CN106709945A - Super-resolution image quality evaluation method - Google Patents

Super-resolution image quality evaluation method Download PDF

Info

Publication number
CN106709945A
CN106709945A CN201710011972.4A CN201710011972A CN106709945A CN 106709945 A CN106709945 A CN 106709945A CN 201710011972 A CN201710011972 A CN 201710011972A CN 106709945 A CN106709945 A CN 106709945A
Authority
CN
China
Prior art keywords
resolution image
image
super
low
block
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710011972.4A
Other languages
Chinese (zh)
Other versions
CN106709945B (en
Inventor
方玉明
鄢杰斌
诸汉炜
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201710011972.4A priority Critical patent/CN106709945B/en
Publication of CN106709945A publication Critical patent/CN106709945A/en
Application granted granted Critical
Publication of CN106709945B publication Critical patent/CN106709945B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20048—Transform domain processing
    • G06T2207/20052—Discrete cosine transform [DCT]

Landscapes

  • Complex Calculations (AREA)

Abstract

The present invention relates to a super-resolution image quality evaluation method. The method comprises: employing a pixel corresponding relation between a low-resolution image and a super-resolution image to obtain a corresponding relation between local image blocks, respectively solving the energy and the texture information of the low-resolution image and the super-resolution image, and obtaining the vision similarity between the changing of the two features. The final super-resolution image quality is obtained through evaluation of the vision similarity of the local image blocks of the low-resolution image and the super-resolution image. The experiment result shows good effects in the quality evaluation of the super-resolution image.

Description

A kind of quality evaluating method for super-resolution image
Technical field
The present invention relates to a kind of super-resolution image quality evaluating method.Belong to multimedia technology field, particularly belong to number Word image and digital technical field of video processing.
Background technology
The super-resolution rebuilding technology of image is widely applied to every field, including Medical Image Processing, infrared imaging, Face/iris recognition, picture editting etc..The purpose is in the case of given one or several low-resolution images, reconstruct Obtain a super-resolution image.In the past more than ten years, researcher proposes many image super-resolution rebuildings and calculates Method, according to the number of low-resolution image, image super-resolution rebuilding algorithm can be divided into the super-resolution rebuilding of multiple image The super-resolution rebuilding algorithm of algorithm and single image
For the super-resolution rebuilding of single image, many different algorithms are presently, there are.Based on traditional interpolation side Method is all by a basic function, such as bilinear interpolation, bicubic interpolation and nearest neighbor algorithm, it is generally the case that these Method is all easy and effective, but because the picture element interpolation of these algorithms is operated, edge and high-frequency region can be caused aliasing occur and imitated Should be with fuzzy distortion.
Because these are based on the limitation of the algorithm of interpolation, the more excellent super-resolution algorithms of many performances are proposed in succession.Base It is on the premise of priori is possessed, to go to obtain super-resolution figure by a regularization loss function in the algorithm rebuild Picture, the priori for being used includes marginal information, sparse prior.Algorithm based on sample learning is by learning one low point Mapping function between resolution image and the image block of super-resolution image, reconstructs super-resolution image.These image block samples Can be obtained by paths such as image, external data base, the blended data sources of input, the learning method of mapping function includes supporting Vector regression, dictionary rarefaction representation and deep learning etc..
For above mentioned Image Super-resolution Reconstruction algorithm, its performance be mainly by small-scale subjective experiment come Checking, quantitative evaluation super resolution image matter quantifier elimination is now also little.But, subjective experiment not only consumes wealth, also takes, And cannot be embedded in super-resolution algorithms goes work to optimize, it is therefore desirable to which the evaluating objective quality studied for super-resolution image is calculated Method.
The content of the invention
In order to make up deficiency at present for super-resolution image Environmental Evaluation Model research, the present invention is based on super-resolution Image quality evaluation model proposes a kind of quality evaluating method for super-resolution image.The present invention passes through following technical side Case is realized：
A kind of quality evaluating method for super-resolution image, it is characterised in that comprise the following steps：
A. first with the pixel corresponding relation between low-resolution image and super-resolution image, topography's block is obtained Between corresponding relation；
B. low-resolution image and the image block corresponding relation of super-resolution image according to gained, seek correspondence image respectively The energy feature and textural characteristics of block, according to the energy feature and textural characteristics tried to achieve, calculate the energy between image block Information change and textural characteristics difference；
C. energy information change between the correspondence image block that comprehensive low-resolution image and super resolution image are included with And textural characteristics difference, obtain super-resolution image quality.
Further, the energy feature and textural characteristics are the visual signature of image, and visual signature passes through discrete cosine Conversion is obtained, and calculation is as follows：
Wherein C is discrete cosine transformation matrix, and N represents tile size, C1And C2N values be respectively equal to f line number and row Number, f represents an image block, and F represents the matrix obtained by discrete cosine transform by f, wherein first in matrix F element F (1,1) is DC coefficients, represents the energy feature of image, and the other elements in matrix F except first element F (1,1) outward are AC systems Number, represents the textural characteristics of image.
Further, the energy variation between image is calculated by the DC coefficients between image block, it is concretely comprised the following steps：
A. low-resolution image is divided into n blocks, the image block set of low-resolution image is expressed as b, b={ b1,b2…, bn}；
B. super-resolution image is divided into n blocks, the image block set of super-resolution image is expressed as b ', b '={ b '1, b′2…,b′n}；
C. the image block energy feature of low-resolution image and high-definition picture is obtained, is calculated with formula (1) and formula (2) To DC coefficients represent image block energy feature, the DC coefficients being calculated for the image block included in set b are by collecting Close D to represent, D={ D1,D2…,Dn, block b1Corresponding DC coefficients are D1, block bnCorresponding DC coefficients are Dn；For set b ' institutes Comprising the DC coefficients that are calculated of image block by set D ' expressions, D '={ D '1,D′2…,D′n, i.e. block b '1Corresponding DC systems Number is D '1, block b 'nCorresponding DC coefficients are D 'n；
D. the energy variation of low-resolution image and super resolution image, low-resolution image and super-resolution image pair are obtained Energy variation between the image block answered can be calculated by formula (3)：
Wherein, bkWith b 'kRepresent low-resolution image and corresponding two image blocks of super-resolution image, C1=6.5025,WithThe average energy value of respective image block is represented respectively, and m is representedWithAverage, computational methods are as follows:
DkRepresent low-resolution image block bkCorresponding DC coefficients, SkRepresent image block bkSize；D′kRepresent super-resolution Image block b 'kCorresponding DC coefficients, S 'kRepresent image block b 'kSize, k ∈ { 1,2 ..., n },
The energy variation F of low-resolution image and super resolution imagee(LR, HR) can be calculated as follows：
Wherein bk∈{b1,b2…bn}；b′k∈{b′1,b′2…b′n, LR represents low-resolution image, and HR represents super-resolution Rate image.
Further, obtain low-resolution image and surpass by what the textural characteristics difference between corresponding image block was calculated Textural characteristics difference between image in different resolution, it is comprised the concrete steps that：
A. for low-resolution image in a size be Nb*NbImage block, by formula (1) and formula (2) discrete cosine Obtained after conversionIndividual AC coefficients, are represented with vectorial A,
B. for super-resolution image in corresponding size be Nb′*Nb′Image block, can be with by discrete cosine transform ObtainIndividual AC coefficients, are represented with vectorial A ',
C. the textural characteristics difference between low-resolution image and the correspondence image block of super-resolution image is by the AC that obtains The average and variance of coefficient are represented, calculated by formula (7)：
Wherein C2=6.5025, C3=58.5225, mAAnd mA′Vector A and vector A ' averages, d are represented respectivelyAAnd dA′Respectively Represent the variance of vector A and vector A ', bkRepresent k-th image block of low resolution figure, b 'kRepresent the kth of super-resolution image Individual image block, ft(bk,b′k) represent the textural characteristics of low-resolution image k-th image block corresponding with super-resolution image Difference；
D. the textural characteristics difference between low-resolution image and super-resolution image, low-resolution image and oversubscription are obtained Textural characteristics difference between resolution image is superimposed by the textural characteristics difference between correspondence image block and obtained, can be by formula (8) Calculate：
Wherein, LR represents low-resolution image, and HR represents super-resolution image, bk∈{b1,b2…bn, represent low resolution Rate image is divided into { b1,b2…bnThis n image block；b′k∈{b′1,b′2…b′n, represent that super-resolution image is divided into { b '1, b′2…b′nThis n image block；Ft(LR, HR) represents that the textural characteristics between low-resolution image and super-resolution image are poor It is different.
Further, described energy and textural characteristics are combined, calculates mass fraction to predict the matter of super resolution image Amount, the mass fraction can be calculated by formula (9)：
Q=Fe(LR,HR)Ft(LR, HR) (9) wherein, Fe(LR, HR) and Ft(LR, HR) is respectively low-resolution image Energy feature and textural characteristics difference value and super-resolution image between, Q represent super-resolution image fraction.
Technique effect of the invention：The DC coefficients that are obtained by discrete cosine transform respectively and AC coefficients calculate image The difference of energy information and texture information, the quality of final super-resolution image is calculated by the difference of energy information and texture information Obtain.The objective quality scores for obtaining are higher with subjective scores correlation.
Brief description of the drawings
Fig. 1 is algorithm frame figure of the invention.
Specific embodiment
Below in conjunction with the accompanying drawings, technical scheme is described in further detail.
Process of the invention is as shown in figure 1, it includes step：
Step 1：Set up topography's block corresponding relation.According to low-resolution image and the local location of super-resolution image Relation, obtains the image block corresponding relation of low-resolution image and super-resolution image.The wherein big I of low-resolution image I*j is expressed as, i represents the line number of low-resolution image, and j represents the columns of low-resolution image；The size of super-resolution image It is represented by (k*i) * (k*j), k>1, the k*i line number for representing super-resolution image, k*j represents the columns of super resolution image.Take An image block (1 in low-resolution image:i1,1:j1), then corresponding to the image block (1 in super resolution image:k*i1,1:k* j1)。(1:i1,1:j1) represent low-resolution image in size be i1*j1Image block, the value of the pixel abscissa of image block Scope is (1, i1), the span of ordinate is (1, j1)；(1:k*i1,1:k*j1) represent super-resolution image in size be (k*i1)*(k*j1) image block, k>1, the span of image block pixel abscissa is (1, k*i1), the value of ordinate Scope is (1, k*j1)。
Step 2：Extract feature.The DCT coefficient that is obtained by discrete cosine transform extracts feature：Two energy of image block Amount and textural characteristics are all extracted by DCT coefficient, so as to set up Environmental Evaluation Model.DCT coefficient is by DC coefficients and AC coefficient sets Into.In each block, DC coefficient tables show the energy information of image block, and AC coefficients represent texture features in block.Every image Energy and textural characteristics represented with DC coefficients and AC coefficients respectively.As formula (1 '), (2 ') represent：
DC={ dc1,dc2,dc3,…dcn} (1′)
ACn={ ac1,ac2,ac3,…acn(2 ') wherein DC represents image DC coefficient vectors, i.e., one image can be by n Individual DC coefficients represent that an image block corresponds to a DC coefficient, dcnThe DC coefficients of n-th image block are represented respectively.ACnRepresent the The n AC coefficient vector of image block, note n-th image block size is N*N, then the number of the AC coefficients for including is N2-1.I.e. one Image can be divided into n blocks, and each block size is N*N, each image block by the available 1 DC coefficient of discrete cosine transform and N2- 1 AC coefficient.
Step 3：Calculate feature difference.Capacity volume variance between low-resolution image and super-resolution image by
The mean value computation of DC coefficients is obtained.Textural characteristics difference between low-resolution image and super-resolution image can be by The average and variance of the AC coefficients for obtaining are represented.For the low-resolution image and corresponding super-resolution image that give, Discrete cosine transform is carried out to low-resolution image and super-resolution image, DC coefficient vectors, the DC of low-resolution image is obtained Coefficient vector is represented that the DC coefficient vectors of super-resolution image are by D ' expressions by D.Then low-resolution image and super resolution image Capacity volume variance can be drawn by formula (3 ')：
Wherein, bkWith b 'kRepresent low-resolution image and corresponding two image blocks of super-resolution image.C1=6.5025,WithThe average energy value of respective image block is represented respectively, and m is represented and taken averaging operation.Low-resolution image and super-resolution The energy variation F of imagee(LR, HR) can be calculated as follows：
Wherein bk∈{b1,b2…bn}；b′k∈{b′1,b′2…b′n, LR represents low-resolution image, and HR represents super-resolution Rate image.
Textural characteristics difference in super-resolution image quality evaluating method, is poor by the textural characteristics of correspondence image block Different superposition is obtained, and correspondence image block textural characteristics difference is calculated by the AC coefficients obtained by discrete cosine transform.It is right A size in low-resolution image is Nb*NbImage block, by being obtained after discrete cosine transformIndividual AC systems Number, is represented with vectorial A,A corresponding image block N in for super-resolution imageb′, warp Crossing discrete cosine transform can obtainIndividual AC coefficients, are represented with vectorial A ',Low point Textural characteristics difference between resolution image and super-resolution image correspondence image block can be by the average of AC coefficients for obtaining and side Difference is represented.It is calculated by formula (5 ')：
Wherein C2=6.5025, C3=58.5225.mAAnd mA′Vector A and vector A ' averages, accordingly, d are represented respectivelyAWith dA′The variance of vector A and vector A ' is represented respectively.bkRepresent k-th image block of low resolution figure, b 'kRepresent super-resolution figure K-th image block of picture.ft(bk,b′k) then represent low-resolution image k-th image block corresponding with super-resolution image Textural characteristics difference.Ft(LR, HR) represents the textural characteristics difference between low-resolution image and super-resolution image, as Textural characteristics difference sum between correspondence image block.Can be calculated by formula (6 ')：
Wherein, LR represents low-resolution image, and HR represents super-resolution image.bk∈{b1,b2…bn, represent low resolution Rate image is divided into b1,b2…bnThis n image block；b′k∈{b′1,b′2…b′n, represent that super-resolution image is divided into b '1,b′2… b′nThis n image block；Ft(LR, HR) represents the textural characteristics difference between low-resolution image and super-resolution image, as Textural characteristics difference sum between correspondence image block.
Step 4：Calculate mass fraction.Super-resolution image quality is by combining low-resolution image and super-resolution image Energy feature and textural characteristics difference, give equal weight, assessment is obtained.Energy feature change meeting in super-resolution image Image vision information is caused to be degenerated, and the change of textural characteristics can cause the distortion of image high-frequency region.Super-resolution image Quality evaluating method combines energy and textural characteristics to predict the mass fraction of super resolution image.Mass fraction can be by formula (7 ') are calculated：
Q=Fe (LR, HR) Ft (LR, HR) (7 ')
The energy feature and textural characteristics that wherein Fe and Ft are respectively between low-resolution image and super-resolution image are poor Different value.
As shown in table 1,2, experiment proves that super-resolution image Environmental Evaluation Model proposed by the present invention is substantially better than at present Other models.Compared by the full reference mass evaluation model common with 7 kinds, the index of assessment：PLCC(Pearson Linear correlation coefficient) and SROC (Spearman correlation coefficient).PLCC Coefficient is that, for measuring two correlations between variable X and Y, X, Y are respectively the subjectivities of image in image quality evaluation Fraction and objective score, span is between [- 1 ,+1].Spearman rank correlation coefficient is used for estimating two variable Xs, Y Between correlation, the wherein correlation between variable can be described using monotonic function.If two the two of variable-value Do not exist two elements of identical in set, then, when one of variable can be expressed as the good of another variable During monotonic function (variation tendency of two variables is identical), the ρ between two variables can reach+1 or -1.Coefficient correlation is used To weigh the linear correlation degree of objective evaluation and subjective assessment, coefficient correlation is bigger, then the performance of evaluating objective quality model It is then better.
Table 1
Components Energy Change Texture Variation Proposed
KRCC 0.4092 0.4996 0.5885
SRCC 0.6001 0.6958 0.8035
Table 2
PSNR SSIM MSSIM NQM VIF MAD NSS-SR Proposed
KRCC 0.3296 0.4502 0.5325 0.5703 0.2786 0.5523 0.0917 0.5885
SRCC 0.4760 0.6203 0.7096 0.7632 0.5226 0.7363 0.1343 0.8035
The above is detailed description of the invention, any equal change done according to scope of the presently claimed invention With modification, it is all scope of the presently claimed invention and is covered.

Claims (5)

1. a kind of quality evaluating method for super-resolution image, it is characterised in that comprise the following steps：
A. first with the pixel corresponding relation between low-resolution image and super-resolution image, between acquisition topography block Corresponding relation；
B. low-resolution image and the image block corresponding relation of super-resolution image according to gained, seek correspondence image block respectively Energy feature and textural characteristics, according to the energy feature and textural characteristics tried to achieve, calculate the energy information between image block Change and textural characteristics difference；
C. energy information change and line between the correspondence image block that comprehensive low-resolution image and super resolution image are included Reason feature difference, obtains super-resolution image quality.
2. the quality evaluating method for super-resolution image according to claim 1, it is characterised in that：The energy is special Textural characteristics of seeking peace are the visual signature of image, and visual signature is obtained by discrete cosine transform, and calculation is as follows：
F = C 1 T · f · C 2 - - - ( 1 )
Wherein C is discrete cosine transformation matrix, and N represents tile size, C1And C2N values be respectively equal to f line number and columns, f Represent an image block, F represents the matrix obtained by discrete cosine transform by f, wherein first in matrix F element F (1, 1) it is DC coefficients, represents the energy feature of image, the other elements in matrix F except first element F (1,1) outward is AC coefficients, Represent the textural characteristics of image.
3. according to claim 2 for super-resolution image quality evaluating method, it is characterised in that：By image block it Between DC coefficients calculate the energy variation between image, it concretely comprises the following steps：
A. low-resolution image is divided into n blocks, the image block set of low-resolution image is expressed as b, b={ b1,b2…,bn}；
B. super-resolution image is divided into n blocks, the image block set of super-resolution image is expressed as b ', b '={ b '1,b′2…, b′n}；
C. the image block energy feature of low-resolution image and high-definition picture, the DC being calculated with formula (1) and (2) are obtained Coefficient represents image block energy feature, and the DC coefficients being calculated for the image block included in set b are by set D tables Show, D={ D1,D2…,Dn, block b1Corresponding DC coefficients are D1, block bnCorresponding DC coefficients are Dn；Included for set b ' The DC coefficients that are calculated of image block by set D ' expressions, D '={ D '1,D′2…,D′n, i.e. block b '1Corresponding DC coefficients are D′1, block b 'nCorresponding DC coefficients are D 'n；
D. the energy variation of low-resolution image and super resolution image is obtained, low-resolution image and super-resolution image are corresponding Energy variation between image block is calculated by formula (3)：
f e ( b k , b k ′ ) = 2 m D k m D k ′ + C 1 m D k 2 + m D k ′ 2 + C 1 - - - ( 3 )
Wherein, bkWith b 'kRepresent low-resolution image and corresponding two image blocks of super-resolution image, C1=6.5025, WithThe average energy value of respective image block is represented respectively, and computational methods are as follows:
m D k = D k S k - - - ( 4 )
m D k ′ = D k ′ S k ′ - - - ( 5 )
DkRepresent low-resolution image block bkCorresponding DC coefficients, SkRepresent image block bkSize；D′kRepresent super-resolution image Block b 'kCorresponding DC coefficients, S 'kRepresent image block b 'kSize, k ∈ { 1,2 ..., n }, low-resolution image and super resolution image Energy variation Fe(LR, HR) is calculated as follows：
F e ( L R , H R ) = Σ ( b k , b k ′ ) f e ( b k , b k ′ ) - - - ( 6 )
Wherein bk∈{b1,b2…bn}；b′k∈{b′1,b′2…b′n, LR represents low-resolution image, and HR represents super-resolution figure Picture.
4. according to Claims 2 or 3 for super-resolution image quality method, it is characterised in that：By corresponding figure As the textural characteristics difference obtained between low-resolution image and super-resolution image that the textural characteristics difference between block is calculated, It is comprised the concrete steps that：
A. for low-resolution image in a size be Nb*NbImage block, by after formula (1) and (2) discrete cosine transform ObtainIndividual AC coefficients, are represented with vectorial A,
B. for super-resolution image in corresponding size be Nb′*Nb' image block, obtained by discrete cosine transformIndividual AC coefficients, are represented with vectorial A ',
C. the textural characteristics difference between low-resolution image and the correspondence image block of super-resolution image is by the AC coefficients that obtain Average and variance represent, calculated by formula (7)：
f t ( b k , b k ′ ) = ( 2 m A m A ′ + C 2 ) ( 2 d A d A ′ + C 3 ) ( m A 2 + m A ′ 2 + C 2 ) ( d A 2 + d A ′ 2 + C 3 ) - - - ( 7 )
Wherein C2=6.5025, C3=58.5225, mAAnd mA′Vector A and vector A ' averages, d are represented respectivelyAAnd dA′Represent respectively The variance of vectorial A and vector A ', bkRepresent k-th image block of low resolution figure, b 'kRepresent k-th figure of super-resolution image As block, ft(bk,b′k) represent that the textural characteristics of low-resolution image k-th image block corresponding with super-resolution image are poor It is different；
D. the textural characteristics difference between low-resolution image and super-resolution image, low-resolution image and super-resolution are obtained Textural characteristics difference between image is superimposed by the textural characteristics difference between correspondence image block and obtained, and is calculated by formula (8)：
F t ( L R , H R ) = Σ ( b k , b k ′ ) f t ( b k , b k ′ ) - - - ( 8 )
Wherein, LR represents low-resolution image, and HR represents super-resolution image, bk∈{b1,b2…bn, represent low-resolution image It is divided into { b1,b2…bnThis n image block；b′k∈{b′1,b′2…b′n, represent that super-resolution image is divided into { b '1,b′2…b ′nThis n image block；Ft(LR, HR) represents the textural characteristics difference between low-resolution image and super-resolution image.
5. according to claim 4 for super-resolution image quality method, it is characterised in that：Will described this energy and line Reason feature is combined, and calculates mass fraction to predict the quality of super resolution image, and the mass fraction is calculated by formula (9)：
Q=Fe(LR,HR)Ft(LR,HR) (9)
Wherein, Fe(LR, HR) and Ft(LR, HR) be respectively energy feature between low-resolution image and super-resolution image and Textural characteristics difference value, Q represents super-resolution image fraction.
CN201710011972.4A 2017-01-09 2017-01-09 A kind of quality evaluating method for super-resolution image Active CN106709945B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710011972.4A CN106709945B (en) 2017-01-09 2017-01-09 A kind of quality evaluating method for super-resolution image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710011972.4A CN106709945B (en) 2017-01-09 2017-01-09 A kind of quality evaluating method for super-resolution image

Publications (2)

Publication Number Publication Date
CN106709945A true CN106709945A (en) 2017-05-24
CN106709945B CN106709945B (en) 2018-03-06

Family

ID=58907934

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710011972.4A Active CN106709945B (en) 2017-01-09 2017-01-09 A kind of quality evaluating method for super-resolution image

Country Status (1)

Country Link
CN (1) CN106709945B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109345525A (en) * 2018-09-27 2019-02-15 方玉明 One kind removing ghost high dynamic range images quality evaluating method
CN109410177A (en) * 2018-09-28 2019-03-01 深圳大学 A kind of image quality analysis method and system of super-resolution image
CN109636727A (en) * 2018-12-17 2019-04-16 辽宁工程技术大学 A kind of super-resolution rebuilding image spatial resolution evaluation method
CN109978840A (en) * 2019-03-11 2019-07-05 太原理工大学 A kind of method of discrimination of the quality containing texture image based on convolutional neural networks
CN109993694A (en) * 2017-12-29 2019-07-09 Tcl集团股份有限公司 A kind of method and device generating super-resolution image
CN111091158A (en) * 2019-12-25 2020-05-01 科大讯飞股份有限公司 Method, device and equipment for classifying image quality of auxiliary images
CN112954356A (en) * 2021-01-27 2021-06-11 西安万像电子科技有限公司 Image transmission processing method and device, storage medium and electronic equipment

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070046666A1 (en) * 2005-08-31 2007-03-01 Yoshiyuki Kokojima Video rendering apparatus and method and program
CN102306374A (en) * 2011-08-30 2012-01-04 西安交通大学 Method for rebuilding super-resolution human face image by position block nonlinear mapping
CN102436655A (en) * 2011-09-02 2012-05-02 清华大学 Super-resolution reconstruction image quality evaluation method based on SVD (singular value decomposition)
CN104123705A (en) * 2014-07-22 2014-10-29 北华大学 Super-resolution reconstructed image quality Contourlet domain evaluation method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070046666A1 (en) * 2005-08-31 2007-03-01 Yoshiyuki Kokojima Video rendering apparatus and method and program
CN102306374A (en) * 2011-08-30 2012-01-04 西安交通大学 Method for rebuilding super-resolution human face image by position block nonlinear mapping
CN102436655A (en) * 2011-09-02 2012-05-02 清华大学 Super-resolution reconstruction image quality evaluation method based on SVD (singular value decomposition)
CN104123705A (en) * 2014-07-22 2014-10-29 北华大学 Super-resolution reconstructed image quality Contourlet domain evaluation method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
沙磊 等: "利用离散余弦变换进行图像压缩的方法", 《成都理工学院学报》 *
黄慧娟 等: "基于SVD的超分辨率重建图像质量无参考评价方法", 《计算机辅助设计与图形学学报》 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109993694A (en) * 2017-12-29 2019-07-09 Tcl集团股份有限公司 A kind of method and device generating super-resolution image
CN109345525A (en) * 2018-09-27 2019-02-15 方玉明 One kind removing ghost high dynamic range images quality evaluating method
CN109345525B (en) * 2018-09-27 2022-03-29 方玉明 Ghost-removed high-dynamic-range image quality evaluation method
CN109410177A (en) * 2018-09-28 2019-03-01 深圳大学 A kind of image quality analysis method and system of super-resolution image
WO2020062901A1 (en) * 2018-09-28 2020-04-02 深圳大学 Method and system for analyzing image quality of super-resolution image
CN109410177B (en) * 2018-09-28 2022-04-01 深圳大学 Image quality analysis method and system for super-resolution image
CN109636727A (en) * 2018-12-17 2019-04-16 辽宁工程技术大学 A kind of super-resolution rebuilding image spatial resolution evaluation method
CN109636727B (en) * 2018-12-17 2022-11-15 辽宁工程技术大学 Super-resolution reconstruction image spatial resolution evaluation method
CN109978840A (en) * 2019-03-11 2019-07-05 太原理工大学 A kind of method of discrimination of the quality containing texture image based on convolutional neural networks
CN111091158A (en) * 2019-12-25 2020-05-01 科大讯飞股份有限公司 Method, device and equipment for classifying image quality of auxiliary images
CN111091158B (en) * 2019-12-25 2024-04-30 科大讯飞股份有限公司 Classification method, device and equipment for image quality of teaching auxiliary image
CN112954356A (en) * 2021-01-27 2021-06-11 西安万像电子科技有限公司 Image transmission processing method and device, storage medium and electronic equipment

Also Published As

Publication number Publication date
CN106709945B (en) 2018-03-06

Similar Documents

Publication Publication Date Title
CN106709945A (en) Super-resolution image quality evaluation method
CN110211045B (en) Super-resolution face image reconstruction method based on SRGAN network
CN106952228A (en) The super resolution ratio reconstruction method of single image based on the non local self-similarity of image
CN105631807B (en) The single-frame image super-resolution reconstruction method chosen based on sparse domain
CN103871041B (en) The image super-resolution reconstructing method built based on cognitive regularization parameter
CN106127688B (en) A kind of super-resolution image reconstruction method and its system
CN106934766A (en) A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation
CN102842115B (en) Based on the compressed sensing image super-resolution rebuilding method of double dictionary study
CN107633486A (en) Structure Magnetic Resonance Image Denoising based on three-dimensional full convolutional neural networks
CN106204449A (en) A kind of single image super resolution ratio reconstruction method based on symmetrical degree of depth network
CN107481236A (en) A kind of quality evaluating method of screen picture
CN107341765A (en) A kind of image super-resolution rebuilding method decomposed based on cartoon texture
CN102243711A (en) Neighbor embedding-based image super-resolution reconstruction method
CN108846800A (en) A kind of non-reference picture quality appraisement method of image super-resolution rebuilding
CN105160667A (en) Blind image quality evaluation method based on combining gradient signal and Laplacian of Gaussian (LOG) signal
CN108053396A (en) A kind of more distorted image quality without with reference to evaluation method
CN104657962A (en) Image super-resolution reconstruction method based on cascading linear regression
CN103020940B (en) Local feature transformation based face super-resolution reconstruction method
CN111626927A (en) Binocular image super-resolution method, system and device adopting parallax constraint
CN104021523A (en) Novel method for image super-resolution amplification based on edge classification
CN105957029A (en) Magnetic resonance image reconstruction method based on tensor dictionary learning
CN116797461A (en) Binocular image super-resolution reconstruction method based on multistage attention-strengthening mechanism
CN104408697A (en) Image super-resolution reconstruction method based on genetic algorithm and regular prior model
CN101609503A (en) People's face oversubscription image processing method based on double-current shape alignment
CN110084750A (en) Single image super-resolution method based on multilayer ridge regression

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant