference image and a virtual viewpoint image for multiple times to obtain a first wavelet subband set corresponding to the matched reference image and a second wavelet subband set corresponding to the virtual viewpoint image; calculating gradient amplitude similarity between third-level wavelet subbands of the first wavelet subband set and the second wavelet subband set, and using the gradient amplitude similarity to calculate a texture distortion score of the virtual viewpoint image; calculating structural similarity between fifth-level wavelet subbands of the first wavelet subband set and the second wavelet subband set, and using the structural similarity to calculate a structure distortion score of the virtual viewpoint image; and calculating a quality score of the virtual viewpoint image according to the texture distortion score and the structure distortion score. The application can accurately evaluate the quality of the virtual viewpoint image.

Description

Technical Field

[0001] This invention relates to the field of image quality assessment technology, and in particular to a virtual viewpoint image quality assessment method and system based on dual-tree complex wavelet transform. Background Technology

[0002] In recent years, the rise of multi-point and free-viewpoint video has provided viewers with more immersive viewing experiences. These 3D applications allow users to view the same scene from different angles, but this also brings many problems, such as significant information redundancy and storage and transportation costs. Therefore, Depth-Image-Based-Rendering...

[0003] The DIBR (Deep Image Reduction) technology emerged to address this need. It requires only a texture map and depth map captured by the camera, along with images from a portion of the viewpoint. All other viewpoint images are rendered using DIBR, significantly reducing the storage and transportation costs of 3D video. DIBR primarily consists of two processes: 3D warping and hole filling. 3D warping first maps pixels from the texture map to 3D space through the original viewpoint, and then back-maps them to the target virtual viewpoint using the depth map. Due to the change in perspective, some objects that were invisible at the original viewpoint become visible at the target viewpoint; these areas are called de-occlusion regions, where black hole distortion is prone to occur. The second step is hole filling. However, due to the de-occlusion regions, some imperfect rendering algorithms use the texture surrounding the black hole to repair it, which can easily lead to various distortions such as ghosting and distortion. Furthermore, the texture or depth map in DIBR technology significantly impacts the quality of the virtual viewpoint image. While DIBR technology has brought tremendous development to 3D video applications, it has also introduced new problems, such as the quality of virtual viewpoint images. Therefore, quality evaluation algorithms for virtual viewpoints have emerged.

[0004] Existing technologies propose quality assessment models for virtual viewpoint images, but most of these models may only consider a single type of distortion (such as structural or texture distortion). With advancements in rendering techniques, many distortions that previously appeared in virtual viewpoint images no longer exist, thus rendering some previously well-performing quality assessment models inapplicable. Summary of the Invention

[0005] The primary objective of this invention is to overcome the problems existing in the prior art and provide a virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform. This invention can improve the accuracy of virtual viewpoint image evaluation.

[0006] As another objective of the present invention, a system adapted to the method based on the aforementioned objective is also provided.

[0007] As a further object of the present application, a non-volatile storage medium suitable for storing a computer program implemented according to the method described is provided.

[0008] To achieve the above primary object, the present application provides a virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform, comprising:

[0009] Step S1: obtaining a reference image and a virtual viewpoint image to be evaluated;

[0010] Step S2: using an algorithm to perform feature matching on the reference image to obtain a matched reference image;

[0011] Step S3: performing multi-scale dual-tree complex wavelet transform on the matched reference image and the virtual viewpoint image to obtain a first set of wavelet subbands corresponding to the matched reference image and a second set of wavelet subbands corresponding to the virtual viewpoint image, wherein the first set of wavelet subbands and the second set of wavelet subbands each contain multiple levels of wavelet subbands;

[0012] Step S4: calculating the gradient amplitude similarity between the third level wavelet subband of the first set of wavelet subbands and the third level wavelet subband of the second set of wavelet subbands, and using the gradient amplitude similarity to calculate the texture distortion score of the virtual viewpoint image;

[0013] Step S5: calculating the structural similarity between the fifth level wavelet subband of the first set of wavelet subbands and the fifth level wavelet subband of the second set of wavelet subbands, and using the structural similarity to calculate the structural distortion score of the virtual viewpoint image;

[0014] Step S6: calculating the quality score of the virtual viewpoint image according to the texture distortion score and the structural distortion score.

