CN108510496A - The fuzzy detection method that SVD based on Image DCT Domain is decomposed - Google Patents
The fuzzy detection method that SVD based on Image DCT Domain is decomposed Download PDFInfo
- Publication number
- CN108510496A CN108510496A CN201810313311.1A CN201810313311A CN108510496A CN 108510496 A CN108510496 A CN 108510496A CN 201810313311 A CN201810313311 A CN 201810313311A CN 108510496 A CN108510496 A CN 108510496A
- Authority
- CN
- China
- Prior art keywords
- image
- follows
- fuzzy
- dct
- dct domain
- 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
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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- 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]
-
- 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)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of fuzzy detection method that the SVD based on Image DCT Domain is decomposed.The gradient map of testing image is calculated first, the marginal information of image can be obtained from gradient map, then gradient map is carried out piecemeal, and dct transform is carried out, because the ac coefficient of DCT domain reflects the edge and clarity of image, then difference matrix analyzes the ac coefficient information of DCT domain, by the singular value for calculating difference matrix, and receptance function is constructed to indicate the fog-level of the image of block, the sum of normalized image block response finally is gone with mean value and variance, to eliminate the influence of picture material.Experiment shows that the fuzzy score that this method obtains and human eye are highly consistent to the subjective assessment score of image.The features such as detection model of the present invention broadens in view of the edge during image blur, and clarity dies down, and the influence of picture material is effectively eliminated, therefore Detection accuracy is very high, and also detection efficiency is fast, the method that overall performance is better than forefathers.
Description
Technical field
The present invention relates to image fuzzy detection fields, it is proposed that a kind of fuzzy detection that the SVD based on Image DCT Domain is decomposed
Method, what this method can be quickly and accurately detects blurred picture.
Background technology
One of the carrier that digital picture is transmitted as information, plays important role in daily life or work.Than
If the universal of mobile phone is so that mobile phone photograph becomes one of people's daily entertainment project;Satellite remote sensing images are to agricultural, industry and ring
The facility etc. that border is brought.But digital picture inevitably introduces some mistakes during acquisition, compression, transimission and storage
Very, this not only influences visual experience, but also may bring huge loss.It is a kind of the most common distortion class that image is fuzzy
Type, therefore, blurred image detection are increasingly paid attention to by people.
Although human eye has the ability for distinguishing blurred picture and clear image, there is the shortcomings of time-consuming, heavy workload,
Therefore, being detected to blurred picture using computer is just particularly important.Currently, having there is many image fuzzy detection sides
Method is generally divided into spatial domain method, frequency domain method and mixing domain method.Generally speaking, image is fuzzy will produce broader edge, therefore
Most of spatial domain method is all based on the border width of image.Such as Marziliano proposes the algorithm based on Sobel operators,
The edges Sobel of this method detection image vertical direction first, then obtain the width of image border, most by Local Extremum
The vague definition of image is average edge width afterwards;As Ferzli and Karam proposes a kind of visual fuzzy (Just
Noticable Blur, JNB) model, then the edge block and smooth block of this method image determining first calculate the block of edge block
Border width obtains the fog-level of image finally by JNB models.In view of spatial domain is to the limitation of image feature representation,
Image is then transformed into frequency domain by many methods, such as the domains DWT or DCT domain.By analyzing the frequency domain nonzero coefficient distribution of image,
Marichal etc. proposes then a kind of method based on Image DCT Domain, this method obtain each 8x8 first by image block
The DCT coefficient of block estimates that image is fuzzy finally by the weighted histogram of DCT nonzero coefficients.Tong etc. proposes that one kind is based on
The method of wavelet field, this method first classify to edge by multi-scale wavelet domain coefficient, then according to the rule of proposition
Judge whether image obscures, the fog-level of last image can be obtained by the number of fuzzy edge.Currently, having scholar
It is proposed that the hybrid domain algorithm for being combined in spatial domain with frequency domain, such as Vu also proposed a kind of comprehensive evaluation algorithm based on hybrid domain
(S3), frequency domain character is obtained by Sigmoid functional transformations, spatial feature, last blur estimation is obtained by local variation
By taking preceding two weighted averages to obtain;Such as a kind of algorithm based on discrete orthogonal moments that Li et al. proposes, this method is led to first
Gradient information is crossed to estimate the edge of image, then obtains frequency domain information using discrete orthogonal moments, finally calculates the discrete of image
The sum of orthogonal moment indicates the fog-level of image.Illustrate that the algorithm that spatial domain and frequency domain combine has better effect simultaneously.
