CN104835172A - No-reference image quality evaluation method based on phase consistency and frequency domain entropy - Google Patents
No-reference image quality evaluation method based on phase consistency and frequency domain entropy Download PDFInfo
- Publication number
- CN104835172A CN104835172A CN201510261242.0A CN201510261242A CN104835172A CN 104835172 A CN104835172 A CN 104835172A CN 201510261242 A CN201510261242 A CN 201510261242A CN 104835172 A CN104835172 A CN 104835172A
- Authority
- CN
- China
- Prior art keywords
- frequency domain
- theta
- image
- phase equalization
- sigma
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/149—Segmentation; Edge detection involving deformable models, e.g. active contour models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/97—Determining parameters from multiple pictures
Abstract
The invention relates to a no-reference image quality evaluation method based on phase consistency and frequency domain entropy, which is characterized by the steps of using frequency domain entropy histogram curve features and phase consistency histogram curve features of an image under different scales to act as features of the image quality through down-sampling processing, feature extraction processing and feature fusion processing; and mapping the acquired features into an image quality score by using a two-step framework for no-reference image quality evaluation. According to the invention, the histogram curve features of a phase consistency value of the image are used to replace the histogram curve features of the space entropy in an SSEQ algorithm, thereby being capable of carrying out quality evaluation on the image quickly and robustly under the condition that the image is distorted with the type being unknown, improving the consistency between an algorithm prediction score and subjective scoring, reducing the computation time complexity of the algorithm, and having higher prediction accuracy and lower computation time complexity.
Description
Technical field
The invention belongs to image quality evaluation technical field, especially a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy.
Background technology
Objective image quality is evaluated object and is to set up the model that can carry out image quality evaluation and these models can be provided experience with the subjective vision of people close picture quality of trying one's best to mark.Along with the development of the communication technology and the generally application of digital picture, Objective image quality evaluation has become the important problem in numerous image applications, such as image acquisition, transmission, compression, recovery and enhancing.Because subjective picture quality evaluation method can not be used in real time and automated system, the Objective image quality evaluation method that exploitation can weigh picture quality automatically and robustly becomes more and more necessary.Based on the need of undistorted original image, Objective image quality evaluation method can be divided three classes: complete with reference to Objective image quality evaluation method, half with reference to Objective image quality evaluation method with without with reference to Objective image quality evaluation method.
Because undistorted image can not obtain under most of reality scene, become unique method that can be embedded in image quality evaluation application system without with reference to Objective image quality evaluation method.Early stage Objective image quality evaluation method hypothesis is only stood the distortion of particular type by altimetric image, such as, for the Objective image quality evaluation method of JP2K or JPEG distortion.Usually, these methods extract the feature for certain distortion relevant to visual quality loss, but need the type of distortion knowing image in advance due to these class methods, and therefore, its range of application is subject to serious restriction.
In contrast, universal is evaluate picture quality when type of distortion the unknown without reference Objective image quality evaluation method.Most universal image quality evaluating method is the framework based on training and study.Moorthy and Bovik proposes BIQI algorithm (Moorthy A K, Bovik A C.A two-stepframework for constructing blind image quality indices [J] .SignalProcessing Letters, IEEE, 2010,17 (5): 513-516.), this algorithm proposes the two step frameworks being used for non-reference picture quality appraisement, namely first to the classification being carried out type of distortion by altimetric image, then the evaluating objective quality for certain distortion type is carried out to image.They proposed to be DIIVINE algorithm (the Moorthy A K based on two step frameworks equally afterwards, Bovik A C.Blind image quality assessment:From natural scene statistics to perceptual quality [J] .Image Processing, IEEE Transactions on, 2011,20 (12): 3350-3364.), it extracts natural scene statistical nature from wavelet field, reaches good performance.Saad etc. propose BLIINDS-II algorithm (Saad M A, BovikA C, Charrier C.Blind image quality assessment:A natural scenestatistics approach in the DCT domain [J] .Image Processing, IEEETransactions on, 2012,21 (8): 3339-3352.), it utilizes the natural scene statistical nature of block DCT domain, and proposes a single step framework fast.BRISQUE algorithm extracts natural scene statistical nature from spatial domain, and this makes the time complexity of algorithm greatly reduce, and this algorithm reaches better estimated performance simultaneously.SSEQ (the Liu L that L.Liu etc. propose, Liu B, Huang H, et al.No-reference imagequality assessment based on spatial and spectral entropies [J] .SignalProcessing:Image Communication, 2014,29 (8): 856-863.) extract natural scene statistical nature from histogram curve and make use of two step frameworks of non-reference picture quality appraisement, reaching good estimated performance.SSEQ method make use of the histogram curve feature of the histogram curve characteristic sum space entropy of frequency domain entropy, but the histogram curve feature of space entropy and picture quality correlativity difference and histogram curve feature calculation time complexity based on block space entropy is high.
Summary of the invention
The object of the invention is to overcome the deficiencies in the prior art, a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy be provided, solve that existing method estimated performance is poor, the high problem of complexity computing time.
The present invention solves its technical matters and takes following technical scheme to realize:
Based on a non-reference picture quality appraisement method for phase equalization and frequency domain entropy, comprise the following steps:
Step 1, by down-sampling process, feature extraction process and Fusion Features process, using the feature of the frequency domain entropy histogram curve characteristic sum phase equalization histogram curve feature under image different scale as picture quality;
Step 2, utilize two step frameworks of non-reference picture quality appraisement by the Feature Mapping obtained for picture quality is marked.
And described down-sampling disposal route is: input picture carries out down-sampling with k multiplying power, obtain the image of n yardstick of input picture, namely original scale image, k multiplying power down-sampled images ..., k
n-1multiplying power down-sampled images.
And described feature extraction disposal route is: the value calculating the value of the frequency domain entropy of the block based on B × B size and the phase equalization based on pixel on each scalogram picture.
And described Fusion Features disposal route is: by phase equalization value and frequency domain entropy being arranged by ascending order respectively, obtain sorted phase equalization value PC=(pc
1, pc
2..., pc
m) and frequency domain entropy F=(fe
1, fe
2..., fe
n), wherein pc
iand fe
jbe phase equalization value and local frequency domain entropy, m and n is the number of the number of pixels of each yardstick and the block of B × B size respectively; Then ratio fusion is carried out to PC and F, namely obtain from the q% element at PC and F extraction center respectively
with
wherein q is one [0,100] constant in scope, finally the degree of bias value of the mean value of PCp and Fp and PC and F is formed jointly the final feature f=(mean (PCp) of each graphical rule, skew (PC), mean (Fp), skew (F)), wherein mean represents and gets averaging operation, and skew represents and gets degree of bias Value Operations.
And the detailed process of described step 2 is; The first step in two step frameworks is a type of distortion sorter, utilizes it to calculate by the possible probability of altimetric image for each type of distortion; Second step is the regression model for each certain distortion type, and this regression model obtains by utilizing each distorted image and their subjective scoring to train; Obtained by altimetric image two vector dot obtained by two step frameworks of finally marking, two vectors are the quality score vector of often kind of type of distortion that the possibility vector sum of each type of distortion that type of distortion sorter provides provides for the regression model of certain distortion type respectively.
And the frequency domain entropy of the described block based on B × B size is:
Wherein P (i, j) is defined as:
Wherein C (i, j) is the DCT coefficient at (i, j) place in B × B block, 1 < i≤B, and 1 < j≤B, DC coefficient is removed.
And the calculation procedure of the described phase equalization based on pixel is:
First, the picture of each yardstick and following two dimensional logarithmic Gabor filter are carried out convolution,
Wherein ω
0the centre frequency of wave filter, σ
rcontrol filter bandwidht, θ
j=j π/J, j={0,1 ..., J-1} is the deflection of wave filter, and J is direction number, σ
θdetermine the angular bandwidth of wave filter;
Then, use
represent above-mentioned two dimensional logarithmic wave filter at the n-th yardstick θ
jodd symmetry wave filter on direction and even symmetry wave filter, two wave filters are conjugation; By regulating ω
0and θ
j, G
2one group can be obtained with image convolution at each location of pixels x place to respond
wherein
with
that two conjugate filters are at the n-th yardstick θ
jthe response at x pixel place on direction; N-th yardstick θ
jlocal acknowledgement's amplitude on direction is
θ
jthe local energy in direction is
Wherein
Be defined as in the phase equalization at x pixel place:
Wherein λ is a very little positive constant, and the span of PC (x) is between 0-1.
Advantage of the present invention and good effect are:
The histogram curve feature of the present invention by utilizing the histogram curve feature of image phase consistance value to replace space entropy in SSEQ algorithm, can fast and robustly when the unknown of image fault type, quality assessment is carried out to image, improve the consistance between algorithm predicts mark and people's subjective scoring, reduce complexity computing time of algorithm, there is higher prediction accuracy and lower complexity computing time.
Accompanying drawing explanation
Fig. 1 is the curve histogram of the phase equalization value of original image and its 5 kinds of type of distortion images;
Fig. 2 is the curve histogram of the frequency domain entropy of original image and its 5 kinds of type of distortion images.
Embodiment
Below in conjunction with accompanying drawing, the embodiment of the present invention is further described:
Based on a non-reference picture quality appraisement method for phase equalization and frequency domain entropy, comprise the following steps:
Step 1, by down-sampling process, feature extraction process and Fusion Features process, using the feature of the frequency domain entropy histogram curve characteristic sum phase equalization histogram curve feature under image different scale as picture quality.
In this step, comprise down-sampling process, feature extraction process and Fusion Features processing procedure, be described respectively below:
(1) down-sampling processing procedure: input picture carries out down-sampling twice with 2 multiplying powers, obtains the image of 3 yardsticks of input picture, i.e. original scale image, 2 multiplying power down-sampled images and 4 multiplying power down-sampled images;
(2) feature extraction processing procedure: calculate the value based on the value of the frequency domain entropy of the block based on 8 × 8 sizes and the phase equalization based on pixel on each scalogram picture;
Its frequency domain entropy is defined as
Wherein P (i, j) is the frequency that normalized DCT coefficient occurs, it is defined as
Wherein C (i, j) is the DCT coefficient at (i, j) place in 8 × 8 pieces, wherein 1 < i≤B, 1 < j≤B, and DC coefficient is removed;
The phase equalization calculated based on pixel mainly contains following steps:
First by the picture of each yardstick and following two dimensional logarithmic Gabor filter convolution,
Wherein ω
0the centre frequency of wave filter, σ
rcontrol filter bandwidht, θ
j=j π/J, j={0,1 ..., J-1} is the deflection of wave filter, and J is direction number, σ
θdetermine the angular bandwidth of wave filter;
Then, use
represent above-mentioned two dimensional logarithmic wave filter at the n-th yardstick θ
jodd symmetry wave filter on direction and even symmetry wave filter, two wave filters are conjugation.By regulating ω
0and θ
j, G
2one group can be obtained with image convolution at each location of pixels x place to respond
wherein
with
that two conjugate filters are at the n-th yardstick θ
jthe response at x pixel place on direction.N-th yardstick θ
jlocal acknowledgement's amplitude on direction is
θ
jthe local energy in direction is
Wherein
Be defined as in the phase equalization at x pixel place
Wherein λ is a very little positive constant, and the span of PC (x) is between 0-1.
(3) Fusion Features processing procedure: by phase equalization value and frequency domain entropy being arranged by ascending order respectively, obtain sorted PC=(pc
1, pc
2..., pc
m), F=(fe
1, fe
2..., fe
n), wherein pc
iand fe
jthe number of to be phase equalization value and local frequency domain entropy m and n the be number of pixels of each yardstick and the block of 8 × 8 sizes; Then ratio fusion is carried out to PC and F, obtain from 60% element at PC and F extraction center respectively
with
then the mean value of PCp and Fp and the degree of bias value of PC and F form the final feature f=(mean (PCp) of each graphical rule jointly, skew (PC), mean (Fp), skew (F)), wherein mean represents and gets averaging operation, and skew represents and gets degree of bias Value Operations.
Step 2, utilize two step frameworks of non-reference picture quality appraisement by the Feature Mapping obtained for picture quality is marked.The detailed process of this step is:
Eachly in the present invention obtained 3 yardsticks by altimetric image by down-sampling, from each yardstick, be extracted 4 features, therefore proposed 12 dimensional features from each altogether by altimetric image; Utilize two step frameworks of non-reference picture quality appraisement by maps feature vectors for picture quality is marked; The first step in two step frameworks is a type of distortion sorter, it can calculate by the possible probability of altimetric image for each type of distortion, second step is the regression model for each certain distortion type, and it obtains by utilizing each distorted image and their subjective scoring to train; Obtained by altimetric image two vector dot obtained by two step frameworks of finally marking, two vectors are the quality score vector of often kind of type of distortion that the possibility vector sum of each type of distortion that type of distortion sorter provides provides for the regression model of certain distortion type respectively; Utilize support vector machine (SVM) to set up sorter in this example, utilize support vector regression (SVR) to set up regression model.
In order to be described effect of the present invention, below experimental demonstration is carried out to the present invention.
(1), test data set and Performance Evaluating Indexes
Fig. 1 gives the curve histogram of the phase equalization value of original image and its 5 kinds of type of distortion images, and Fig. 2 gives the curve histogram of the frequency domain entropy of original image and its 5 kinds of type of distortion images.In figure, " Original " corresponding original graph, its subjective scoring DMOS value is 0; " JP2K " corresponding original graph stands the figure of jpeg2000 distortion, and its DMOS value is 67.6968; " JPEG " corresponding original graph stands the figure of jpeg distortion, and its DMOS value is 70.5024; " Noise " corresponding original graph stands the figure of white noise distortion, and its DMOS value is 55.0675; " FF " corresponding original graph stands the figure of fast attenuation channel distortion, and its DMOS value is 64.7162.
LIVE image quality evaluation data set is tested, and LIVE data set comprises 29 test patterns and 5 kinds of type of distortion---jpeg 2000 compression artefacts (JP2K), jpeg compression artefacts (JPEG), additive white Gaussian noise distortion (Noise), Gaussian Blur distortion (BLUR) and fast attenuation channel distortion (FF).Data centralization often plants type of distortion 5 to 6 specified distortion level.Data set provides the subjective difference Mean Opinion Score (DMOS) of distorted picture.
Utilize Spearman rank correlation coefficient (SROCC) and Pearson's linearly dependent coefficient (LCC) as the evaluation index of model prediction accuracy.SROCC weighs the monotonicity of predicted data, and it only carries out computing to the sort order of prediction mark.LCC can be considered to be the prediction accuracy of model.The span of SROCC and LCC is [0,1], and the estimated performance of their the higher expression model of value is more close to the subjective assessment of people.
2, overall performance test
The reference picture of each random selecting LIVE data centralization 80% and their distorted image are as training set, remaining 20% reference picture and their distorted image carry out training and testing as test set, using the median of the result of 1000 training and testings as final Performance figure, experimental result lists in table one and table two.
Table one carries out the median of the SROCC of 1000 training and testings on LIVE data set
JP2K | JPEG | Noise | Blur | FF | ALL | |
SSEQ | 0.8971 | 0.9427 | 0.9640 | 0.9155 | 0.8536 | 0.9121 |
This method | 0.9059 | 0.9453 | 0.9689 | 0.9333 | 0.8572 | 0.9226 |
Table two carries out the median of the LCC of 1000 training and testings on LIVE data set
JP2K | JPEG | Noise | Blur | FF | ALL | |
SSEQ | 0.9026 | 0.9608 | 0.9695 | 0.9289 | 0.8745 | 0.9177 |
This method | 0.9101 | 0.9516 | 0.9750 | 0.9397 | 0.8808 | 0.9245 |
As can be seen from the above results, this method except the LCC value under JPEG type of distortion is lower than except existing SSEQ algorithm, in other independent type of distortion and SROCC and LCC of all type of distortion hybrid detection (ALL) all higher than existing SSEQ algorithm.This illustrates that this method has better estimated performance.
3, feature set performance test
In order to prove phase equalization histogram curve feature (by the average of the phase equalization histogram curve of altimetric image on 3 yardsticks and the degree of bias) the proposed by the invention validity for non-reference picture quality appraisement, respectively by phase equalization histogram curve feature, use in frequency domain entropy histogram curve feature (by the average of the histogram curve of the frequency domain entropy of altimetric image on 3 yardsticks and the degree of bias) and SSEQ algorithm space entropy histogram curve feature (by the average of the histogram curve of the space entropy of altimetric image on 3 yardsticks and the degree of bias) apply to separately in non-reference picture quality appraisement two step framework, and the reference picture of each random selecting LIVE data centralization 80% and their corresponding distorted images are as training set, remaining 20% reference picture and their corresponding distorted images carry out training and testing as test set, using the Performance figure of the median of 1000 training test results as each feature set, experimental result lists in table three and table four.
Table three different characteristic collection carries out the median of the SROCC of 1000 training tests on LIVE data set
JP2K | JPEG | Noise | Blur | FF | ALL | |
The entropy feature in frequency domain collection | 0.9157 | 0.9268 | 0.9301 | 0.9217 | 0.8449 | 0.8933 |
Phase equalization feature set | 0.8458 | 0.8630 | 0.9528 | 0.8892 | 0.8258 | 0.7826 |
Space entropy feature set | 0.6404 | 0.8725 | 0.9146 | 0.7448 | 0.6040 | 0.7568 |
Table four different characteristic collection carries out the median of the SROCC of 1000 training tests on LIVE data set
JP2K | JPEG | Noise | Blur | FF | ALL | |
The entropy feature in frequency domain collection | 0.9224 | 0.9302 | 0.9406 | 0.9298 | 0.8703 | 0.9004 |
Phase equalization feature set | 0.8338 | 0.8810 | 0.9662 | 0.8852 | 0.8519 | 0.7880 |
Space entropy feature set | 0.6584 | 0.8919 | 0.9202 | 0.7720 | 0.6788 | 0.7689 |
As can be seen from table three and table four, as SROCC and LCC of more all type of distortion hybrid detection (ALL), the performance of the entropy feature in frequency domain collection is best, phase equalization feature set performance is apparently higher than space entropy feature set, and this phase equalization feature performance when being applied to two step framework of non-reference picture quality appraisement illustrating that the present invention proposes obviously is better than the space entropy feature used in SSEQ.
4, feature extraction time complexity test
Test pattern is LIVE data centralization " buildings.bmp ", and test platform is a desktop computer having four core CPU, dominant frequency 3.30GHz, 8GB internal memory.To image zooming-out 1000 features, the averaging time that each feature extraction consumes lists in table five.
The averaging time of table five 1000 feature extractions
SSEQ | 2.112s |
This method | 1.943s |
As can be seen from the above table, the extraction required time of the phase equalization histogram curve characteristic sum frequency domain entropy histogram curve feature that this method is used is obviously less than the space entropy histogram curve characteristic sum frequency domain entropy histogram curve feature that SSEQ utilizes.
In order to prove the superiority of frequency domain entropy histogram curve feature calculation time complexity proposed by the invention further, three kinds of histogram curve features that " buildings.bmp " image zooming-out SSEQ and this method are used 1000 times, and overall time number percent shared by each feature extraction is listed in table six.
Table six extracts three kinds of time consuming ratios of feature set
Time scale shared by feature extraction | |
Frequency domain entropy histogram curve feature | 28.92% |
Phase equalization histogram curve feature | 33.39% |
Space entropy histogram curve feature | 37.69% |
The time that phase equalization histogram curve feature calculation proposed by the invention as can be seen from Table VI consumes obviously is less than space entropy histogram curve feature.
5, classification accuracy
The median of the type of distortion classification accuracy of this method on LIVE data set in 1000 training and testings lists in table seven, and this classification accuracy is exactly the classification accuracy of the first step (i.e. type of distortion sorter) in two step frameworks.
The median of the classification accuracy of table seven 1000 training and testings
JP2K | JPEG | Noise | Blur | FF | ALL | |
SSEQ | 66.67 | 86.84 | 100.00 | 66.67 | 46.67 | 73.13 |
This method | 71.43 | 93.12 | 100.00 | 63.33 | 43.33 | 74.53 |
Can find out that overall (ALL) classification accuracy of this method is a little more than SSEQ from this table.
6, data set independence test
The test of data set independence is to prove that the premium properties of this method is independent of data set.This method is carried out training on whole LIVE data set and tests on TID2008 data set.TID2008 data set comprises the distorted image of 25 reference pictures and 1700 17 kinds of type of distortion.There are 24 to be natural scene image in 25 images, have in 17 kinds of type of distortion four kinds identical with the type of distortion of LIVE data centralization (JP2K, JPEG, Noise, Blur).The SROCC value of the type of distortion that four kinds of 24 natural scene images identical with LIVE data set being carried out test is listed in table eight.
Table eight is trained and the SROCC tested on TID2008 data set on whole LIVE data set
JP2K | JPEG | Noise | Blur | ALL | |
SSEQ | 0.9112 | 0.8629 | 0.7986 | 0.8531 | 0.8501 |
This method | 0.8618 | 0.9065 | 0.7765 | 0.8589 | 0.8840 |
As can be seen from Table VIII, this method still can provide higher SROCC value on TID data set, and this illustrates that this method has good data set independence, and the overall performance of this method on TID2008 (ALL) is still obviously better than SSEQ.
It is emphasized that; embodiment of the present invention is illustrative; instead of it is determinate; therefore the present invention includes the embodiment be not limited to described in embodiment; every other embodiments drawn by those skilled in the art's technical scheme according to the present invention, belong to the scope of protection of the invention equally.
Claims (7)
1., based on a non-reference picture quality appraisement method for phase equalization and frequency domain entropy, it is characterized in that comprising the following steps:
Step 1, by down-sampling process, feature extraction process and Fusion Features process, using the feature of the frequency domain entropy histogram curve characteristic sum phase equalization histogram curve feature under image different scale as picture quality;
Step 2, utilize two step frameworks of non-reference picture quality appraisement by the Feature Mapping obtained for picture quality is marked.
2. a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy according to claim 1, it is characterized in that: described down-sampling disposal route is: input picture carries out down-sampling with k multiplying power, obtain the image of n yardstick of input picture, namely original scale image, k multiplying power down-sampled images ..., k
n-1multiplying power down-sampled images.
3. a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy according to claim 1, is characterized in that: described feature extraction disposal route is: the value calculating the value of the frequency domain entropy of the block based on B × B size and the phase equalization based on pixel on each scalogram picture.
4. a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy according to claim 1, it is characterized in that: described Fusion Features disposal route is: by phase equalization value and frequency domain entropy being arranged by ascending order respectively, obtain sorted phase equalization value PC=(pc
1, pc
2..., pc
m) and frequency domain entropy F=(fe
1, fe
2..., fe
n), wherein pc
iand fe
jbe phase equalization value and local frequency domain entropy, m and n is the number of the number of pixels of each yardstick and the block of B × B size respectively; Then ratio fusion is carried out to PC and F, namely obtain from the q% element at PC and F extraction center respectively
with
wherein q is one [0,100] constant in scope, finally the degree of bias value of the mean value of PCp and Fp and PC and F is formed jointly the final feature f=(mean (PCp) of each graphical rule, skew (PC), mean (Fp), skew (F)), wherein mean represents and gets averaging operation, and skew represents and gets degree of bias Value Operations.
5. a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy according to claim 1, is characterized in that: the detailed process of described step 2 is; The first step in two step frameworks is a type of distortion sorter, utilizes it to calculate by the possible probability of altimetric image for each type of distortion; Second step is the regression model for each certain distortion type, and this regression model obtains by utilizing each distorted image and their subjective scoring to train; Obtained by altimetric image two vector dot obtained by two step frameworks of finally marking, two vectors are the quality score vector of often kind of type of distortion that the possibility vector sum of each type of distortion that type of distortion sorter provides provides for the regression model of certain distortion type respectively.
6. a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy according to claim 3, is characterized in that: the frequency domain entropy of the described block based on B × B size is:
Wherein P (i, j) is defined as:
Wherein C (i, j) is the DCT coefficient at (i, j) place in B × B block, 1 < i≤B, and 1 < j≤B, DC coefficient is removed.
7. a kind of non-reference picture quality appraisement method based on phase equalization and frequency domain entropy according to claim 3, is characterized in that: the calculation procedure of the described phase equalization based on pixel is:
First, the picture of each yardstick and following two dimensional logarithmic Gabor filter are carried out convolution,
Wherein ω
0the centre frequency of wave filter, σ
rcontrol filter bandwidht, θ
j=j π/J, j={0,1 ..., J-1} is the deflection of wave filter, and J is direction number, σ
θdetermine the angular bandwidth of wave filter;
Then, use
represent above-mentioned two dimensional logarithmic wave filter at the n-th yardstick θ
jodd symmetry wave filter on direction and even symmetry wave filter, two wave filters are conjugation; By regulating ω
0and θ
j, G
2one group can be obtained with image convolution at each location of pixels x place to respond
wherein
with
that two conjugate filters are at the n-th yardstick θ
jthe response at x pixel place on direction; N-th yardstick θ
jlocal acknowledgement's amplitude on direction is
θ
jthe local energy in direction is
Wherein
Be defined as in the phase equalization at x pixel place:
Wherein λ is a very little positive constant, and the span of PC (x) is between 0-1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510261242.0A CN104835172A (en) | 2015-05-21 | 2015-05-21 | No-reference image quality evaluation method based on phase consistency and frequency domain entropy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510261242.0A CN104835172A (en) | 2015-05-21 | 2015-05-21 | No-reference image quality evaluation method based on phase consistency and frequency domain entropy |
Publications (1)
Publication Number | Publication Date |
---|---|
CN104835172A true CN104835172A (en) | 2015-08-12 |
Family
ID=53813037
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510261242.0A Pending CN104835172A (en) | 2015-05-21 | 2015-05-21 | No-reference image quality evaluation method based on phase consistency and frequency domain entropy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104835172A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108376394A (en) * | 2018-01-12 | 2018-08-07 | 上海大学 | A kind of camera Atomatic focusing method and system based on frequency domain histogram analysis |
CN109345520A (en) * | 2018-09-20 | 2019-02-15 | 江苏商贸职业学院 | A kind of quality evaluating method of image definition |
CN110070539A (en) * | 2019-04-28 | 2019-07-30 | 重庆大学 | Image quality evaluating method based on comentropy |
CN112132757A (en) * | 2020-05-14 | 2020-12-25 | 北京理工大学 | General image restoration method based on neural network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011250013A (en) * | 2010-05-25 | 2011-12-08 | Nippon Telegr & Teleph Corp <Ntt> | Image quality evaluation method, image quality evaluation apparatus, and program |
CN104104943A (en) * | 2013-04-10 | 2014-10-15 | 江南大学 | No-reference JPEG2000 compressed image quality evaluation method based on generalized regression neural network |
-
2015
- 2015-05-21 CN CN201510261242.0A patent/CN104835172A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011250013A (en) * | 2010-05-25 | 2011-12-08 | Nippon Telegr & Teleph Corp <Ntt> | Image quality evaluation method, image quality evaluation apparatus, and program |
CN104104943A (en) * | 2013-04-10 | 2014-10-15 | 江南大学 | No-reference JPEG2000 compressed image quality evaluation method based on generalized regression neural network |
Non-Patent Citations (2)
Title |
---|
LIN ZHANG ET AL: "FSIM: A Feature Similarity Index for Image Quality Assessment", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
LIXIONG LIU ET AL: "No-reference image quality assessment based on spatial and spectral entropies", 《SIGNAL PROCESSING: IMAGE COMMUNICATION》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108376394A (en) * | 2018-01-12 | 2018-08-07 | 上海大学 | A kind of camera Atomatic focusing method and system based on frequency domain histogram analysis |
CN108376394B (en) * | 2018-01-12 | 2021-08-10 | 上海大学 | Camera automatic focusing method and system based on frequency domain histogram analysis |
CN109345520A (en) * | 2018-09-20 | 2019-02-15 | 江苏商贸职业学院 | A kind of quality evaluating method of image definition |
CN110070539A (en) * | 2019-04-28 | 2019-07-30 | 重庆大学 | Image quality evaluating method based on comentropy |
CN112132757A (en) * | 2020-05-14 | 2020-12-25 | 北京理工大学 | General image restoration method based on neural network |
CN112132757B (en) * | 2020-05-14 | 2022-08-16 | 北京理工大学 | General image restoration method based on neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109325550B (en) | No-reference image quality evaluation method based on image entropy | |
CN108428227B (en) | No-reference image quality evaluation method based on full convolution neural network | |
CN104134204B (en) | Image definition evaluation method and image definition evaluation device based on sparse representation | |
CN104023230B (en) | A kind of non-reference picture quality appraisement method based on gradient relevance | |
CN105208374A (en) | Non-reference image quality objective evaluation method based on deep learning | |
CN109978854B (en) | Screen content image quality evaluation method based on edge and structural features | |
CN102547368B (en) | Objective evaluation method for quality of stereo images | |
CN101976444B (en) | Pixel type based objective assessment method of image quality by utilizing structural similarity | |
CN101562675B (en) | No-reference image quality evaluation method based on Contourlet transform | |
CN103996192A (en) | Non-reference image quality evaluation method based on high-quality natural image statistical magnitude model | |
CN106447647A (en) | No-reference quality evaluation method of compression perception recovery images | |
CN107481236A (en) | A kind of quality evaluating method of screen picture | |
CN103200421A (en) | No-reference image quality evaluation method based on Curvelet transformation and phase coincidence | |
CN104318545A (en) | Foggy weather polarization image quality evaluation method | |
CN104835172A (en) | No-reference image quality evaluation method based on phase consistency and frequency domain entropy | |
CN104103064A (en) | Reference-free noise image quality evaluation method based on gradient similarity | |
CN109003265A (en) | A kind of non-reference picture assessment method for encoding quality based on Bayes's compressed sensing | |
CN108053396A (en) | A kind of more distorted image quality without with reference to evaluation method | |
CN106815839A (en) | A kind of image quality blind evaluation method | |
CN112184672A (en) | No-reference image quality evaluation method and system | |
CN107948635A (en) | It is a kind of based on degenerate measurement without refer to sonar image quality evaluation method | |
Wu et al. | Blind quality assessment for screen content images by combining local and global features | |
CN105894507A (en) | Image quality evaluation method based on image information content natural scenario statistical characteristics | |
Khalid et al. | Gaussian process-based feature-enriched blind image quality assessment | |
CN103996188A (en) | Full-reference-type image quality evaluation method based on Gabor weighted characteristics |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
EXSB | Decision made by sipo to initiate substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150812 |
|
WD01 | Invention patent application deemed withdrawn after publication |