[0015] Further, the algorithm in step S2 includes a Speeded-Up Robust Features feature matching algorithm and a RANSAC algorithm, wherein the Speeded-Up Robust Features feature matching algorithm is used to match feature points, and the RANSAC algorithm is used to eliminate inaccurate feature points.

[0016] Further, step S3 specifically comprises:

[0017] Step S3.1: performing multi-scale dual-tree complex wavelet transform on the matched reference image and the virtual viewpoint image to obtain a first low-frequency subband and six first high-frequency subbands of different directions corresponding to the matched reference image, and a second low-frequency subband and six second high-frequency subbands of different directions corresponding to the virtual viewpoint image;

[0018] Step S3.2: performing multiple dual-tree complex wavelet transform on the first low-frequency subband and the second low-frequency subband to obtain the first set of wavelet subbands and the second set of wavelet subbands, each time of transform being performed on a low-frequency wavelet subband obtained in the last time of transform, and each time of transform obtaining one low-frequency wavelet subband and six high-frequency wavelet subbands of different directions.

[0019] Further, the six different directions are ±15°, ±45° and ±75°, respectively.

[0020] Further, the step S3.2 includes performing at least four times of dual-tree complex wavelet transform.

[0021] Further, the step S4 specifically includes:

[0022] Step S4.1: calculating image gradient amplitudes of the wavelet subbands using a Prewitt filter, specifically:

[0023]

[0024] wherein, is a convolution kernel in a horizontal direction, is a convolution kernel in a vertical direction;

[0025] Step S4.2: calculating gradient amplitudes of the third-level wavelet subbands of the matched reference image and the third-level wavelet subbands of the virtual view image at a pixel (i,j), specifically:

[0026]

[0027] wherein, r is the third-level wavelet subband of the matched reference image after dual-tree complex wavelet transform, d is the third-level wavelet subband of the virtual view image after dual-tree complex wavelet transform, M r (i,j) is a gradient amplitude value of the wavelet subband of the reference image at the pixel (i,j), M d (i,j) is a gradient amplitude value of the wavelet subband of the virtual view image at the pixel (i,j), denotes a convolution operation;

[0028] Step S4.3: calculating gradient amplitude similarity of the third-level wavelet subbands of the reference image and the virtual view image one by one, denoted as GMS, in the following manner:

[0029]

[0030] wherein, c is a non-zero constant;

[0031] Step S4.4: performing average pooling on the obtained gradient amplitude similarities of the six directions to obtain an average pooling result, denoted as Q’ T , specifically,

[0032]

[0033] where M, N are the height and width of the wavelet sub-band image respectively;

[0034] Step S4.5: summing the six wavelet sub-bands to obtain a texture distortion score, denoted as Q T , specifically,

[0035]

[0036] is the gradient amplitude similarity index of the six different direction wavelet sub-bands of the third level, is the weight of the six different direction wavelet sub-bands of the third level and is set to 1.

[0037] Further, step S5 specifically comprises:

[0038] Step S5.1: calculating the luminance contrast, contrast contrast and structure contrast of the fifth level wavelet sub-bands of the reference image and the virtual viewpoint image after feature matching respectively, and the specific calculation formula is as follows:

[0039]

[0040] wherein, is the luminance contrast; is the contrast contrast; is the structure contrast; , , are the mean, variance and covariance of the fifth level wavelet sub-band respectively; the covariance represents and the trend of common change; and are the fifth level wavelet sub-bands of the reference image and the virtual viewpoint image after feature matching respectively; , , are all constant constants, and ;

[0041] Step S5.2: calculating the structure distortion score of the virtual viewpoint image according to the three indexes of luminance contrast, contrast contrast and structure contrast, denoted as Q S , specifically,

[0042]

[0043] Wherein, k is the fifth level six different direction wavelet sub-band structure similarity index; and Both take 0.2; γ takes 0.6; Indicate that six wavelet sub-bands give the same weight.

[0044] Further, step S6 is specifically: calculating the quality score Q of the virtual viewpoint image according to the texture distortion score and the structure distortion score, and the calculation formula is as follows:

[0045] .

[0046] In order to achieve another object of the present application, the present application provides a virtual viewpoint image quality evaluation system based on dual-tree complex wavelet transform, comprising:

[0047] The acquisition module is used for acquiring a reference image and a virtual viewpoint image to be evaluated;

[0048] The matching module is used for performing feature matching on the reference image using an algorithm to obtain a matched reference image;

[0049] The transformation module is used for performing multiple multi-scale dual-tree complex wavelet transform on the matched reference image and the virtual viewpoint image to obtain a first wavelet sub-band set corresponding to the matched reference image and a second wavelet sub-band set corresponding to the virtual viewpoint image, wherein the first wavelet sub-band set and the second wavelet sub-band set both contain multiple level wavelet sub-bands;

[0050] The first calculation module is used for calculating the gradient amplitude similarity between the third level wavelet sub-band of the first wavelet sub-band set and the third level wavelet sub-band of the second wavelet sub-band set, and calculating the texture distortion score of the virtual viewpoint image using the gradient amplitude similarity;

[0051] The second calculation module is used for calculating the structure similarity between the fifth level wavelet sub-band of the first wavelet sub-band set and the fifth level wavelet sub-band of the second wavelet sub-band set, and calculating the structure distortion score of the virtual viewpoint image using the structure similarity;

[0052] The third calculation module is used for calculating the quality score of the virtual viewpoint image according to the texture distortion score and the structure distortion score.

[0053] In order to achieve another object of the present application, the present application provides a computer readable storage medium having a computer program of a virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform stored thereon, and the computer program of the virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform is processed to realize the steps of the virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform.

[0054] Compared with the prior art, the present application has the beneficial effects that:

[0055] The reference image and the virtual viewpoint image are subjected to multi-scale and multi-direction feature decomposition through the dual-tree complex wavelet transform, which is more in line with the human visual system perception process than the traditional two-dimensional wavelet transform, and can improve the evaluation accuracy of the virtual viewpoint image.

[0056] The reference image is subjected to offset compensation first, and then the dual-tree complex wavelet transform is performed on the reference image and the virtual viewpoint image after feature matching to obtain a plurality of wavelet subbands of different directions and scales, the gradient amplitude similarity of the third-level wavelet subband is calculated to measure the texture distortion of the virtual viewpoint image, and the structural similarity of the fifth-level wavelet subband is calculated to measure the structural distortion, so that the texture distortion and the structural distortion which are more likely to occur in the virtual viewpoint image can be measured more accurately, the types of distortion are considered more comprehensively, and the evaluation accuracy of the virtual viewpoint image is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 is a flowchart of a virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform according to an embodiment of the present application;

[0058] Figure 2 is a block diagram of a virtual viewpoint image quality evaluation system based on dual-tree complex wavelet transform according to an embodiment of the present application;

[0059] Figure 3 is a structural block diagram of a virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform according to an embodiment of the present application. DETAILED DESCRIPTION

[0060] The specific embodiments of the present application will be further described in detail below in combination with the drawings and examples. The following examples are used to illustrate the present application, but are not used to limit the scope of the present application.

[0061] Example 1

[0062] As shown in Figure 1 , 3 , a virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform according to a preferred embodiment of the present application comprises:

[0063] Step S1: obtaining a reference image and a virtual viewpoint image to be evaluated;

[0064] In one specific embodiment, the reference image and the virtual viewpoint image to be evaluated are obtained from a known data set.

[0065] Step S2: feature matching of the reference image using an algorithm to obtain a matched reference image;

[0066] In one embodiment, the algorithm in step S2 includes a Speeded-Up Robust Features (SURF) feature matching algorithm and a RANSAC algorithm, the SURF feature matching algorithm is used to match feature points, and the RANSAC algorithm is used to eliminate inaccurate feature points. The SURF+RANSAC is used to match the reference image to solve the pixel offset problem of the reference image and the virtual viewpoint image, highlight the image features, and reduce the influence on the full-reference image quality evaluation model.

[0067] Step S3: performing multi-scale dual-tree complex wavelet transform on the matched reference image and the virtual viewpoint image multiple times to obtain a first set of wavelet subbands corresponding to the matched reference image and a second set of wavelet subbands corresponding to the virtual viewpoint image, the first set of wavelet subbands and the second set of wavelet subbands each include multiple levels of wavelet subbands;

[0068] In one embodiment, step S3 specifically includes:

[0069] Step S3.1: performing multi-scale dual-tree complex wavelet transform on the matched reference image and the virtual viewpoint image to obtain a first low-frequency subband and six first high-frequency subbands of different directions corresponding to the matched reference image, and a second low-frequency subband and six second high-frequency subbands of different directions corresponding to the virtual viewpoint image; the six different directions are ±15°, ±45°, and ±75°.

[0070] Step S3.2: performing multi-scale dual-tree complex wavelet transform on the first low-frequency subband and the second low-frequency subband to obtain the first set of wavelet subbands and the second set of wavelet subbands, each time of transform is performed on the low-frequency wavelet subband obtained in the last time, and each time of transform obtains a low-frequency wavelet subband and six high-frequency wavelet subbands of different directions, and in one embodiment, the dual-tree complex wavelet transform is performed at least four times.

[0071] Step S4: calculating a gradient amplitude similarity between a third level wavelet subband of the first set of wavelet subbands and a third level wavelet subband of the second set of wavelet subbands, and using the gradient amplitude similarity to calculate a texture distortion score of the virtual viewpoint image;

[0072] In one embodiment, step S4 specifically includes:

[0073] Step S4.1: calculating the image gradient amplitude of the wavelet subband using a Prewitt filter, specifically:

[0074]

[0075] wherein, is a horizontal direction convolution kernel, is a vertical direction convolution kernel;

[0076] Step S4.2: Calculate the gradient amplitudes of the third level wavelet subbands of the matched reference image and the third level wavelet subbands of the virtual viewpoint image at pixel (i,j), specifically as follows:

[0077]

[0078] wherein, r is the third level wavelet subband of the matched reference image after double tree complex wavelet transform, d is the third level wavelet subband of the virtual viewpoint image after double tree complex wavelet transform, M r (i,j) is the gradient amplitude value of the wavelet subband of the reference image at pixel (i,j), M d (i,j) is the gradient amplitude value of the wavelet subband of the virtual viewpoint image at pixel (i,j), represents a convolution operation;

[0079] Step S4.3: Calculate the gradient amplitude similarity of the third level wavelet subbands of the reference image and the virtual viewpoint image one by one, denoted as GMS, in the following manner:

[0080]

[0081] wherein, c is a non-zero constant;

[0082] Step S4.4: Average pool the gradient amplitude similarities of the six directions to obtain an average pooling result, denoted as Q ’ T , specifically as follows:

[0083]

[0084] wherein, M and N are the height and width of the wavelet subband image, respectively;

[0085] Step S4.5: Sum the six wavelet subbands to obtain a texture distortion score, denoted as Q T , specifically as follows:

[0086]

[0087] is the gradient amplitude similarity index of the third level six different direction wavelet subbands, is the weight of the third level six different direction wavelet subbands and is set to 1.

[0088] Step S5: calculating the structural similarity between the fifth level wavelet subband of the first set of wavelet subbands and the fifth level wavelet subband of the second set of wavelet subbands, and using the structural similarity to calculate a structural distortion score of the virtual view image;

[0089] In one specific embodiment, step S5 specifically comprises:

[0090] Step S5.1: calculating three indexes of luminance contrast, contrast contrast and structure contrast of the fifth level wavelet subband of the reference image and the virtual view image respectively after feature matching, and the specific calculation formula is as follows:

[0091]

[0092] Wherein, is the luminance contrast; is the contrast contrast; is the structure contrast; , , are the mean, variance and covariance of the fifth level wavelet subband respectively; the covariance represents the trend of common variation with ; and are the fifth level wavelet subbands of the reference image and the virtual view image respectively after feature matching; , , are all constant constants, and ;

[0093] Step S5.2: calculating the structural distortion score of the virtual view image according to the three indexes of luminance contrast, contrast contrast and structure contrast, denoted as Q S , and the specific calculation formula is as follows:

[0094]

[0095] Wherein, k is the structural similarity index of the fifth level six different direction wavelet subbands; and are both 0.2; γ is 0.6; , which indicates that the six wavelet subbands give the same weight.

[0096] Step S6: calculating the quality score of the virtual view image according to the texture distortion score and the structural distortion score.

[0097] In one specific embodiment, step S6 specifically comprises: calculating the quality score Q of the virtual view image according to the texture distortion score and the structural distortion score, and the calculation formula is as follows:

[0098] .

[0099] And, using the PLCC, SRCC, RMSE performance indicators to evaluate the performance of the virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform proposed in the application, experiments are carried out on the public data set IETR, according to the above experimental steps, the consistency of the objective quality score and the true label score of the application on the experimental data is calculated, and compared with other virtual viewpoint image quality evaluation models, the performance of the model is judged. In the experiment, the comparison methods involved are APT, SEQUSS, SSDP, LOGS, MP-PSNR-reduce, MNSS, BDWT. Among them, the higher the PLCC, SRCC represents the better the model performance, the lower the RMSE represents the better the performance, and the results are shown in the following table:

[0100]

[0101] Example two

[0102] As Figure 2 shown, a virtual viewpoint image quality evaluation system based on dual-tree complex wavelet transform according to an embodiment of the application comprises:

[0103] The acquisition module is configured to acquire a reference image and a virtual viewpoint image to be evaluated.

[0104] The matching module is configured to use an algorithm to perform feature matching on the reference image to obtain a matched reference image.

[0105] The transformation module is configured to perform multiple multi-scale dual-tree complex wavelet transforms on the matched reference image and the virtual viewpoint image to obtain a first set of wavelet subbands corresponding to the matched reference image and a second set of wavelet subbands corresponding to the virtual viewpoint image, wherein the first set of wavelet subbands and the second set of wavelet subbands each contain multiple levels of wavelet subbands.

[0106] The first calculation module is configured to calculate the gradient amplitude similarity between the third level wavelet subband of the first set of wavelet subbands and the third level wavelet subband of the second set of wavelet subbands, and use the gradient amplitude similarity to calculate the texture distortion score of the virtual viewpoint image.

[0107] The second calculation module is configured to calculate the structural similarity between the fifth level wavelet subband of the first set of wavelet subbands and the fifth level wavelet subband of the second set of wavelet subbands, and use the structural similarity to calculate the structural distortion score of the virtual viewpoint image.

[0108] The third calculation module is configured to calculate the quality score of the virtual viewpoint image according to the texture distortion score and the structural distortion score.

[0109] The embodiment decomposes the reference image and the virtual viewpoint image in multi-scale and multi-direction through double-tree complex wavelet transform, is more in line with the human visual system perception process compared with the traditional two-dimensional wavelet transform, and can improve the evaluation accuracy of the virtual viewpoint image.

[0110] The reference image is offset compensated first, and then the double-tree complex wavelet transform is performed on the reference image and the virtual viewpoint image matched in features to obtain a plurality of wavelet subbands in different directions and scales, the gradient amplitude similarity of the third level wavelet subband is calculated to measure the texture distortion of the virtual viewpoint image, and the structural similarity of the fifth level wavelet subband is calculated to measure the structural distortion, so that the texture distortion and the structural distortion which are more in the virtual viewpoint image can be more accurately measured, the types of distortion are more comprehensively considered, and the evaluation accuracy of the virtual viewpoint image is effectively improved.

[0111] Embodiment three

[0112] The embodiment also provides a computer readable storage medium, which stores a computer program of the virtual viewpoint image quality evaluation method based on double-tree complex wavelet transform.

[0113] In summary, the embodiment of the present application provides a virtual viewpoint image quality evaluation method and system based on double-tree complex wavelet transform, which decomposes the reference image and the virtual viewpoint image in multi-scale and multi-direction through double-tree complex wavelet transform, is more in line with the human visual system perception process compared with the traditional two-dimensional wavelet transform, and can improve the evaluation accuracy of the virtual viewpoint image; the reference image is offset compensated first, and then the double-tree complex wavelet transform is performed on the reference image and the virtual viewpoint image matched in features to obtain a plurality of wavelet subbands in different directions and scales, the gradient amplitude similarity of the third level wavelet subband is calculated to measure the texture distortion of the virtual viewpoint image, and the structural similarity of the fifth level wavelet subband is calculated to measure the structural distortion, so that the texture distortion and the structural distortion which are more in the virtual viewpoint image can be more accurately measured, the types of distortion are more comprehensively considered, and the evaluation accuracy of the virtual viewpoint image is effectively improved.

[0114] The above only describes the preferred embodiments of the present application, and it should be noted that, for those skilled in the art, without departing from the technical principles of the present application, several improvements and replacements can be made, and these improvements and replacements should also be considered as the protection scope of the present application.

Claims

1. A virtual view image quality evaluation method based on dual-tree complex wavelet transform, characterized by, The method comprises the following steps: Step S1: obtaining a reference image and a virtual viewpoint image to be evaluated; Step S2: performing feature matching on the reference image using an algorithm to obtain a matched reference image; Step S3: performing multi-scale dual-tree complex wavelet transform on the matched reference image and the virtual viewpoint image multiple times to obtain a first set of wavelet subbands corresponding to the matched reference image and a second set of wavelet subbands corresponding to the virtual viewpoint image, wherein the first set of wavelet subbands and the second set of wavelet subbands each comprise multiple levels of wavelet subbands; Step S4: calculating the gradient magnitude similarity between the third level wavelet subband of the first set of wavelet subbands and the third level wavelet subband of the second set of wavelet subbands, and using the gradient magnitude similarity to calculate the texture distortion score of the virtual viewpoint image; Step S5: calculating the structural similarity between the fifth level wavelet subband of the first set of wavelet subbands and the fifth level wavelet subband of the second set of wavelet subbands, and using the structural similarity to calculate the structural distortion score of the virtual viewpoint image; Step S6: calculating the quality score of the virtual viewpoint image according to the texture distortion score and the structural distortion score.

2. The virtual view image quality assessment method based on dual-tree complex wavelet transform according to claim 1, characterized in that, The algorithm in step S2 comprises a Speeded-Up Robust Features feature matching algorithm and a RANSAC algorithm, wherein the Speeded-Up Robust Features feature matching algorithm is used to match feature points, and the RANSAC algorithm is used to eliminate feature points with inaccurate matching.

3. The virtual view image quality assessment method based on dual-tree complex wavelet transform according to claim 1, characterized in that, Step S3 specifically comprises: Step S3.1: performing multi-scale dual-tree complex wavelet transform on the matched reference image and the virtual viewpoint image to obtain a first low-frequency subband and six first high-frequency subbands of different directions corresponding to the matched reference image, and a second low-frequency subband and six second high-frequency subbands of different directions corresponding to the virtual viewpoint image; Step S3.2: performing multi-time dual-tree complex wavelet transform on the first low-frequency subband and the second low-frequency subband to obtain the first set of wavelet subbands and the second set of wavelet subbands, wherein each time of transform is performed on the low-frequency wavelet subband obtained in the previous time, and each time of transform obtains a low-frequency wavelet subband and six high-frequency wavelet subbands of different directions.

4. The virtual view image quality assessment method based on dual-tree complex wavelet transform according to claim 3, characterized in that, The six different directions are ±15°, ±45° and ±75°.

5. The virtual view image quality assessment method based on dual-tree complex wavelet transform according to claim 3, characterized in that, In step S3.2, at least four times of dual-tree complex wavelet transform are performed.

6. The virtual view image quality assessment method based on dual-tree complex wavelet transform according to claim 1, wherein, Step S4 specifically comprises: Step S4.1: calculating the image gradient magnitude of the wavelet subband using a Prewitt filter, specifically as follows: wherein, is a convolution kernel in a horizontal direction, is a convolution kernel in a vertical direction; Step S4.2: calculating the gradient magnitude of the third level wavelet subband of the matched reference image and the third level wavelet subband of the virtual viewpoint image at pixel (i, j), specifically as follows: wherein r is the third level wavelet subband of the matched reference image after double tree complex wavelet transform, d is the third level wavelet subband of the virtual view image after double tree complex wavelet transform, M r (i,j) is the gradient magnitude value of the wavelet subband of the reference image at pixel (i,j), M d (i,j) is the gradient magnitude value of the wavelet subband of the virtual view image at pixel (i,j), denotes a convolution operation; Step S4.3: calculating the gradient magnitude similarity between the third level wavelet subbands of the reference image and the virtual viewpoint image one by one, denoted as GMS, in the following manner: where c is a non-zero constant; Step S4.4: average pooling is performed on the obtained gradient magnitude similarities of the six directions to obtain an average pooling result, denoted as , specifically: where M and N are the height and width of the wavelet subband image, respectively; Step S4.5: Summing the six wavelet subbands to obtain a texture distortion score, denoted as Q T Specifically, a gradient magnitude similarity of six different directional wavelet subband indices of the third level, a weight of six different directional wavelet subbands of the third level and is set to 1.

7. The virtual view image quality assessment method based on dual-tree complex wavelet transform according to claim 6, wherein, Step S5 specifically comprises: Step S5.1: Calculate the three indexes of the fifth level wavelet sub-band of the reference image and the virtual viewpoint image after feature matching, respectively, which are luminance contrast, contrast contrast and structure contrast, and the specific calculation formula is as follows: in, For brightness contrast; For contrast comparison; For structural comparison; , , These are the mean, variance, and covariance of the fifth-level wavelet subband; covariance represent and A common trend of change; and These are the fifth-level wavelet sub-bands of the reference image after feature matching and the virtual viewpoint image, respectively; , , All are constants, and ; Step S5.2: Calculate the structure distortion score of the virtual viewpoint image according to the three indexes of the brightness contrast, the contrast contrast and the structure contrast, denoted as Q S , specifically: where k is the fifth level six different direction wavelet sub-band structure similarity index; and Both take 0.2; γ takes 0.6; , indicates that six wavelet sub-bands give the same weight.

8. The virtual view image quality assessment method based on dual-tree complex wavelet transform according to claim 7, characterized in that, Step S6 is specifically: calculating the quality score Q of the virtual viewpoint image according to the texture distortion score and the structure distortion score, and the calculation formula is as follows: 。 9. A virtual view image quality evaluation system based on dual tree complex wavelet transform, characterized by, Comprise: An acquisition module is configured to acquire a reference image and a virtual viewpoint image to be evaluated; A matching module is configured to perform feature matching on the reference image using an algorithm to obtain a matched reference image; A transformation module is configured to perform multi-scale dual-tree complex wavelet transformation on the matched reference image and the virtual viewpoint image multiple times to obtain a first set of wavelet sub-bands corresponding to the matched reference image and a second set of wavelet sub-bands corresponding to the virtual viewpoint image, wherein the first set of wavelet sub-bands and the second set of wavelet sub-bands each contain multiple levels of wavelet sub-bands; A first calculation module is configured to calculate the gradient amplitude similarity between the third level wavelet sub-band of the first set of wavelet sub-bands and the third level wavelet sub-band of the second set of wavelet sub-bands, and use the gradient amplitude similarity to calculate the texture distortion score of the virtual viewpoint image; A second calculation module is configured to calculate the structural similarity between the fifth level wavelet sub-band of the first set of wavelet sub-bands and the fifth level wavelet sub-band of the second set of wavelet sub-bands, and use the structural similarity to calculate the structure distortion score of the virtual viewpoint image; A third calculation module is configured to calculate the quality score of the virtual viewpoint image according to the texture distortion score and the structure distortion score.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer storage program is executed by the processor to realize the virtual viewpoint image quality evaluation method based on dual-tree complex wavelet transform according to any one of claims 1 to 8.

Images