Existing image fuzzy detection method has very much, they all utilize blurred picture feature:With the mould for increasing image
The edge of paste degree, image can become wider, profile increasingly unobvious.
Invention content
The characteristics of by comparing the above method, it is proposed that a kind of detection method that the SVD based on Image DCT Domain is decomposed, it should
Method combines image spatial domain and frequency domain information.The gradient map of image is calculated first, and the marginal information of image can be from gradient map
In obtain, gradient map is then carried out piecemeal, and carry out dct transform, because the ac coefficient of DCT domain reflects the edge of image
And clarity, the ac coefficient information of DCT domain is then analyzed with difference matrix, by calculating the singular value of difference matrix, and
Receptance function is constructed to indicate the fog-level of the image of block, is finally normalized with mean value and variance to eliminate picture material
Influence.
Steps are as follows for technical scheme of the present invention:
Step 1:The gradient map of image to be detected is calculated, and piecemeal is carried out to gradient image, the size of block is p × p.
Step 2:Dct transform is carried out to each gradient image block, DCT coefficient is obtained and removes DC coefficient.
Step 3:Calculate the difference matrix horizontally and vertically of DCT coefficient.
Step 4:The singular value of difference matrix is calculated, and the response of block is obtained by receptance function.
Step 5:Response summation to all pieces obtains the response E of entire image.
Step 6:Carrying out piecemeal to image, (the same step 1) of block size calculates the mean value and variance of each image block, to all
The mean value and variance of block are summed respectively, obtain the mean value C and variance V of entire image.
Step 7:E in step 5 is normalized in the C and V obtained with step 6, obtains final fuzzy score S.
Step 8:By comparing the size of S and the threshold value T of selection, image is divided into clear and fuzzy two classes.
Beneficial effects of the present invention:
Present invention incorporates the information in image spatial domain and frequency domain, effectively raise the accuracy rate of fuzzy detection, make up only
Use spatial domain or the defect of frequency domain.Image edge information is obtained in spatial domain, the fog-level for obtaining image in a frequency domain (is rung
Answer), the intuitive of image spatial information (si) had not only been remained in this way, but also remained the validity of image frequency domain information.In addition, of the invention
Operation also is normalized to the response that image frequency domain obtains, eliminates the influence of picture material.
Description of the drawings
Fig. 1 algorithm flow charts.
Fig. 2 clear image samples.
Fig. 3 blurred picture samples.
The fuzzy score of 58 width images in Fig. 4 LIVE image libraries.
The testing result of Fig. 5 clear images sample and blurred picture sample.
Specific implementation mode
Below in conjunction with the accompanying drawings, specific embodiments of the present invention are described in further detail.Based on Image DCT Domain
The fuzzy detection method that SVD is decomposed, specific steps description are as shown in Figure 1:
Step 1:The gradient map of image to be detected is calculated, and piecemeal is carried out to gradient image, the size of block is p × p.
Step 2:Dct transform is carried out to each gradient image block, DCT coefficient is obtained and removes DC coefficient.
Step 3:Calculate the difference matrix both horizontally and vertically of DCT coefficient.
Step 4:The singular value of difference matrix is calculated, and the response of block is obtained by receptance function.
Step 5:Response summation to all pieces, obtains the overall response E of entire image.
Step 6:Carrying out piecemeal to image, (the same step 1) of block size calculates the mean value and variance of each image block, to all
The mean value and variance of block are summed respectively, obtain the mean value C and variance V of entire image.
Step 7:E in step 5 is normalized in the C and V obtained with step 6, obtains final fuzzy score S.
Step 8:By comparing the size of S and the threshold value T of selection, image is divided into clear and fuzzy two classes.
Step 1 is specific as follows:The gradient map G of testing image I is calculated, formula is as follows:
Ix=I* [- 101], Iy=I* [- 101]T
Wherein * is convolution operation, then carries out piecemeal to gradient image G, and each piecemeal is set as Bk, k=(1,2 ..., N),
N is the total number of image block, and block size is p × p, p=6 in experiment.
Step 2 is specific as follows:By image block BkTransform to DCT domain Dk, and removing DC coefficient, formula is as follows:
Dk=DCT (Bk)
Wherein i, j ∈ { 1 ... p }.
Step 3 is specific as follows:Difference matrix horizontally and vertically is calculated separately, formula is as follows:
Wherein i ∈ 1 ..., p }, j ∈ 1 ..., p-1 }.
Wherein i ∈ 1 ..., p-1 }, j ∈ 1 ..., p }.
Step 4 is specific as follows:The singular value of difference matrix is calculated, formula is as follows:
(:) matrix is indicated to change into a column vector, the size of F is p (p-1) × 2, then carries out singular value decomposition to F
Obtain singular value s1,s2, the energy of image block is obtained by receptance function e.
ek=s1×s2-α(s1+s2)2
Wherein α is constant, α=0.01 in experiment.
Step 5 is specific as follows:The response for calculating all pieces of testing image is summed, and the overall response E of whole image is obtained,
Formula is as follows:
Wherein N is the total number of image block.
Step 6 is specific as follows:To testing image piecemeal, process calculates every piece of mean value c with step 1kWith variance vk, so
Calculate the sum of all pieces of mean value C afterwards, the sum of variance V, formula is as follows:
Step 7 is specific as follows:E is normalized with V and C, obtains final fuzzy score S, formula is as follows:
Step 8 is specific as follows:The threshold value T between clear image and blurred picture is determined first.It can be obtained clearly by step 7
The fuzzy score S of clear image (see attached drawing 2) and blurred picture (see attached drawing 3) analyzes the data S that this two classes image obtains, can be with
Find out that clear image is more much higher than the S of blurred picture (see attached drawing 5).In order to ensure the abundant of experimental data, in LIVE images
Great amount of samples (blurred picture and each half of clear image quantity) is chosen in library to be tested, and determines last threshold value T (see attached drawing
4).The threshold value of final choice is T=15 (line of black in attached drawing 4), when T > 15 indicate that image is clearly, when T≤15 is indicated
Image is fuzzy.
Claims (9)
1. the fuzzy detection method that the SVD based on Image DCT Domain is decomposed calculates the gradient image of testing image first, image
Marginal information can be obtained from gradient map, gradient map then be carried out piecemeal, and carry out dct transform, because of the exchange of DCT domain
Coefficient reflects the edge and clarity of image, and the ac coefficient information of DCT domain is then analyzed with difference matrix, and calculates difference
The singular value of sub-matrix constructs receptance function by singular value to indicate the fog-level of the image of block, finally uses mean value and side
Difference is normalized to eliminate the influence of picture material;It is as follows:
Step 1:The gradient map of image to be detected is calculated, and piecemeal is carried out to gradient image, the size of block is p × p;
Step 2:Dct transform is carried out to each gradient image block, DCT coefficient is obtained and removes DC coefficient;
Step 3:Calculate the difference matrix horizontally and vertically of DCT coefficient;
Step 4:The singular value of difference matrix is calculated, and the response of block is obtained by receptance function;
Step 5:Response summation to all pieces, obtains the response E of entire image;
Step 6:Carrying out piecemeal to image, (the same step 1) of block size calculates the mean value and variance of each image block, to all pieces
Mean value and variance are summed respectively, obtain the mean value C and variance V of entire image;
Step 7:E in step 5 is normalized in the C and V obtained with step 6, obtains final fuzzy score S;
Step 8:By comparing the size of S and the threshold value T of selection, image is divided into clear and fuzzy two classes.
2. the fuzzy detection method that the SVD according to claim 1 based on Image DCT Domain is decomposed, it is characterised in that step 1
Detailed process is as follows:
The gradient map G of testing image I is calculated, formula is as follows:
Ix=I* [- 101], Iy=I* [- 101]T
Wherein * is convolution operation, then carries out piecemeal to gradient image G, and each piecemeal is set as Bk, k=(1,2 ..., N), N is
The total number of image block, block size are p × p, p=6 in experiment.
3. the fuzzy detection method that the SVD according to claim 2 based on Image DCT Domain is decomposed, it is characterised in that step 2
Detailed process is as follows:
By image block BkTransform to DCT domain Dk, and removing DC coefficient, formula is as follows:
Dk=DCT (Bk)
Wherein i, j ∈ { 1...p }.
4. the fuzzy detection method that the SVD according to claim 3 based on Image DCT Domain is decomposed, it is characterised in that step 3
Detailed process is as follows:
Difference matrix horizontally and vertically is calculated separately, formula is as follows:
Wherein i ∈ 1 ..., p }, j ∈ 1 ..., p-1 };
Wherein i ∈ 1 ..., p-1 }, j ∈ 1 ..., p }.
