CN108550146A - A kind of image quality evaluating method based on ROI - Google Patents
A kind of image quality evaluating method based on ROI Download PDFInfo
- Publication number
- CN108550146A CN108550146A CN201810319898.7A CN201810319898A CN108550146A CN 108550146 A CN108550146 A CN 108550146A CN 201810319898 A CN201810319898 A CN 201810319898A CN 108550146 A CN108550146 A CN 108550146A
- Authority
- CN
- China
- Prior art keywords
- reference picture
- image
- gradient
- block
- sub
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A kind of image quality evaluating method based on ROI.Quality for evaluating different distorted images.Step is：(1) image gray processing is handled；(2) reference, the brightness l (x, y) of distorted image correspondence image block, contrast c (x, y) and structure s (x, y) comparison function are calculated；(3) extract gradient information, calculate reference, distorted image gradient map, and calculate corresponding gradient comparison function g (x, y)；(4) four kinds of comparison functions are combined, calculate reference, distorted image correspondence image block GSSIM values, and then obtain local similarity figure；(5) the vision significance model based on phase spectrum is utilized, the ROI figures of reference picture is calculated, further obtains weight map, be weighted processing using weight map portion's similitude figure of playing a game, obtain evaluation of estimate R_GSSIM.The present invention tests on the databases such as LIVE and TID2008, the results showed that this innovatory algorithm not only has good consistency with the subjective perception of human eye, but also computation complexity is relatively low.
Description
Technical field
The present invention relates to a kind of image quality evaluating method being based on ROI (Region of Interest, abbreviation ROI),
The image processing field for belonging to image quality evaluation is mainly used in and carries out quality evaluation, main target to all kinds of distorted images
Objective image quality evaluating method is exactly designed, is allowed to be consistent with the visual perception of people.
Background technology
Image quality evaluation is a meaningful research topic in image processing field.Image quality evaluation is divided into master
See image quality evaluating method and Objective image quality evaluation method.Subjective picture quality evaluation method includes double stimulation damages point
Grade method, double stimulation continuous mass stagings, single stimulation continuous mass staging.For subjective picture quality evaluation method,
Method for objectively evaluating is using more acurrate, extensive.Objective image quality evaluation method can be divided into be commented with reference to evaluation method, half reference entirely
Valence method and without with reference to evaluation method, entirely more mature with reference to evaluation method at present, the most commonly used is mean square errors in numerous methods
Poor method (MSE) and Y-PSNR method (PSNR).They directly count the grey scale pixel value of reference picture and distorted image
It calculates, calculating is simple, meaning is clear, but cannot be consistent well with the subjective feeling of people.With to human visual system
(HVS) what is recognized gos deep into, and certain characteristics that people start with human visual system are evaluated (referring to Pang Quan, Wang Zhenhua, Geng
It is beautiful large etc..Wang et al. is (referring to WANG Zhou, BOVIK A C, SHEIKH HR, et al.Image quality
assessment:from error visibility to structural similarity[J].IEEE Trans on
Image Processing, 2004,13 (4):600-612.；WANG Zhou, BOVIK A C.A universal image
Quality index [J] .IEEE Trans Signal Precessing, 2002,9 (3):It is similar 91-94.) to propose structure
Property theoretical (SSIM), the brightness of reference picture, contrast and structural information are compared, preferable result has been obtained.He is false
If HVS is good at the structural information in extraction scene, image is carried out by evaluating and testing the degree of degeneration of structural information of distorted image
Evaluation, is widely used.Many scholars are improved on the basis of SSIM, if Wang et al. is (referring to WANG
Zhou, SIMONCELLI P, BOVIC A C.Multiscale structural similarity for image
Quality assessment [C] //Proc of the 37th Asilomar Conference on Signals,
Systems and Computers.2003:Multi-scale model similitude (MSSIM) 1398-1402.) is proposed, is compared
The single better result of scale；L i et al. are (referring to LI Chao-feng, BOVIC A C.Three-component
weighted structural similarity index[C]//Proc of SPIE.2009:72420Q-72420Q-9.)
3-SSIM is proposed, the comparison to the brightness of entire image, contrast and structure in SSIM, is changed in edge, texture peace
Skating area domain calculates separately and assigns different weights, obtains final evaluation result；Chen et al. (referring to CHEN Guan-hao,
YANG Chun-ling, XIE Sheng-li.Gradient-basedstructure similarity for image
quality assessment[C]//Proc of International Conference on Image
Processing.2006:2929-2932.) propose the structural similarity (GSSIM) based on gradient, in SSIM to contrast
Comparison with structure is changed to calculate the gradient image of reference picture and distorted image.Wan et al. (referring to YANG Wan,
WU Le-hua, FAN Y, et al.A mothod of image quality assessment based on region of
interest[C]//Proc of the 7th World Congress on Intelligent Control and
Automation.2008:6840-6843.) propose the method based on area-of-interest；Aja-Fernandez et al. (referring to
AJA-FERNANDEZ S, ESTEPAR R S, ALBE ROLA-LOPEZ C, et al.Image quality assessment
based on local variance[C]//Proc of the28th Annual International Conference
on Engineering in Medicine and Biology Society.Berlin:Springer-Verlag, 2006:
4815-4818.) propose the quality evaluating method based on local contrast；The numerous clanging or clanking sound in hole is (referring to the numerous clanging or clanking sound combinations HVS in hole and similar
Property measurement image quality evaluation estimate [J] Journal of Image and Graphics, 2011,16 (7):1184-1191.) propose one kind
The image quality evaluation that HVS and similarity measurement combine is estimated；Leaf contains nanmu et al. (referring to Ye Shengnan, Su Kaina, Xiao's wound
Image quality evaluation [J] electronic letters, vols that the such as cypress are extracted based on structural information, 2008,36 (5):856-861.) propose base
In the image quality evaluating method of structural information extraction.
Invention content
The technology of the present invention solves the problems, such as：Overcome the deficiencies of the prior art and provide a kind of image quality evaluation based on ROI
Method can carry out quality evaluation to the image of various distortions, be carried out in a variety of image data bases such as LIVE2 and TID2008,
Experiment show innovatory algorithm R_GSSIM of the present invention not only with the subjective perception of human eye have good consistency, but also with compared with
Low computation complexity.
Technical solution of the invention：A kind of image quality evaluating method based on ROI, including：
The conversion of step (1), gray-scale map：Convert reference picture and distorted image to gray-scale map respectively, and by gray-scale map
It is divided into M*N sub-block, the operation carried out for reference picture and distorted image in subsequent step is gray scale corresponding for the two
Scheme the operation carried out；
Step (2), the test of SSIM similarities：Similarity survey is carried out to reference picture and distorted image by SSIM measuring systems
Examination, similarity test include：Brightness contrast, contrast comparison and Structure Comparison, respectively by calculate brightness comparison function l (x,
Y), contrast comparison function c (x, y) and structure comparison function s (x, y) is obtained；Assuming that reference picture is X, distorted image is
Y, x, y are respectively corresponding sub-block in image X and Y；Wherein brightness comparison function l (x, y), contrast comparison function c (x, y)
And structure comparison function s (x, y) can pass through brightness contrast module, contrast contrast module, the structure in SSIM measuring systems
Contrast module is realized；
The extraction of step (3), gradient information：The gradient map for calculating separately reference picture and distorted image, according to gradient map
Calculate gradient comparison function g (x, y)；
The information that step (2) and step (3) obtain is merged, merges and obtain reference picture and distortion map by step (4)
The local similarity figure of picture；
The calculating of step (5), objective evaluation value R_GSSIM：It is calculated first with the vision significance based on phase spectrometry
Model, calculates the ROI figures of reference picture, and is divided into M*N sub-block, obtains weight map；Then utilize weight map to step
(4) the local similarity figure obtained is weighted processing, obtains final objective evaluation value R_GSSIM.
Further, in the step (3) the step of the extraction of gradient information, including：
Step (31), the gradient map for calculating separately reference picture and distorted image；
Gradient map is divided into M*N sub-block by step (32)；
Step (33), the gradient comparison function g (x, y) for calculating reference picture gradient map sub-block corresponding with distorted image.
Further, the extraction of the gradient information realizes that process is as follows：
Gradient operator in the step (31) in gradient nomography using improved all directions to Sobel operators：Using
Size be 5 × 5 eight direction templates, eight directions be respectively 0 °, 22.5 °, 45 °, 67.5 °, 90 °, 112.5 °, 135 ° and
157.5°.Gradient operator used be improved all directions to Sobel operators, calculate the gradient map of image, use Sobel operators
Carry out gradient extraction, and then calculate gradient comparison function, obtain respectively reference picture and distorted image two gradient map GM1 and
GM2 and gradient comparison function；
Assume that reference picture is X in the step (33), distorted image Y, x and y are respectively reference picture X and distortion Y
In corresponding sub-block, define reference picture sub-block x and distorted image sub-block y gradient comparison function be g (x, y)；
According to the definition of gradient comparison function, it is known that g (x, y) meets following three characteristics：
(a) comparative：G (x, y)=g (y, x)；
(b) boundedness：g(x,y)≤1；
(c) maximum value uniqueness：G (x, y)=1, and if only if reference picture sub-block x and the complete phases of distorted image sub-block y
Meanwhile value is just 1.
Further, merge in the step (4), merge the local similarity figure for obtaining reference picture and distorted image
Realization process is as follows：
Step (41), since brightness comparison function, contrast ratio are compared with Dentistry numbers, structure comparison function and gradient comparison function
Independently of each other, thus the present invention by brightness comparison function l (x, y), contrast comparison function c (x, y), structure comparison function s (x,
Y) it merged using multiplication mode with gradient comparison function g (x, y), merge and obtain the image matter based on gradient-structure similarity
Measure evaluation model GSSIM (x, y)；
Step (42) repeats the GSSIM values that step (41) calculates each image block, then to the GSSIM of all image blocks
Value is averaging, and obtains the final objective evaluation value MGSSIM (x, y) of distorted image.MGSSIM values are bigger, then show distorted image
More similar to reference picture, i.e. the quality of distorted image is better.
Further, the step of step (5) the objective evaluation value R_GSSIM is calculated include：
Step (51) is first with phase spectrometry (the Phase spectrum of Fourier based on Fourier transformation
Transform, PFT) vision significance computation model, calculate reference picture ROI figure；In order to improve the efficiency of calculating,
In actual calculating process, usually realized using fast two-dimensional fourier transformation FFT, corresponding function be fft2, ifft2,
It is as follows：
(1.1) by Fourier transform by reference picture f (x, y) by space field transformation to frequency domain：
(1.2) the phase spectrum A (u, v) of reference picture is calculated；
(1.3) phase spectrum of reference picture is subjected to inverse Fourier transform, obtains the regarding based on phase spectrum in spatial domain
Feel Saliency maps：SMap (x, y), i.e. ROI scheme；
Secondly, the ROI figures of reference picture are divided into M*N sub-block, obtain weight map；It is the same with SSIM algorithms, utilize M*N
Gauss window sliding piecemeal is carried out to vision significance figure SMap (x, y), if x, y are respectively reference picture and distorted image pair
The sub-block for answering position, with sub-block x, SMap (x, y) sub-block of y same sizes and position is s, and it is W to define the weights at the position,
Obtain weight map；
Step (52), the local similarity figure obtained to step (4) using weight map are weighted processing, obtain final
Objective evaluation value R_GSSIM.Innovatory algorithm of the present invention meets symmetry, and value range is 0 to 1, and value is bigger, then shows distortion map
The quality of picture is better, and value is smaller, then shows that the quality of distorted image is poorer.
The advantages of the present invention over the prior art are that：
(1) present invention incorporates image gradient information first on the basis of former SSIM algorithms, gives a kind of based on gradient
The image quality evaluation algorithm of structural similarity (GSSIM).The algorithm includes not only to be compared using brightness compared with SSIM algorithms
The SSIM similarity informations that function, contrast comparison function, structural information comparison function calculate, it is often more important that increase use
The gradient information that gradient comparison function calculates.It is in LIVE2 image data bases the experimental results showed that, GSSIM algorithm performances are excellent
In SSIM algorithms.
(2) further, the present invention calculates GSSIM using vision significance graph model (i.e. ROI figures) as weighting function
Method is advanced optimized, and a kind of improved image quality evaluation algorithm R_GSSIM based on structural similarity is finally obtained.
It is on the databases such as LIVE and TID2008 the experimental results showed that, innovatory algorithm R_GSSIM of the present invention not only with human eye subjectivity feel
Know with good consistency, and there is lower computation complexity.
Description of the drawings
Fig. 1 is the system assumption diagram of present system；
Fig. 2 is that the SSIM in present system tests systematic realizing program；
Fig. 3 be all directions used of extraction gradient information in present system to Sobel operator convolution masks；
Specific implementation mode
The present invention is described in detail below in conjunction with the accompanying drawings.
As shown in Figure 1, the present invention relates to a kind of image quality evaluating method based on ROI, this method is suitable for numerous mistakes
The quality evaluation of true image, it is in a variety of image data bases such as LIVE2 and TID2008 the experimental results showed that, the present invention improve
Algorithm R_GSSIM not only has good consistency with the subjective perception of human eye, but also has lower computation complexity.It is described
System includes the conversion of gray-scale map, converts reference picture and distorted image to gray-scale map respectively；The test of SSIM similarities (including
Brightness contrast item, contrast comparative run, structural information compare item)；The extraction of gradient information (gradient information compares item)；Merge,
Fusion obtains local similarity figure；The ROI figures that reference picture is calculated using PFT models, are obtained weight map, finally utilize weight map
Processing is weighted to local similarity figure, obtains final objective evaluation value R_GSSIM.This method overcomes to be based on originally
The shortcomings that inaccuracy of SSIM quality evaluating methods, can accurately evaluate all types of distorted images, and accuracy is high,
With lower computation complexity, it is more in line with human-eye visual characteristic, preferably meets the demand of people.
It is entire to realize that process is as follows：
(1) reference picture and distorted image are separately converted to gray-scale map processing, the algorithm of image gray processing mainly have with
It lower 3 kinds, can be achieved, respectively maximum value process, mean value method, weighted average method；
(2) SSIM similarities are tested, by SSIM measuring systems it is found that the measurement of similarity can be by three kinds of contrast module groups
At respectively：Brightness, contrast, structure.Finally this three modularity function is defined, process is as shown in Figure 2；
(3) extraction of gradient information, gradient operator used be improved all directions to Sobel operators, calculate image
Gradient map uses Sobel operators and carries out gradient extraction, and then calculates gradient comparison function, operator template used such as figure (3)；
(4) merge, fusion obtains local similarity figure：Original three functions of SSIM, including brightness comparison function, comparison
Comparison function, structural information comparison function are spent along with acquisition gradient comparison function can since four comparison functions are relatively independent
Merge, permeate a local similarity figure；
(5) acquisition of weight map and R_GSSIM：1. the acquisition of weight map：Utilize the vision significance meter based on phase spectrum
Model is calculated, the ROI figures of reference picture are calculated, and is divided into M*N sub-block, obtains weight map；2. calculating objective evaluation value R_
GSSIM is weighted processing to obtained local similarity figure using weight map, obtains final objective evaluation value.
Above-mentioned each module realizes that function is as follows：
1.SSIM similarities are tested
The analytic process of the module is as shown in Figure 2：
(1) three kinds of modularity functions are defined：
(1.1) firstly, for discrete signal, we are used as the estimation of brightness measurement with average gray：
Brightness contrast function l (x, y) is about μx, μyFunction.
(1.2) known by measuring system and average gray value is removed from signal, for discrete signal x- μxIt can be used
Standard deviation does contrast estimator：
Contrast contrast function c (x, y) is exactly about σx, σyFunction.
(1.3) signal is divided by by the standard deviation of oneself, and Structure Comparison function is just defined asWithLetter
Number.
(1.4) three contrast modules are combined into a complete similarity measure function：
S (x, y)=f (l (x, y), c (x, y), s (x, y)) (3)
S (x, y) should meet following three conditions：
(1.4.1) symmetry：S (x, y)=S (y, x)；
(1.4.2) boundedness:S(x,y)<=1；
(1.4.3) maximum value uniqueness：When x=y, S (x, y)=1.
(2) three contrast functions are defined：
(2.1) brightness contrast function：
Constant C1It is in order to avoid μx 2+μy 2The unstable of system is caused when close to 0.
Particularly, we select C1=(K1L)2, L is gradation of image series, for 8-bit gray-scale maps, L=255, K1<<1
Formula (7) meets above three condition.
(2.2) contrast comparison function：
Constant C2=(K2L)2, and K2<<1.Formula (8) still meets above three condition.
(2.3) Structure Comparison function：
Wherein
(2.4) three combination of function are got up, and SSIM exponential functions are obtained：
SSIM (x, y)=[l (x, y)]α[c(x,y)]β[s(x,y)]γ (8)
Here α, β, γ>0, for adjusting the importance of three intermodules.Reduced form in order to obtain, if α=β=γ,
C3=0.5C2It obtains：
2. the extraction of gradient information
The analytic process of the module is as follows：
Gradient operator used be improved all directions to Sobel operators, the convolution mask of the operator is as shown in Figure 3.
(1) gradient map of image is calculated：The algorithm use size for 5 × 5 eight direction templates.Eight directions are respectively
0 °, 22.5 °, 45 °, 67.5 °, 90 °, 112.5 °, 135 ° and 157.5 °.
If image I, then gradient image Grad_map can be by calculating the greatest gradients that be obtained on eight directions of image I
It is worth to.
WhereinIt is the gradient operator in eight directions,For convolution symbol.
(2) assume that reference picture is X, distorted image Y, x and y are respectively corresponding sub-block in image X and Y, X ' and
Y ' is the gradient image of image X and Y, and x ' and y ' indicate there is same size and position with x and y in gradient image X ' and Y respectively
Image subblock.
(3) gradient comparison function is calculated：Use for reference the structural similarity thought that Zhou Wang et al. are proposed, present invention definition
The gradient comparison function of reference image block x and distorted image block y is：
Wherein λx'And λy'Constant C is added in the mean value of respectively x ' and y '4It is to increase the stability of g (x, y), prevent
There is the case where denominator is zero.
According to the definition of gradient comparison function, it is known that g (x, y) meets following three characteristics：
(a) comparative：G (x, y)=g (y, x)；
(b) boundedness：g(x,y)≤1；
(c) maximum value uniqueness：G (x, y)=1, and if only if reference picture sub-block x and the complete phases of distorted image sub-block y
Meanwhile value is just 1.
3. single merging, fusion obtain local similarity figure
The analytic process of the module is as follows：
(1) since brightness comparison function, contrast comparison function, structure comparison function and gradient comparison function are mutually only
It is vertical, therefore this four are combined together by the present invention by the way of being multiplied, and obtain the image matter based on gradient-structure similarity
Measuring evaluation model is：
GSSIM (x, y)=[l (x, y)]α·[c(x,y)]β·[s(x,y)]γ·[g(x,y)]δ (13)
Wherein, α, beta, gamma, δ are the constant more than zero, for adjusting the significance level of this several part.In actual experiment
In, α=β of the present invention=γ=δ=1.
(2) utilize formula (13) that the GSSIM values of each image block can be calculated, then to all image blocks
GSSIM values are averaging, and obtain the final objective evaluation value of distorted image.
Wherein, GSSIM (xj,yj) be j-th of sub-block similarity value, M be image block total number.MGSSIM values are bigger,
Then show that distorted image is more similar to reference picture, i.e., the quality of distorted image is better.
4. the acquisition of weight map and R_GSSIM
The analytic process of the module is as follows：
The present invention utilizes phase spectrometry (the Phase spectrum of Fourier based on Fourier transformation
Transform, PFT) calculate the vision significance figure (ROI figures) of image.In order to improve the efficiency of calculating, in actual calculating
In the process, it is usually realized using fast two-dimensional fourier transformation FFT, corresponding function is fft2, ifft2.Specific steps are such as
Under：
(1) PFT methods are utilized to calculate the vision significance figure of reference picture, the vision significance based on phase spectrum calculates mould
Steps are as follows for the specific implementation of type：
(1.1) by Fourier transform by image f (x, y) by space field transformation to frequency domain；
F (u, v)=f { f (x, y) } (15)
(1.2) phase spectrum of image is calculated：
A (u, v)=angle (F (u, v)) (16)
(1.3) phase spectrum of image is subjected to inverse Fourier transform, obtains the ROI figures based on phase spectrum in spatial domain：
Wherein, F and F-1Fourier direct transform and the inverse fourier transform for respectively representing image, can be in the hope of by angle ()
The phase spectrum of image is obtained, g (x, y) is a 2-d gaussian filters device,Indicate that convolution algorithm, SMap (x, y) represent ROI figures；
(2) weight is defined:It is the same with SSIM algorithms, using M*N Gauss window to vision significance figure SMap (x, y) into
Row sliding piecemeal, if x, y are respectively the sub-block of reference picture and distorted image corresponding position, with sub-block x, y same sizes and position
SMap (x, the y) sub-block set is s, then it is W that the present invention, which defines the weights at the position,：
Wherein, K is the number of pixel in s, piFor the value in s at ith pixel point, ωiIt is 1.5 M*N for standard deviation
I-th of value after normalization in Gauss weighting windows.
(3) the improved image quality evaluating method index R_GSSIM based on ROI eventually is most obtained:
Wherein WjIndicate the weights of j-th of image subblock, GSSIM (xj,yj) it is similarity values of j-th of Wai as sub-block.M
For the total number of image block.Innovatory algorithm R_GSSIM of the present invention meets symmetry, and value range is 0 to 1, and value is bigger, then shows
The quality of distorted image is better, and value is smaller, then shows that the quality of distorted image is poorer.
Claims (6)
1. a kind of image quality evaluating method based on ROI, it is characterised in that：Including step：Wherein：
The conversion of step (1), gray-scale map：It converts reference picture and distorted image to gray-scale map respectively, and gray-scale map is divided into
M*N sub-block, the operation carried out for reference picture and distorted image in subsequent step be for both corresponding gray-scale map into
Capable operation；
Step (2), the test of SSIM similarities：Similarity test is carried out to reference picture and distorted image by SSIM measuring systems,
Similarity is tested：Brightness contrast, contrast comparison and Structure Comparison, respectively by calculate brightness comparison function l (x, y),
Contrast comparison function c (x, y) and structure comparison function s (x, y) is obtained；Assuming that reference picture is X, distorted image Y, x
It is respectively corresponding sub-block in image X and Y with y；
The extraction of step (3), gradient information：The gradient map for calculating separately reference picture and distorted image, calculates according to gradient map
Gradient comparison function g (x, y)；
The information that step (2) and step (3) obtain is merged, merges and obtain reference picture and distorted image by step (4)
Local similarity figure；
The calculating of step (5), objective evaluation value R_GSSIM：First with the vision significance computation model based on phase spectrometry,
The ROI figures of reference picture are calculated, and are divided into the sub-block of M*N, obtain weight map；Then weight map is utilized to obtain step (4)
To local similarity figure be weighted processing, obtain final objective evaluation value R_GSSIM.
2. the image quality evaluating method based on ROI according to claim 1, it is characterised in that：Gray scale in the step (1)
The method for transformation of figure includes：Maximum value process, mean value method, weighted average method.
3. the image quality evaluating method according to claim 1 based on ROI, it is characterised in that：It is terraced in the step (3)
The step of spending information extraction, including：
Step (31), the gradient map for calculating separately reference picture and distorted image；
Gradient map is divided into M*N sub-block by step (32)；
Step (33), the gradient comparison function g (x, y) for calculating reference picture gradient map sub-block corresponding with distorted image.
4. the image quality evaluating method based on ROI according to claim 3：The extraction of the gradient information realizes process such as
Under：
Gradient operator in the step (31) in gradient nomography using improved all directions to Sobel operators：Using size
For 5 × 5 eight direction templates, eight directions are respectively 0 °, 22.5 °, 45 °, 67.5 °, 90 °, 112.5 °, 135 ° and 157.5 °；
Assume that reference picture is X in the step (33), distorted image Y, x and y are respectively phase in reference picture X and distortion Y
Corresponding sub-block, the gradient comparison function for defining reference picture sub-block x and distorted image sub-block y are g (x, y)；
According to the definition of gradient comparison function, it is known that g (x, y) meets following three characteristics：
(a) comparative：G (x, y)=g (y, x)；
(b) boundedness：g(x,y)≤1；
(c) maximum value uniqueness：G (x, y)=1, when reference picture sub-block x is identical with distorted image sub-block y,
Value is just 1.
5. the image quality evaluating method based on ROI according to claim 1：Merge in the step (4), fusion is joined
The realization process for examining the local similarity figure of image and distorted image is as follows：
Step (41), by brightness comparison function l (x, y), contrast comparison function c (x, y), structure comparison function s (x, y) and ladder
Degree comparison function g (x, y) is merged using multiplication mode, is merged and obtain the image quality evaluation based on gradient-structure similarity
Model GSSIM (x, y)；
Step (42) repeats the GSSIM values that step (41) calculates each image block, then asks the GSSIM values of all image blocks
It is average, obtain the final objective evaluation value MGSSIM (x, y) of distorted image.
6. the image quality evaluating method based on ROI according to claim 1：Step (5) the objective evaluation value R_GSSIM
Calculating the step of include：
Step (51) first, using the vision significance computation model of the phase spectrometry of Fourier transformation, calculates reference picture
ROI schemes；It is as follows：
(1.1) by Fourier transform by reference picture f (x, y) by space field transformation to frequency domain：F (u, v)=f { f (x, y) }；
(1.2) the phase spectrum A (u, v) of reference picture is calculated；
(1.3) phase spectrum of reference picture is subjected to inverse Fourier transform, the vision based on phase spectrum obtained in spatial domain is aobvious
Work property figure：SMap (x, y), i.e. ROI scheme；Secondly, the ROI figures of reference picture are divided into M*N sub-block, obtain weight map；It utilizes
The Gauss window of M*N carries out sliding piecemeal to vision significance figure SMap (x, y), if x, y are respectively reference picture and distortion map
As the sub-block of corresponding position, with sub-block x, SMap (x, y) sub-block of y same sizes and position is s, defines the weights at the position
For W, weight map is obtained；
Step (52), the local similarity figure obtained to step (4) using weight map are weighted processing, and it is final objective to obtain
Evaluation of estimate R_GSSIM.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810319898.7A CN108550146A (en) | 2018-04-11 | 2018-04-11 | A kind of image quality evaluating method based on ROI |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810319898.7A CN108550146A (en) | 2018-04-11 | 2018-04-11 | A kind of image quality evaluating method based on ROI |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108550146A true CN108550146A (en) | 2018-09-18 |
Family
ID=63514431
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810319898.7A Pending CN108550146A (en) | 2018-04-11 | 2018-04-11 | A kind of image quality evaluating method based on ROI |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108550146A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109754391A (en) * | 2018-12-18 | 2019-05-14 | 北京爱奇艺科技有限公司 | A kind of image quality evaluating method, device and electronic equipment |
| CN111062916A (en) * | 2019-12-05 | 2020-04-24 | 苏州大学 | Microscopic image definition evaluation method and device |
| CN111598837A (en) * | 2020-04-21 | 2020-08-28 | 中山大学 | Full-reference image quality evaluation method and system suitable for visual two-dimensional code |
| CN112215145A (en) * | 2020-10-12 | 2021-01-12 | 招商局重庆交通科研设计院有限公司 | Tunnel lighting lamp cleaning effect monitoring method based on image comparison |
| CN112291540A (en) * | 2019-07-22 | 2021-01-29 | 奇景光电股份有限公司 | Method and apparatus for performing automatic exposure control in a depth sensing system |
| CN112700485A (en) * | 2020-12-31 | 2021-04-23 | 重庆电子工程职业学院 | Image depth information extraction method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102169576A (en) * | 2011-04-02 | 2011-08-31 | 北京理工大学 | Quantified evaluation method of image mosaic algorithms |
| US8620092B2 (en) * | 2010-03-04 | 2013-12-31 | Hewlett-Packard Development Company, L.P. | Determining similarity of two images |
| CN104021545A (en) * | 2014-05-12 | 2014-09-03 | 同济大学 | Full-reference color image quality evaluation method based on visual saliency |
-
2018
- 2018-04-11 CN CN201810319898.7A patent/CN108550146A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8620092B2 (en) * | 2010-03-04 | 2013-12-31 | Hewlett-Packard Development Company, L.P. | Determining similarity of two images |
| CN102169576A (en) * | 2011-04-02 | 2011-08-31 | 北京理工大学 | Quantified evaluation method of image mosaic algorithms |
| CN104021545A (en) * | 2014-05-12 | 2014-09-03 | 同济大学 | Full-reference color image quality evaluation method based on visual saliency |
Non-Patent Citations (1)
| Title |
|---|
| 于明明: "基于结构相似度图像质量评价算法的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109754391A (en) * | 2018-12-18 | 2019-05-14 | 北京爱奇艺科技有限公司 | A kind of image quality evaluating method, device and electronic equipment |
| CN109754391B (en) * | 2018-12-18 | 2021-10-22 | 北京爱奇艺科技有限公司 | Image quality evaluation method and device and electronic equipment |
| CN112291540A (en) * | 2019-07-22 | 2021-01-29 | 奇景光电股份有限公司 | Method and apparatus for performing automatic exposure control in a depth sensing system |
| CN112291540B (en) * | 2019-07-22 | 2022-03-18 | 奇景光电股份有限公司 | Method and apparatus for performing automatic exposure control in a depth sensing system |
| CN111062916A (en) * | 2019-12-05 | 2020-04-24 | 苏州大学 | Microscopic image definition evaluation method and device |
| CN111062916B (en) * | 2019-12-05 | 2023-12-29 | 苏州大学 | Definition evaluation method and device for microscopic image |
| CN111598837A (en) * | 2020-04-21 | 2020-08-28 | 中山大学 | Full-reference image quality evaluation method and system suitable for visual two-dimensional code |
| CN111598837B (en) * | 2020-04-21 | 2023-05-05 | 中山大学 | Full-reference image quality evaluation method and system suitable for visualized two-dimensional code |
| CN112215145A (en) * | 2020-10-12 | 2021-01-12 | 招商局重庆交通科研设计院有限公司 | Tunnel lighting lamp cleaning effect monitoring method based on image comparison |
| CN112700485A (en) * | 2020-12-31 | 2021-04-23 | 重庆电子工程职业学院 | Image depth information extraction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108550146A (en) | A kind of image quality evaluating method based on ROI | |
| CN106920232B (en) | Gradient similarity graph image quality evaluation method and system based on conspicuousness detection | |
| CN109523506B (en) | Full-reference stereo image quality objective evaluation method based on visual salient image feature enhancement | |
| CN109255358B (en) | 3D image quality evaluation method based on visual saliency and depth map | |
| CN101976444B (en) | Pixel type based objective assessment method of image quality by utilizing structural similarity | |
| CN105957054B (en) | A kind of image change detection method | |
| CN104134204A (en) | Image definition evaluation method and image definition evaluation device based on sparse representation | |
| CN103475897B (en) | Adaptive image quality evaluation method based on distortion type judgment | |
| CN110400293B (en) | No-reference image quality evaluation method based on deep forest classification | |
| CN101562675B (en) | No-reference image quality evaluation method based on Contourlet transform | |
| CN110070539A (en) | Image quality evaluating method based on comentropy | |
| CN109242834A (en) | It is a kind of based on convolutional neural networks without reference stereo image quality evaluation method | |
| Jia et al. | Image denoising via sparse representation over grouped dictionaries with adaptive atom size | |
| CN105976351B (en) | Stereo image quality evaluation method based on central offset | |
| CN109003265A (en) | A kind of non-reference picture assessment method for encoding quality based on Bayes's compressed sensing | |
| CN109978854A (en) | A kind of screen content image quality measure method based on edge and structure feature | |
| CN109344845A (en) | A kind of feature matching method based on Triplet deep neural network structure | |
| CN109544504A (en) | Screen picture quality evaluating method based on rarefaction representation | |
| CN108257125A (en) | A kind of depth image quality based on natural scene statistics is without with reference to evaluation method | |
| CN112184672A (en) | No-reference image quality evaluation method and system | |
| CN106023214B (en) | Image quality evaluating method and system based on central fovea view gradient-structure similitude | |
| CN104103064A (en) | Reference-free noise image quality evaluation method based on gradient similarity | |
| CN108399620A (en) | A kind of image quality evaluating method based on low-rank sparse matrix decomposition | |
| CN104217430B (en) | Image significance detection method based on L1 regularization | |
| CN104144339B (en) | A kind of matter based on Human Perception is fallen with reference to objective evaluation method for quality of stereo images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180918 |
|
| WD01 | Invention patent application deemed withdrawn after publication |