Method and device for image restoration

Abstract

The present invention provides a method for image restoration. The method may include: obtaining a target image to be restored and a plurality of initial reference images; selecting a plurality of reference images from the plurality of initial reference images based on the suitability of each of the plurality of initial reference images for assisting in restoring the target image; transforming each of the plurality of reference images into a registered image having the same viewing angle as the target image; and combining information of each registered image to restore the target image. In addition, the present invention also provides an apparatus for image restoration. Through the present invention, the target image can be accurately and reasonably reconstructed with a small amount of calculation, and the restoration result has an accurate structure and a natural texture.

Description

Technical Field

[0001] The present invention relates to image processing, and more particularly, to a method and apparatus for image restoration. Background Art

[0002] Image restoration technology is to make reasonable guesses about the damaged areas of an image according to certain rules, and restore it to a complete image with continuous structure and natural texture. Different restoration algorithms will produce different results for different regional defects, so many excellent technologies have been derived in the field of image restoration. These restoration algorithms are divided into two categories: classic traditional algorithms and new algorithms that have emerged in recent years: traditional algorithms include methods based on partial differential equations, methods based on texture synthesis, and methods based on sparse representation, and new algorithms include deep learning algorithms and multi-view restoration algorithms.

[0003] Traditional image restoration algorithms are studied on the basis of a single image. Due to the limited information of the source area, when some areas with weak texture and high structure are missing, it is difficult to find image blocks with similar texture in the known information area. This type of algorithm performs unsatisfactory when repairing images with large missing areas, and there are often situations where texture extension or large areas are blurred. In recent years, scholars have introduced deep learning theory to use high-level semantic information to repair missing areas based on the accumulated prior information. Deep learning has powerful representation capabilities, but a large number of training samples may only achieve good results for scene images in a specific range. Further research is needed on the universality of this method.

[0004] Multi-view image restoration technology combines the known information of multiple reference images with similar scenes to the target image but different perspectives, matches the points of interest between the target image and the reference view image, aligns the reference view image to the same perspective as the target image, and then extracts the useful information to fill the target image. At present, most of the multi-view related restoration algorithms are based on video sequences, selecting frames with similar video frames and small perspective changes, or requiring depth map information to assist in restoration, which leads to certain limitations of such algorithms. In image restoration algorithms that only study color images, the reference view image is given by the user, or the viewpoint-invariant image search method is used to automatically search for reference images on the Internet. Some scholars have proposed using the minimum global energy function to search for blocks similar to the target image blocks in a set of registered images. This algorithm has a wide range of applicability, but it is quite time-consuming in calculation.

[0005] Accordingly, there is a need in the art for an image restoration technology that has a wide range of applications, requires less computation and can accurately and reasonably reconstruct a target image. Summary of the invention

[0006] This Summary is provided to introduce some concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

[0007] In view of the defects in the prior art described above, an object of the present invention is to efficiently repair an image with damaged areas with a small amount of calculation.

[0008] According to one aspect of the present invention, a method for image restoration is provided, which may include: obtaining a target image to be restored and multiple initial reference images; selecting multiple reference images from the multiple initial reference images based on the suitability of each of the multiple initial reference images to assist in restoring the target image; transforming each of the multiple reference images into a registered image having the same viewing angle as the target image; and combining information of each registered image to restore the target image.

[0009] According to another aspect of the present invention, a device for image restoration is provided, which may include: a memory; and a processor coupled to the memory, wherein the processor is configured to: obtain a target image to be restored and a plurality of initial reference images; select a plurality of reference images from the plurality of initial reference images based on the suitability of each of the plurality of initial reference images to assist in restoring the target image; transform each of the plurality of reference images into a registered image having the same viewing angle as the target image; and combine information of each registered image to restore the target image.

[0010] According to another aspect of the present invention, a device for image restoration is provided, which may include: an image acquisition module, configured to obtain a target image to be restored and a plurality of initial reference images; a reference image selection module, configured to select a plurality of reference images from the plurality of initial reference images based on the suitability of each of the plurality of initial reference images for assisting in restoring the target image; an image registration module, configured to transform each of the plurality of reference images into a registered image having the same viewing angle as the target image; and an image combination module, configured to combine information of each registered image to restore the target image.

[0011] By adopting the technical solution provided by the present invention, the target image can be accurately and reasonably reconstructed with a small amount of calculation, and the restoration result has an accurate structure and natural texture.

[0012] These and other features and advantages will become apparent by reading the following detailed description and by reference to the associated drawings.It is to be understood that the foregoing general description and the following detailed description are illustrative only and are not restrictive of the aspects of what is claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to understand the manner in which the above features of the present invention are used in detail, the above briefly summarized contents can be described in more detail with reference to various embodiments, some of which are shown in the accompanying drawings. However, it should be noted that the accompanying drawings only show some typical aspects of the present invention and should not be considered to limit its scope, because the description may allow for other equally effective aspects.

[0014] Figure 1 A flow chart illustrating a method for image inpainting according to aspects of the present invention is illustrated.

[0015] Figure 2 A schematic diagram illustrating using a reference image to assist a target image in repairing remaining holes according to various aspects of the present invention is illustrated.

[0016] Figure 3 A block diagram of an apparatus for image restoration according to aspects of the present invention is illustrated.

[0017] Figure 4 A block diagram illustrating an example of a hardware implementation of an apparatus for image restoration according to aspects of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be described in detail below in conjunction with the accompanying drawings, and the features of the present invention will be further revealed in the following specific description.

[0019] As mentioned above, currently, image restoration based on multiple views is mostly based on video frame sequences or combined with depth maps to assist color images for restoration, which has a narrow scope of application. In addition, the user specifies or selects appropriate reference view images using the viewpoint-invariant image search method. The image quality in the selection results is inconsistent, or the similarity is not high enough, making the restoration result less than ideal. In order to automatically obtain the most suitable reference image for restoring the target image, and to have abundant available information to assist in restoring the remaining holes, the present invention proposes a balance rule to measure the applicability of the reference image for assisting the restoration of the target image. Secondly, after combining the reference images to repair the large area of ​​missing areas, if there are still remaining holes to be repaired due to blind spots in the viewing angle or insufficient image information, for the remaining holes, the present invention makes full use of the reference image information and uses the balance rule to select a reference image with the highest applicability to assist in repairing the remaining holes, and finally can reasonably repair the image with large area missing from coarse to fine.

[0020] Figure 1 1 illustrates a flow chart of a method 100 for image restoration according to aspects of the present invention. In some examples, the method 100 may be performed by Figure 3 The device 300 described in Figure 4In some examples, the method 100 may be performed by any suitable device or apparatus for performing the functions or algorithms described below.

[0021] At block 110, the method 100 may include obtaining a target image to be restored and a plurality of initial reference images. In the present invention, the target image to be restored may refer to an image having a damaged region (e.g., a missing region) due to some reason (e.g., a coding error, a transmission error, interference during transmission, etc.). The initial reference images may be images used to restore the target image, which may be sourced from the Internet, may be specified or given by a user, or may be automatically searched on the Internet using a viewpoint-invariant image search method, etc.

[0022] At box 120, method 100 may include selecting multiple reference images from the multiple initial reference images based on the suitability of each of the multiple initial reference images to assist in repairing the target image. In the present invention, suitability can represent the degree to which effective information can be provided for repairing the damaged area of ​​the target image. The higher the suitability, the richer the effective information that can be provided. In one embodiment, the suitability of all initial reference images can be sorted, and a group of reference images with the highest suitability can be selected from them for repairing the target image. For example, the five reference images with the highest suitability can be selected from the initial reference images according to the sorting of suitability for repairing the target image. Of course, less than or more than five reference images can also be selected for repairing the target image.

[0023] In one embodiment, the suitability of each initial reference image for assisting in repairing a target image can be determined based on the image quality of the initial reference image, the similarity between the initial reference image and the target image, and the distance between the initial reference image and the target image. The image quality of the reference image can ensure the accuracy of feature point selection when registering images, and a blurred image will weaken the quality of the repaired image; the similarity with the target image can ensure the texture consistency of the final repair effect; and a smaller image distance can increase the number of feature points that can be matched when registering images, thereby obtaining a higher registration accuracy. In one example, the reference image f can be evaluated by the following formula: i Assisted restoration of the target image f t The applicability of S(f i , f t ):

[0024]

[0025] Where s(f i , f t ) represents the reference image f i and the target image f tThe similarity between the two images (for example, the similarity can be expressed as the reference image f i and the target image f t The Euclidean distance between i , f t ) represents the reference image f i and the target image f t The distance between (for example, the distance can be obtained by using the reference image f i and the target image f t The rotation distance between i ) and q(f t ) represent the reference images f i and the target image f t The image quality of the target image is determined by its ratio, where w 1 and w 2 Represents the weight parameter, which is a positive number, w 1 +w 2 =1 (for example, w 1 Take 0.4, w 2 Take 0.6). For the evaluation of image quality, a gradient-based calculation method can be used. The gradient operator is used to calculate the gradient values ​​of all pixels, and the proportion of relatively large gradients is statistically calculated. A higher proportion indicates that the image has a large number of display edges, which means that the image has higher clarity and image quality.

[0026] At box 130, method 100 may include transforming each of the selected multiple reference images into a registered image having the same viewing angle as the target image. As known to those skilled in the art, the purpose of image registration is to compare or fuse images acquired under different conditions for the same object (for example, the images may come from different acquisition devices, taken at different times, different shooting angles, etc.). Specifically, for two images in a set of image data sets, one image is mapped to another image by finding a spatial transformation so that the points corresponding to the same spatial position in the two images correspond one to one, thereby achieving the purpose of information fusion. The operation at box 130 may be implemented using a known image registration method (for example, based on grayscale information method, transform domain method, feature-based method, etc.). In one embodiment, the operation at box 130 may include: segmenting each reference image into multiple planar scenes; finding feature matching points using a scale-invariant feature transform SIFT detector; and using multiple local homographies in the multiple planar scenes to transform the reference image into the same viewing angle as the target image to obtain a registered image. In one example, segmenting each reference image into multiple planar scenes can be achieved by simply segmenting the image based on a grid. In this way, the entire image can be well registered without severe distortion and obvious edge gaps.

[0027] At block 140, method 100 may include combining information of each registered image to repair the target image. In one example, a pixel value (e.g., average pixel value, weighted average pixel value, etc.) of the registered image corresponding to a specific position in the damaged region of the target image may be added to the specific position in the damaged region of the target image to repair the target image. In another example, known image fusion techniques may be used to combine information of each registered image to repair the target image.

[0028] In another example, the information of each registered image can be combined according to the Markov random field MRF to repair the target image. For example, each registered image is set as a label L, and a label L is assigned to each pixel p at each position in the damaged area. p , label L p Corresponds to the registered image assigned to pixel p. Using the data cost term E 1 (p,L p ) to indicate that L p The cost assigned to pixel p, while the smoothing cost term E 2 (p, q, L p , L q ) is to assign labels L to adjacent pixels p and q respectively p and L q The cost function is as follows:

[0029]

[0030] Where φ represents the set of all pixels in the mask area of ​​the target area, (p, q) is a pair of adjacent pixels. The 4-neighborhood system is used here, q represents the pixel point in the 4-neighborhood system of p, and ζ represents the set of all these pixel pairs in the mask area. The adjacent pixels include unmasked pixels, which are also considered as contributions to the total energy because it can be used as a constraint on the masked pixels through the smoothing cost term.

[0031] The data cost item needs to meet two objectives. One is that the pixels around the mask area of ​​the target image should be similar to those of the original target image. The other is that the pixels inside the mask area should be similar to the unoccluded background information to avoid inserting new occlusions into the image. The purpose of the smoothing cost item is to encourage the merging of a few large areas rather than multiple small areas, ensuring that the region boundaries from different reference view images appear where the structural features are not obvious.

[0032] The cost function of each registered image is calculated for each pixel p at each position in the damaged area, and then the registered image corresponding to the minimum cost function among these cost functions is used to repair the pixel p at this position, that is, the pixel value of the registered image corresponding to the minimum cost function at this position is used as the pixel value of the pixel p at the corresponding position of the target image. In this way, the damaged area of ​​the target image is repaired pixel by pixel, thereby obtaining a reconstructed target image.

[0033] Since the brightness of the same scene in different images may vary to some extent, and all pixels in the missing area of ​​the target image come from the planar scene area of ​​the reference image, there may be a brightness inconsistency between the restored missing area and the entire target image, i.e., "ghosting". Therefore, in an optional step of method 100, after combining the information of each registered image to restore the target image, the Poisson image fusion technique may be further used to process the target image, so as to deal with the "ghosting phenomenon".

[0034] Generally speaking, after executing the above steps of method 100, most areas of the target image have been repaired. However, in some cases, there may still be a few remaining holes that need to be repaired. To this end, in an optional step of method 100, after combining the information of each registered image to repair the target image, a reference image with the highest applicability among the multiple reference images can be further selected, and then the target image can be further repaired using a texture synthesis method based on the reference image with the highest applicability. In one embodiment, first, the missing boundary of the remaining area of ​​the target image is set to δφ, the missing area is set to φ, and an arbitrary pixel point p on the boundary is taken, with point p as the center, to obtain an image block Ψ p , the confidence term C(p) and data term D(p) of the image block are defined as:

[0035]

[0036]

[0037] Among them, |Ψ p | is the image block Ψ p The total number of pixels, is the direction of the iso-illuminance line at point p, n p is the normal vector of point p, and α is the normalization factor. Since the confidence will approach 0 sharply with the increase of iteration times, the data item is also 0 when the iso-illuminance line is perpendicular to the normal vector. In order to prevent the priority from being 0 when either of these two situations occurs, the image block Ψ can be p The priority P(p) is defined as:

[0038] P(p)=αC(p)+βD(p) stα+β=1

[0039] Calculate the priority of the image blocks corresponding to all pixels on the boundary of the missing area, take the block with the largest priority as the block to be repaired, and then use the sum of squared differences (SSD) in the known information area of ​​the target image and the reference image after registration to find the image block that best matches the block to be repaired as the pixel value corresponding to the missing area, such as Figure 2 As shown, x is the block to be repaired, x i is the matching block to search for.

[0040] After selecting the image block with the highest matching degree, the value of the missing pixel position of the corresponding block to be repaired is copied to the block to be repaired as its pixel value. Next, the edge and confidence of its damaged area are updated to further realize the repair of the missing area of ​​the entire image.

[0041] Figure 3 A block diagram of an apparatus 300 for image restoration according to various aspects of the present invention is illustrated. The apparatus 300 may include: an image acquisition module configured to obtain a target image to be restored and a plurality of initial reference images; a reference image selection module configured to select a plurality of reference images from the plurality of initial reference images based on the suitability of each of the plurality of initial reference images to assist in restoring the target image; an image registration module configured to transform each of the plurality of reference images into a registered image having the same viewing angle as the target image; and an image combination module configured to combine information of each registered image to restore the target image.

[0042] Figure 4 A block diagram illustrating an example of a hardware implementation of an apparatus 400 for image restoration according to various aspects of the present invention is illustrated. The apparatus 400 may be implemented using a processing system 414 including one or more processors 404. Examples of processors 404 include microprocessors, microcontrollers, digital signal processors (DSPs), field programmable gate arrays (FPGAs), programmable logic devices (PLDs), state machines, gating logic, discrete hardware circuits, and other suitable hardware configured to perform various functionalities described throughout the present disclosure. In various examples, the apparatus 400 may be configured to perform any one or more of the functions described herein. That is, the processor 404 as utilized in the apparatus 400 may be used to implement the above referenced Figure 1 Describe the method.

[0043] In this example, the processing system 414 can be implemented with a bus architecture generally represented by bus 402. Depending on the specific application and overall design constraints of the processing system 414, the bus 402 may include any number of interconnecting buses and bridges. The bus 402 communicatively couples various circuits including one or more processors (generally represented by processor 404), memory 405, and computer-readable media (generally represented by computer-readable media 406). The bus 402 may also link various other circuits, such as timing sources, peripherals, regulators, and power management circuits, which are well known in the art and therefore will not be further described. The bus interface 408 provides an interface between the bus 402 and the transceiver 410. The transceiver 410 provides a communication interface or device for communicating with various other devices on a transmission medium. Depending on the characteristics of the device, a user interface 412 (e.g., a keypad, a display, a speaker, a microphone, a joystick) may also be provided. Of course, such a user interface 412 is optional and may be omitted in some examples.

[0044] In some aspects, the processor 404 may include a processor configured to obtain a target image to be repaired and multiple initial reference images; select multiple reference images from the multiple initial reference images based on the suitability of each of the multiple initial reference images to assist in repairing the target image; transform each of the multiple reference images into a registered image having the same perspective as the target image; and combine information of each registered image to repair the target image.

[0045] The processor 404 is responsible for managing the bus 402 and general processing, including the execution of software stored on the computer-readable medium 406. The software, when executed by the processor 404, causes the processing system 414 to perform the various functions described for any particular device. The computer-readable medium 406 and memory 405 may also be used to store data that is manipulated by the processor 404 when executing the software.

[0046] One or more processors 404 in the processing system can execute software. Software should be broadly interpreted to mean instructions, instruction sets, codes, code segments, program codes, programs, subroutines, software modules, applications, software applications, software packages, routines, subroutines, objects, executables, threads of execution, procedures, functions, etc., whether described in software, firmware, middleware, microcode, hardware description language, or other terms. The software may reside on a computer-readable medium 406. The computer-readable medium 406 may be a non-transitory computer-readable medium. As examples, non-transitory computer-readable media include magnetic storage devices (e.g., hard disks, floppy disks, magnetic tapes), optical disks (e.g., compact disks (CDs) or digital versatile disks (DVDs)), smart cards, flash memory devices (e.g., card, stick, or key-type drives), random access memory (RAM), read-only memory (ROM), programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), registers, removable disks, and any other suitable medium for storing software and / or instructions that can be accessed and read by a computer. Computer-readable medium 406 may reside in processing system 414, external to processing system 414, or distributed across multiple entities including processing system 414. Computer-readable medium 406 may be implemented in a computer program product. As an example, a computer program product may include a computer-readable medium in packaging materials. Those skilled in the art will recognize how to best implement the described functionality presented throughout this disclosure depending on the specific application and the overall design constraints imposed on the overall system.

[0047] In one or more examples, the computer-readable storage medium 406 may include software configured for various functions, including, for example, functions for image restoration. The software may include instructions that configure the processing system 414 to perform the reference Figure 1 One or more functions described.

[0048] Various aspects of the present invention have been described in detail above. In general, the present invention has a wide range of applications. Ordinary color images can be used as algorithm input to obtain good restoration effects, and can repair missing images of larger areas. A balance rule is used to automatically select a group of reference images with the highest applicability, thereby reducing human interference and making the auxiliary information for target image restoration more available. In addition, for the restoration of remaining holes, only one of the reference images with the highest applicability is used to assist in the restoration of the target image, which not only reduces the amount of calculation, but also accurately and reasonably reconstructs the target image, and the restoration result has an accurate structure and a natural texture.

[0049] It will be appreciated by those skilled in the art that various embodiments of the present invention may be provided as methods, devices, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) in which computer executable program code is stored.

[0050] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices, systems and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0051] Although various aspects of the present invention have been described so far with reference to the accompanying drawings, the above-mentioned methods, systems and devices are only examples, and the scope of the present invention is not limited to these aspects, but is limited only by the attached claims and their equivalents. Various components may be omitted or may be replaced by equivalent components. In addition, the steps may be implemented in an order different from the order described in the present invention. In addition, various components may be combined in various ways. It is also important that, as technology develops, many of the components described may be replaced by equivalent components that appear later. Various modifications to the present disclosure will be apparent to those skilled in the art, and the universal principles defined herein may be applied to other variations without departing from the scope of the present disclosure. Thus, the present disclosure is not limited to the examples and designs described herein, but should be granted the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for image restoration, comprising: Obtaining a target image to be restored and multiple initial reference images; selecting a plurality of reference images from the plurality of initial reference images based on the suitability of each of the plurality of initial reference images for assisting in repairing the target image; transforming each of the plurality of reference images into a registered image having the same viewing angle as the target image; as well as combining information from each of the registered images to inpaint the target image, The suitability of each initial reference image for assisting in repairing the target image is determined by the following formula based on the image quality of the initial reference image, the similarity between the initial reference image and the target image, and the distance between the initial reference image and the target image: Where S(f i , f t ) represents the reference image f i Assisted restoration of the target image f t The applicability of s(f i , f t ) represents the reference image f i and the target image f t The similarity between them, d(f i , f t ) represents the reference image f i and the target image f t The distance between i ) and q(f t ) represent the reference images f i and the target image f t image quality, w1 and w2 represent weight parameters, which are positive numbers, w1+w2=1.

2. The method according to claim 1, characterized in that The multiple reference images include 5 reference images, and the 5 reference images are 5 initial reference images with the highest applicability among the multiple initial reference images.

3. The method according to claim 1, characterized in that Combining the information of each registered image to inpaint the target image is performed according to a Markov random field.

4. The method according to claim 1, characterized in that Transforming each of the multiple reference images into a registered image having the same viewing angle as the target image includes: segmenting each reference image into multiple planar scenes; finding feature matching points using a scale-invariant feature transform (SIFT) detector; and using multiple local homographies in the multiple planar scenes to transform the reference image into the same viewing angle as the target image to obtain a registered image.

5. The method according to claim 1, characterized in that The method further includes further using a Poisson image fusion technique to process the target image after combining information of each registered image to repair the target image.

6. The method according to claim 1, characterized in that The method further includes: further selecting a reference image with the highest applicability among the multiple reference images after combining information of each registered image to restore the target image.

7. The method according to claim 6, characterized in that The method further comprises: based on the reference image with the highest applicability, using a texture synthesis method to further repair the target image.

8. A device for image restoration, comprising: Memory; as well as a processor coupled to the memory, wherein the processor is configured to: Obtaining a target image to be restored and multiple initial reference images; selecting a plurality of reference images from the plurality of initial reference images based on the suitability of each of the plurality of initial reference images for assisting in repairing the target image; transforming each of the plurality of reference images into a registered image having the same viewing angle as the target image; as well as combining information from each of the registered images to inpaint the target image, The suitability of each initial reference image for assisting in repairing the target image is determined by the following formula based on the image quality of the initial reference image, the similarity between the initial reference image and the target image, and the distance between the initial reference image and the target image: Where S(f i , f t ) represents the reference image f i Auxiliary repair target image f t The applicability of s(f i , f t ) represents the reference image f i and the target image f t The similarity between them, d(f i , f t ) represents the reference image f i and the target image f t The distance between i ) and q(f t ) represent the reference images f i and the target image f t image quality, w1 and w2 represent weight parameters, which are positive numbers, w1+w2=1.

9. A device for image restoration, comprising: An image acquisition module, configured to obtain a target image to be restored and a plurality of initial reference images; a reference image selection module configured to select a plurality of reference images from the plurality of initial reference images based on the suitability of each of the plurality of initial reference images for assisting in restoring the target image; an image registration module configured to transform each of the plurality of reference images into a registered image having the same viewing angle as the target image; as well as an image combination module configured to combine information of each registered image to inpaint the target image, The suitability of each initial reference image for assisting in repairing the target image is determined by the following formula based on the image quality of the initial reference image, the similarity between the initial reference image and the target image, and the distance between the initial reference image and the target image: Where S(f i , f t ) represents the reference image f i Auxiliary repair target image f t The applicability of s(f i , f t ) represents the reference image f i and the target image f t The similarity between them, d(f i , f t ) represents the reference image f i and the target image f t The distance between i ) and q(f t ) represent the reference images f i and the target image f t image quality, w1 and w2 represent weight parameters, which are positive numbers, w1+w2=1.

Images