5. the fuzzy detection method that the SVD according to claim 4 based on Image DCT Domain is decomposed, it is characterised in that step 4
Detailed process is as follows:
The singular value of difference matrix is calculated, formula is as follows:
(:) matrix is indicated to change into a column vector, the size of F is p (p-1) × 2, and then carrying out singular value decomposition to F obtains
Singular value s1,s2;The response of image block is obtained by receptance function e;
ek=s1×s2-α(s1+s2)2
Wherein α is constant, α=0.01 in experiment.
6. the fuzzy detection method that the SVD according to claim 5 based on Image DCT Domain is decomposed, it is characterised in that step 5
Detailed process is as follows:
The response for calculating all pieces of testing image is summed, and obtains the overall response E of whole image, formula is as follows:
Wherein N is the total number of image block.
7. the fuzzy detection method that the SVD according to claim 6 based on Image DCT Domain is decomposed, it is characterised in that step 6
Detailed process is as follows:
To testing image piecemeal, process calculates every piece of mean value c with step 1kWith variance vk, then calculate all pieces of mean value
The sum of C, the sum of variance V, formula it is as follows:
8. the fuzzy detection method that the SVD according to claim 7 based on Image DCT Domain is decomposed, it is characterised in that step 7
Detailed process is as follows:
E is normalized with C and V, obtains final fuzzy score S, formula is as follows:
9. the fuzzy detection method that the SVD according to claim 8 based on Image DCT Domain is decomposed, it is characterised in that step 8
Detailed process is as follows:
The threshold value T between clear image and blurred picture is determined first;Clear image and blurred picture can be obtained by step 8
Fuzzy score S analyzes the data S that this two classes image obtains, it can be seen that clear image is more much higher than the S of blurred picture;For
Ensure the abundant of experimental data, great amount of samples is chosen in LIVE image libraries and is tested, determines last threshold value T;Finally
The threshold value selected is T=15, when T > 15 indicate that image is clearly, when T≤15 indicates that image is fuzzy.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810313311.1A CN108510496B (en) | 2018-04-09 | 2018-04-09 | Fuzzy detection method for SVD (singular value decomposition) based on image DCT (discrete cosine transformation) domain |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810313311.1A CN108510496B (en) | 2018-04-09 | 2018-04-09 | Fuzzy detection method for SVD (singular value decomposition) based on image DCT (discrete cosine transformation) domain |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108510496A true CN108510496A (en) | 2018-09-07 |
CN108510496B CN108510496B (en) | 2020-09-22 |
Family
ID=63381063
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810313311.1A Active CN108510496B (en) | 2018-04-09 | 2018-04-09 | Fuzzy detection method for SVD (singular value decomposition) based on image DCT (discrete cosine transformation) domain |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108510496B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110211105A (en) * | 2019-05-24 | 2019-09-06 | 厦门美柚信息科技有限公司 | A kind of blurred picture detection method, apparatus and system |
CN110390700A (en) * | 2019-07-05 | 2019-10-29 | 浙江大学 | One kind being based on the accurate more material decomposition methods of dual intensity CT of the matched image area of block |
CN110619647A (en) * | 2019-09-16 | 2019-12-27 | 中山大学 | Method for positioning fuzzy region of image based on combination of edge point frequency domain and spatial domain characteristics |
CN110782442A (en) * | 2019-10-23 | 2020-02-11 | 国网陕西省电力公司宝鸡供电公司 | Image artificial fuzzy detection method based on multi-domain coupling |
CN114972141A (en) * | 2022-05-13 | 2022-08-30 | 华侨大学 | Double-mode focusing analysis method of re-fuzzy theory |
CN115049658A (en) * | 2022-08-15 | 2022-09-13 | 合肥的卢深视科技有限公司 | RGB-D camera quality detection method, electronic device and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101950422A (en) * | 2010-09-29 | 2011-01-19 | 北京航空航天大学 | Singular value decomposition(SVD)-based image quality evaluation method |
US8111929B2 (en) * | 2008-02-29 | 2012-02-07 | Interra Systems Inc. | Method and system to detect and quantify blockiness in video files |
CN103093430A (en) * | 2013-01-25 | 2013-05-08 | 西安电子科技大学 | Heart magnetic resonance imaging (MRI) image deblurring method based on sparse low rank and dictionary learning |
CN106485703A (en) * | 2016-09-30 | 2017-03-08 | 杭州电子科技大学 | Fuzzy detection method based on image gradient dct transform |
-
2018
- 2018-04-09 CN CN201810313311.1A patent/CN108510496B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8111929B2 (en) * | 2008-02-29 | 2012-02-07 | Interra Systems Inc. | Method and system to detect and quantify blockiness in video files |
CN101950422A (en) * | 2010-09-29 | 2011-01-19 | 北京航空航天大学 | Singular value decomposition(SVD)-based image quality evaluation method |
CN103093430A (en) * | 2013-01-25 | 2013-05-08 | 西安电子科技大学 | Heart magnetic resonance imaging (MRI) image deblurring method based on sparse low rank and dictionary learning |
CN106485703A (en) * | 2016-09-30 | 2017-03-08 | 杭州电子科技大学 | Fuzzy detection method based on image gradient dct transform |
Non-Patent Citations (2)
Title |
---|
LEIDA LI ET AL.: ""No-Reference and Robust Image Sharpness Evaluation Based on Multiscale Spatial and Spectral Features"", 《IEEE TRANSACTIONS ON MULTIMEDIA》 * |
桑庆兵: ""半参考和无参考图像质量评价新方法研究"", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110211105A (en) * | 2019-05-24 | 2019-09-06 | 厦门美柚信息科技有限公司 | A kind of blurred picture detection method, apparatus and system |
CN110390700A (en) * | 2019-07-05 | 2019-10-29 | 浙江大学 | One kind being based on the accurate more material decomposition methods of dual intensity CT of the matched image area of block |
CN110619647A (en) * | 2019-09-16 | 2019-12-27 | 中山大学 | Method for positioning fuzzy region of image based on combination of edge point frequency domain and spatial domain characteristics |
CN110619647B (en) * | 2019-09-16 | 2022-12-30 | 中山大学 | Method for positioning fuzzy region of image based on combination of edge point frequency domain and spatial domain characteristics |
CN110782442A (en) * | 2019-10-23 | 2020-02-11 | 国网陕西省电力公司宝鸡供电公司 | Image artificial fuzzy detection method based on multi-domain coupling |
CN110782442B (en) * | 2019-10-23 | 2023-03-24 | 国网陕西省电力公司宝鸡供电公司 | Image artificial fuzzy detection method based on multi-domain coupling |
CN114972141A (en) * | 2022-05-13 | 2022-08-30 | 华侨大学 | Double-mode focusing analysis method of re-fuzzy theory |
CN115049658A (en) * | 2022-08-15 | 2022-09-13 | 合肥的卢深视科技有限公司 | RGB-D camera quality detection method, electronic device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108510496B (en) | 2020-09-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108510496A (en) | The fuzzy detection method that SVD based on Image DCT Domain is decomposed | |
Qureshi et al. | A critical survey of state-of-the-art image inpainting quality assessment metrics | |
CN102799669B (en) | Automatic grading method for commodity image vision quality | |
Fan et al. | No reference image quality assessment based on multi-expert convolutional neural networks | |
CN108052980B (en) | Image-based air quality grade detection method | |
CN101877127B (en) | Image reference-free quality evaluation method and system based on gradient profile | |
CN103853724B (en) | multimedia data classification method and device | |
CN105740945A (en) | People counting method based on video analysis | |
CN107886507B (en) | A kind of salient region detecting method based on image background and spatial position | |
CN111612741B (en) | Accurate reference-free image quality evaluation method based on distortion recognition | |
CN105007488A (en) | Universal no-reference image quality evaluation method based on transformation domain and spatial domain | |
CN110349099A (en) | A kind of complex scene video shadow Detection and removing method | |
CN112184672A (en) | No-reference image quality evaluation method and system | |
Ma et al. | Efficient saliency analysis based on wavelet transform and entropy theory | |
CN104537381A (en) | Blurred image identification method based on blurred invariant feature | |
Chen et al. | Edge preservation ratio for image sharpness assessment | |
CN110351453A (en) | A kind of computer video data processing method | |
Yang et al. | EHNQ: Subjective and objective quality evaluation of enhanced night-time images | |
CN106600615A (en) | Image edge detection algorithm evaluation system and method | |
Yuan et al. | Locally and multiply distorted image quality assessment via multi-stage CNNs | |
Jin | Complex impulse noise removal from color images based on super pixel segmentation | |
CN109741300B (en) | Image significance rapid detection method and device suitable for video coding | |
CN111275687A (en) | Fine-grained image stitching detection method based on connected region marks | |
Hua et al. | Iterative residual network for image dehazing | |
Li et al. | Blur-specific no-reference image quality assessment: A classification and review of representative methods |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |