CN101540048A - Image quality evaluating method based on support vector machine - Google Patents
Image quality evaluating method based on support vector machine Download PDFInfo
- Publication number
- CN101540048A CN101540048A CN200910082608A CN200910082608A CN101540048A CN 101540048 A CN101540048 A CN 101540048A CN 200910082608 A CN200910082608 A CN 200910082608A CN 200910082608 A CN200910082608 A CN 200910082608A CN 101540048 A CN101540048 A CN 101540048A
- Authority
- CN
- China
- Prior art keywords
- vector machine
- support vector
- image
- class
- training
- 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
Images
Landscapes
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides an image quality evaluating method based on a support vector machine. The method comprises the following steps: first, a preprocessed image sample is selected and extracted according to characteristic value, a processed sample set is respectively divided into a training set and a testing set; secondly, the training set is used for training the support vector machine, the number of the support vector machine is ensured according to a certain level which is required by a system, thus ensuring each support vector machine to be trained, wherein, an input sample is the characteristic value of the image and an output sample is the level of the image quality; thirdly, the trained support vector machine is used for adjusting and optimizing correlation parameters with the testing set and determining the parameter of the decision function of the optimal hyperplane of the support vector machine model; and finally, the support vector machine model which is trained and optimized is used for evaluating the quality level of the image sample. The invention has the advantages of little required sample, fast arithmetic speed, high precision, good performance, strong popularization, etc.
Description
Technical field
The invention belongs to technical field of image processing, be specifically related to a kind of image quality evaluating method based on support vector machine.
Background technology
Correct evaluation to picture quality is research topic highly significant in the image information engineering field.Image quality evaluating method generally is divided into subjective and objective two classes.Image finally is to watch for the people, so its quality evaluation method the most accurately is subjective assessment, but there are problems in actual applications in subjective evaluation method, so people are devoted to design the objective appraisal method untiringly with the requirement of approximate reflection subjective feeling.
Now carried out the pre-appraisal of remote sensing images classification in the world based on the decipher degree, part country has formulated based on the picture element of decipher degree and has estimated and evaluation method and system, and the many military units of China also propose to set up the demand based on the remote sensing images grade scale of decipher degree one after another.The decipher of remote sensing images is that the characteristic information of the various recognition objectives that provide by remote sensing images is analyzed, reasoning and judgement, finally reaches the purpose that recognition objective or scene are understood.Remote sensing images are according to the geological effect of obtaining, and the decipher degree is divided into four classes:
1. decipher is good: geologic body details and tectonic structure profile all can get access to from image, can work out out more complete decipher geological map;
2. decipher is medium: main tectonic structure and the main situation of geologic body can obtain from image, can only work out out rough decipher geological map;
3. decipher difficulty: can only find out part tectonic structure and a small amount of geology key element details, only can work out the decipher geologic scheme of summary;
4. decipher is difficult especially: can only find out a spot of geology key element, can not form complete tectonic structure notion, can't work out the geologic scheme of chi decipher in proportion.
The U.S. national defense aerial reconnaissance is deployed on 1974 and has just issued based on national image interpretation scale standard (the NationalImagery Interpretability Rating Scale that uses, abbreviate NIIRS as) as a kind of quantitative subjective picture quality standard, user's mission requirements are linked up with Remote Sensing Image Quality; This standard was adopted by NATO's (North AtlanticTreaty Organization abbreviates NATO as) in 1978, was called image interpretation scale standard (Imagery InterpretabilityRating Scale abbreviates IIRS as).Nineteen ninety-five has been issued civilian NIIRS, and the intelligence value that it has expressed image has embodied the literal requirement of intelligence community to reconnaissance image, has constituted the standard language that exchanges between user and the development department.China military has also set up Chinese image interpretation scale standard (Chinese Imagery Interpretability Rating Scale, abbreviate CIIRS as), be similar to the NIIRS of the U.S., be kind to estimate and appraisal procedure and system based on the picture element of image interpretation degree, the optimal design of guiding optics remote sensor scientifically.
These present methods are though the characteristic index difference of the development sample image collection of reference all adopts linear regression method to develop the picture quality equation; Although these methods have been set up the picture quality Prediction System more accurately, have shortcomings such as the sample image of needs is many, deal with data is many, arithmetic speed is slow, accuracy has much room for improvement.
Have at present kind of improved method be seek a kind of need sample still less, processing speed is faster, result is more accurate machine learning method, for example: though fuzzy system Fuzzy System and gray system Grey System often use in classification, accuracy is not high; Artificial neural network (Artificial Neural Network, abbreviate ANN as) though on non-linear and classification problem, have good advantage, but generalization ability is poor, local smallest point is arranged, and wherein generalization ability is meant that trained network also can provide the character of suitable output for the input that is not sample set; Hidden Markov method Hidden MarkovModel sets up and the training time requires long.
Summary of the invention
The objective of the invention is at the deficiencies in the prior art: handle that slow, poor reliability, generalization ability are poor, subjective assessment and objectively estimate inconsistent shortcoming, a kind of image quality evaluating method based on support vector machine (SVM, full name are Support Vector Machine) is proposed.
Be the purpose that is able to exactly picture quality be made an appraisal, the method for the invention has following steps:
After image pattern carried out denoising, select its some eigenwert, carry out certain pre-service after the extraction.Estimate rank in conjunction with subjective quality again, constitute sample set this image pattern.Then this sample set is divided into training set and test set two parts.
Wherein, image pattern is the picture after the evaluation of process expert subjective quality, and other number of samples difference of each quality assessment level is little; Pre-service is the normalized of data.
Wherein, some eigenwert is meant contrast C ontrast, entropy Entropy, texture Texture and blur level Blur.
1) contrast C ontrast
Contrast refers in the luminance component of piece image, the measurement of different brightness levels between light and shade zone most bright value and the darkest value, and the big more representative contrast of disparity range is big more, and the more little representative contrast of disparity range is more little.
The simultaneous contrast refers to that brightness value is the poor ratio of gray scale between image object and background, is one of estimating of image contrast.If B
1, B
2Be respectively the brightness value of target and background, then simultaneous contrast C is defined as:
C＝(B
1-B
2)/B
2
The possible degree that target is identified in the size reflection image of C value, the C value is big more, and this target is easy more to be identified.
2) entropy Entropy
The entropy of image is a kind of statistical form of feature, and it has reflected what of average information in the image.
The quantity of information that aggregation characteristic comprised of intensity profile makes P in the one dimension entropy presentation video of image
iGray-scale value is the shared ratio of pixel of i in the presentation video, and the monobasic gray scale entropy H that then defines gray level image is:
The one dimension entropy of image can the presentation video intensity profile aggregation characteristic, but can not reflect the space characteristics that gradation of image distributes, in order to characterize this space characteristics, can on the basis of one dimension entropy, introduce the two-dimensional entropy that the characteristic quantity that can reflect the intensity profile space characteristics comes the composition diagram picture.
3) texture Texture
What image texture reflected is a kind of partial structurtes feature of image, be embodied in certain variation of image slices vegetarian refreshments neighborhood interior pixel point gray level or color, and this variation is that spatial statistics is relevant, and it is made of two key elements of arrangement of texture primitive and primitive.
4) blur level Blur
Fuzzy is when image compresses by wave filter or through vision data, and owing to the decay of spatial domain HFS causes, its feature is exactly losing of edge smearing and detailed information.At present, the blur level of measurement image or video flowing does not also have the method for what maturation, and most of method need be carried out a large amount of loop iteration computings, and is not suitable for Real-Time Evaluation.
The thinking of blur level measuring method, the sharp edges by wave filter or compression can become level and smooth or smearing is arranged, so judge ambiguity by the degree of measuring border extended.The specific algorithm flow process is: seek earlier the vertical strong edge of handling the back image, to handling each the qualified edge in the image of back, all find the reference position and the edge calculation width at edge, blur level is exactly all border widths and the ratio of number of edges so.
Step 2 is determined the number of support vector machine;
The category level N of system requirements determines the number of support vector machine as required, N 〉=2.
Wherein, the category level N of system is the quality assessment rank of image, and excellent such as being divided into, good, neutralization differs from 4 grades, promptly this moment N=4.
Wherein, the category level of described system requirements as required determines that the number of support vector machine is meant: support vector machine is two class sorters, be applied to the above time-like that divides of two classes and two classes, several method is arranged, the category level that supposing the system requires has N 〉=2 kind, for the i ∈ in the N class [1, N] class is separated with other classifications, mainly utilizing 1-a-r is that 1-aginst-rest and 1-a-1 are two kinds of methods of 1-aginst-1.Wherein, 1-a-r is meant the two class sorter for N class problem structure N, i SVM with the training sample in the i class as positive training sample, and with other sample as the training sample of bearing, exporting at last is that class that two class sorters are output as maximum; 1-a-1 is meant all possible two class sorters of structure in N class training sample, and each SVM only trains on the 2 class training samples in the N class, and the result constructs K=N (N-1)/2 sorter altogether.
Step 3, support vector machine training and optimization part;
Utilize training set that each support vector machine is trained respectively.In the training, input is the eigenwert of image pattern, and output is the evaluation rank of picture quality, and promptly this invention utilization level numeral replaces quality assessment, and as excellent with 1 representative, 2 representatives are good, and in 3 representatives, 4 representatives are poor.Utilize test set that the supporting vector machine model correlation parameter that obtains is adjusted optimization then, determine the correlation parameter of the decision function of supporting vector machine model optimal classification face, comprise classification and parameter, penalty factor, Lagrange multiplier and the displacement factor of seeking only kernel function.
Step 4, the support vector machine applying portion;
Application is finished the supporting vector machine model of training image pattern is graded;
Wherein, the described supporting vector machine model of finishing training can both be made quality assessment to arbitrary image of input.
The present invention has following advantage:
1) taken into full account the influential characteristic parameter of picture quality, increased the reliability of image quality evaluation;
2) a kind of image quality evaluation system has been proposed, with subjective assessment and objective estimate effectively combine;
3) utilize that the support vector machine processing speed is fast, the classify accuracy advantages of higher is carried out quality grading;
4) be generalized to two classes of a plurality of support vector machine combinations and classify more than two classes from the classification of two classes of support vector machine.
Description of drawings
Fig. 1 is the method for the invention process flow diagram;
Fig. 2 is sample pre-service and feature extraction schematic flow sheet;
Fig. 3 is a support vector machine theoretical principle synoptic diagram;
Fig. 4 is the inner principle schematic of support vector machine;
Fig. 5 is support vector machine training schematic flow sheet;
Fig. 6 is a support vector machine application flow synoptic diagram.
Embodiment
Below in conjunction with accompanying drawing, the specific embodiment of the present invention is elaborated.
Image quality evaluating method of the present invention can be divided into four steps to be finished, steps flow chart as shown in Figure 1:
As shown in Figure 2, image pattern is carried out extracting contrast C ontrast, entropy Entropy, texture Texture and four eigenwerts of blur level Blur, constitutive characteristic value vector (p1 after the denoising, p2, p3 p4), and carries out normalization so that post-processed to each coordinate.
Image pattern is carried out subjective expert's quality assessment.Meaningful on statistics for guaranteeing subjective assessment, should consider to have untrained " layman " observer when selecting the observer, considering again has pair image technique that " expert " observer of certain experience is arranged; In addition, the observer who participates in scoring will have 20 at least, and test condition should be complementary with service condition as far as possible.The observer judges looked like to make quality by evaluation map according to the experience of oneself; Also can provide one group of standard picture as a reference, help the observer that picture quality is made suitable evaluation; And will estimate good in difference be converted to corresponding classification y ∈ 1,2,3,4} so that and the classification of support vector machine combine.
(p3 p4) imports as the four-dimension of sample for p1, p2, and evaluation result y constitutes sample set as output with the eigenwert vector.To all extract 3/4 out as training set in the pairing 4 kinds of samples of all kinds of evaluation y, all the other are 1/4 as test set.
Step 2 is determined the number of support vector machine;
Support vector machine is based on that two class linear separability problems in the pattern-recognition propose.Two class classification problems are converted into a quadratic programming problem to be solved.Suppose training sample (x
1, y
1) .., (x
1, y
1), x
i∈ R
n, y
i∈+1 ,-1}, i ∈ 1 ..., l}, wherein, R
nBe n dimension real number.As shown in Figure 3, sorting track l1 and sorting track l2 can correctly be divided into these samples type1, type2 two classes, belong to type1, then y
iBe labeled as+1; Belong to type2, then y
iBe labeled as-1.Such sorting track has many, but cut-off rule l1 makes and the interval maximum of two class samples be defined as the optimal classification line that the next optimal classification lineoid that is defined as of higher-dimension situation is called for short the optimal classification face.
Suppose to have the classification lineoid:
(w·x)+b＝0
Wherein, (wx) inner product between vectorial w of expression and the x, x ∈ { x
i.This classification lineoid can have correct two classes of dividing of training sample
Close and be written as y
i[(wx)+b] 〉=1, i=1,2 ..., l.
Make g (x)=(wx)+b, the defining classification function is f (x)=sgn (g (x)), and sgn () is the symbol discriminant function.| g (x) |=1 sample point, separate class linear distance minimum, determined the optimal classification line, be referred to as support vector.The higher-dimension situation is next has determined the optimal classification face.
According to the distance relation of putting the plane, make two class sample interval maximums as shown in Figure 3, need make the 1/||w|| maximum, promptly need to make || the w|| minimum,
Minimum.The problem equivalent of then asking the optimal classification lineoid is in following optimization problem
s.t|y
i[(w·x)+b]≥1，i＝1，2，...，l
The saddle point of separating by following Lagranian functional of this optimization problem provides:
Wherein, a
iBe Lagrangian coefficient, w and b are asked local derviation respectively, get the saddle point place
a
i〉=0, i=1 .., l; In the expansion of w, has only the expansion coefficient a of support vector correspondence
iNon-zero, therefore,
a
i〉=0.
Can be in the hope of the optimal classification function:
The optimal classification function is also referred to as the decision function of optimal classification face.Wherein, the x in the equation
iHave only support vector, promptly i take from 1 ..., the part of l}, a
iBe the pairing Lagrangian coefficient of obtaining of support vector,
It is a constant.The internal arithmetic process is introduced kernel function as shown in Figure 4 during actual the realization, with the inner product of vector kernel function K (x
i, x) replace, promptly
Support vector machine is a kind of two class sorters, and how support vector machine being generalized to two classes effectively and classifying more than two classes is current research focus.The building method of two classes and the above support vector machine classifier of two classes generally comprises two kinds of strategies:
(1) a series of two class svm classifier devices of structure, each sorter is used for identification two classifications wherein, and their differentiation result combined in some way realizes two classes and classify more than two classes;
(2) parametric solution with a plurality of classifying faces merges in the optimization problem, realizes two classes and the above classification of two classes by finding the solution this optimization problem " disposable ".
For these two class methods, second class methods far away more than first kind method, further cause training speed slow for the variable in the optimization problem solution procedure, and nicety of grading is also poor; Therefore existing most methods all belongs to first class.
In first kind, can construct or make up a plurality of two class sorters according to two kinds of algorithms usually and classify:
It is 1-aginst-rest that (1) first kind of algorithm is called 1-a-r, for N two class sorters of N class problem structure, i SVM with the training sample in the i class as positive training sample, and with other sample as negative training sample, output at last is that two class sorters are output as that maximum class.
(2) another kind of algorithm is that 1-a-1 is 1-aginst-1, promptly in N class training sample, construct all possible two class sorters, each SVM only trains on the 2 class training samples in the N class, the result constructs K=N (N-1)/2 sorter altogether, make up these two classes sorters with the ballot method, who gets the most votes's class is the class under the new point; If but two classes have same poll, can select the less class of index value.This ballot method is called the maximum method that is dominant.Two classes and the above classification problem of two classes that solve SVM by this method are a kind of methods relatively more commonly used, and compare with two classes and the above sorting technique of two classes of 1-a-r and directed acyclic graph method DAG_SVM, when handling mass data, the ballot method can access higher nicety of grading; Wherein directed acyclic graph method DAG_SVM is meant a new sample is carried out the branch time-like, at first from the sorter at top, classifier result according to the top, adopt two sorters of lower floor to continue classification, then these two sorters are continued classification with two sorters respectively again, till the affiliated classification of bottom.Though the 1-a-1 algorithm needs with N (N-1)/2 sorter the sample training, so but the sample data of each sorter training is all fewer owing to only taking from two classes, so the time of whole training is also few comparatively speaking, so the most suitable actual two classes and the above problem of two classes of solving.
What utilize among the present invention is the method for 1-a-1, if divide 4 classes, then needs to train 4*3/2=6 support vector machine, all will train separately each support vector machine.
Step 3, support vector machine training and optimization part;
Tentatively determine the correlation parameter of the decision function of optimal classification face by training set.In training process, adopt gridding method to determine optimum kernel function parameter and penalty factor, in the hope of obtaining optimum classifying quality.Wherein, gridding method refers to for several numbers within the specific limits, distinguishing value at regular intervals in the interval separately, forms netted value condition at last, selects optimum solution by more last result.Utilize test set that these parameters are adjusted optimization then, improve accuracy rate.
Support vector machine needs kernel function K (x
i, x)=Φ (x
i) Φ (x) realizes the mapping from the original space to the feature space, promptly can be used as kernel function as long as satisfy the symmetric function of Mercer condition.
(x y) is commonly referred to a kernel function to a binary function K.Given nuclear K (x, y), if having real number λ and nonvanishing function ψ (x) to make establishment
Claim that then λ is an eigenwert of nuclear, claims ψ (x) fundamental function about eigenvalue for nuclear.About Mercer nuclear following theorem is arranged, Mercer theorem: Mercer nuclear K (x y) can be launched into uniformly convergent series of functions:
λ wherein
i, ψ (x) is respectively nuclear K, and (their number may be limited or infinite for x, eigenwert y) and proper vector.
Kernel function commonly used has:
1) polynomial kernel function:
K(x
i，x
j)＝(μx
i·x
j+c)
d d＝1，2，…
2) radially basic kernel function:
K(x
i，x
j)＝exp(-γ||x
i-x
j||
2)
3) Sigmoid kernel function:
K(x
i，x
j)＝tanh(β(x
i·x
j+e)
More than all parameters of each kernel function be defaulted as real number.Because radially the parameter of basic kernel function is few, good classification effect is so the kernel function of support vector machine is selected radially basic kernel function K (x for use in this invention
i, x
j)=exp (γ || x
i-x
j||
2).
Because the optimal classification face may wrongly divide some sample points, has therefore quoted relaxation factor ξ
iWith penalty factor C, be used for the optimal classification face is adjusted.The problem equivalent of optimal classification face is in following optimization problem at this moment
s.t|y
i[(w·x)+b]≥1-ξ
i，i＝1，2，...，l，ξ
i≥0
At this moment, in the parameter of the decision function of the optimal classification face of obtaining, remove α
i〉=0 becomes 0≤α
i≤ C, all the other parameters are all constant, and the citation form of decision function is also in full accord.
As Fig. 5, among the present invention, the sample of support vector machine be image pattern (p1, p2, p3, p4, y), wherein (p3 p4) is pretreated eigenwert vector for p1, p2, and y is the image quality evaluation result.The training study of support vector machine is exactly the parameter γ and the penalty factor C that will find optimum radially basic kernel function so, support vector collection, Lagrangian coefficient a
iWith deviation ratio b.
Among the present invention, select the 1-a-1 algorithm for use, for each two class support vector machines, the image pattern eigenwert vector of these two kinds of quality that selection is corresponding and quality assessment rank are as sample set.Such as, for above-mentioned 1-2 support vector machine, it is 1 and 2 to be that quality assessment is excellent and good sample image that sample set is taken from quality assessment.For the y in the sample set, if from the image of 1 sample, the y order is 1; If from the image of 2 samples, the y order is for-1.For the 1-3 support vector machine, sample set take from quality assessment be 1 and 3 be quality assessment be excellent and in sample image.For y, if from 1 sample, the y order is 1; If from 3 samples, the y order is for-1.For the 1-4 support vector machine, it is 1 and 4 to be that quality assessment is excellent and poor sample image that sample set is taken from quality assessment.For y, if from 1 sample, the y order is 1; If from 4 samples, the y order is for-1.For the 2-3 support vector machine, sample set take from quality assessment be 2 and 3 be quality assessment be good and in sample image.For y, if from 2 samples, the y order is 1; If from 3 samples, the y order is for-1.For the 2-4 support vector machine, it is 2 and 4 to be that quality assessment is good and poor sample image that sample set is taken from quality assessment.For y, if from 2 samples, the y order is 1; If from 4 samples, the y order is for-1.For the 3-4 support vector machine, it is 3 and 4 to be that quality assessment is the sample image of neutralization difference that sample set is taken from quality assessment.For y, if from 3 samples, the y order is 1; If from 4 samples, the y order is for-1.In training process, at first adopt gridding method to determine the parameter γ and the penalty factor C of optimum radially basic kernel function, to guarantee higher classification accuracy.By the study of training set, search out the support vector collection then, tentatively determine Lagrange multiplier a
iWith displacement factor b, utilize test set that the supporting vector machine model correlation parameter that obtains is adjusted optimization again, finally determine the parameter of the decision function of supporting vector machine model optimal classification face.
Step 4, the support vector machine applying portion;
Among the present invention, owing to adopted the sorting technique of 1-a-1, so for every detected image, want in the decision function of each supporting vector machine model of substitution, finally from the result of these supporting vector machine models, determine the class categories of image by the ballot method, thereby determine the quality assessment of image, as Fig. 6.
In sum, compare traditional nonlinear fitting, the image quality evaluating method processing speed that the present invention adopts is faster, can Better by property, generalization ability is stronger, subjective assessment and objective estimate also can be basically identical.
Claims (3)
1, a kind of image quality evaluating method based on support vector machine is characterized in that, this method comprises following steps:
Step 1 is set up sample set;
At first, image pattern carried out denoising after, extract contrast C ontrast, entropy Entropy, texture Texture and four eigenwerts of blur level Blur of image, constitutive characteristic value vector (p1, p2, p3, p4), and to each coordinate carry out normalization so that post-processed;
Then, (p3 p4) as the four-dimension input of sample, carries out subjective expert's quality assessment to image pattern for p1, p2, and evaluation result y constitutes sample set as output with all kinds of eigenwert vectors;
At last, all extract 3/4 out as training set from the pairing sample set of all kinds of evaluation rank y, all the other are 1/4 as test set;
Step 2 is determined the number of support vector machine;
The number of support vector machine is determined in the level n of system 〉=2 as required; Wherein, the level n of system is the quality assessment rank of image;
Wherein, the rank of described system as required determines that the number of support vector machine is meant: support vector machine is two class sorters, be applied to the above time-like that divides of two classes, the category level that supposing the system requires has N 〉=2 kind, for with the i ∈ in the N class [1, N] class separates with other classifications, adopts 1-a-1 method construct two classes and the above support vector machine classifier of two classes among the present invention;
Step 3, support vector machine training and optimization part;
At first, utilize training set that each support vector machine is trained respectively, tentatively determine the correlation parameter of the decision function of optimal classification face; In the training, input is the eigenwert of image pattern, and output is the evaluation rank of picture quality, and promptly this invention utilization level numeral replaces quality assessment; Wherein adopt gridding method to determine optimum radially basic kernel function K (x
i, x
j)=exp (γ ‖ x
i-x
j‖
2) parameter γ and penalty factor C, in the hope of obtaining optimum classifying quality;
Then,, search out the support vector collection, tentatively determine Lagrange multiplier a by the study of training set
iWith displacement factor b;
At last, utilize test set that the supporting vector machine model correlation parameter that obtains is adjusted optimization, improve accuracy rate; Determine the correlation parameter of the decision function of supporting vector machine model optimal classification face;
Step 4, the support vector machine applying portion;
Application is finished the supporting vector machine model of training image pattern is graded;
Adopt the sorting technique of 1-a-1, for every detected image, in the decision function of each supporting vector machine model of substitution, finally from the result of these supporting vector machine models, determine the class categories of image, thereby determine the quality assessment rank of image by the ballot method;
Wherein, the described supporting vector machine model of finishing training is all exported quality assessment rank to this image to arbitrary image of input.
The valency rank.
2, a kind of image quality evaluating method according to claim 1 based on support vector machine, it is characterized in that, image pattern carried out subjective expert's quality assessment be meant described in the step 1, the observer who participates in scoring will have 20 at least, and test condition and service condition are complementary or are approximate; The observer judges looked like to make quality by evaluation map according to the experience of oneself; Perhaps provide one group of standard picture simultaneously as a reference, help the observer that picture quality is made suitable evaluation; And will estimate good in the difference be converted to corresponding classification y ∈ 1,2,3,4}, wherein y is an evaluation result.
3, a kind of image quality evaluating method according to claim 1 based on support vector machine, it is characterized in that, 1-a-1 described in the step 2 is the two class sorters that 1-aginst-1 is meant structure all situations in N class training sample, every class is only trained on the 2 class training samples in the N class, the result constructs K=N (N-1)/2 sorter altogether, with the ballot method is that the maximum method of being dominant makes up these two classes sorters, and who gets the most votes's class is the class under the new point; If two classes have same poll, select the less class of index value; Wherein K is the quantity of sorter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100826082A CN101540048B (en) | 2009-04-21 | 2009-04-21 | Image quality evaluating method based on support vector machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100826082A CN101540048B (en) | 2009-04-21 | 2009-04-21 | Image quality evaluating method based on support vector machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101540048A true CN101540048A (en) | 2009-09-23 |
| CN101540048B CN101540048B (en) | 2010-08-11 |
Family
ID=41123224
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009100826082A Expired - Fee Related CN101540048B (en) | 2009-04-21 | 2009-04-21 | Image quality evaluating method based on support vector machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101540048B (en) |
Cited By (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101847210A (en) * | 2010-06-25 | 2010-09-29 | 哈尔滨工业大学 | Multi-group image classification method based on two-dimensional empirical modal decomposition and wavelet denoising |
| CN101853401A (en) * | 2010-06-25 | 2010-10-06 | 哈尔滨工业大学 | Multi-packet image classification method based on two-dimensional empirical mode decomposition |
| CN101887523A (en) * | 2010-06-21 | 2010-11-17 | 南京邮电大学 | Method for detecting image spam email by picture character and local invariant feature |
| CN102156871A (en) * | 2010-02-12 | 2011-08-17 | 中国科学院自动化研究所 | Image classification method based on category correlated codebook and classifier voting strategy |
| CN102170130A (en) * | 2011-04-26 | 2011-08-31 | 华北电力大学 | Short-term wind power prediction method |
| CN101695035B (en) * | 2009-10-21 | 2012-07-04 | 成都市华为赛门铁克科技有限公司 | Flow rate identification method and device thereof |
| CN102749585A (en) * | 2011-04-21 | 2012-10-24 | 李昌 | Multi-layer SVM (support vector machine) based storage battery on-line monitoring method |
| CN102750695A (en) * | 2012-06-04 | 2012-10-24 | 清华大学 | Machine learning-based stereoscopic image quality objective assessment method |
| CN102799669A (en) * | 2012-07-17 | 2012-11-28 | 杭州淘淘搜科技有限公司 | Automatic grading method for commodity image vision quality |
| CN102833085A (en) * | 2011-06-16 | 2012-12-19 | 北京亿赞普网络技术有限公司 | System and method for classifying communication network messages based on mass user behavior data |
| CN103065118A (en) * | 2011-10-21 | 2013-04-24 | 索尼公司 | Image blurring detection method and device |
| CN103487463A (en) * | 2013-07-30 | 2014-01-01 | 中国标准化研究院 | Honey detection method using grid optimization-based selection of parameters of support vector machine classifier |
| CN103487464A (en) * | 2013-07-30 | 2014-01-01 | 中国标准化研究院 | Single-discrimination-based extraction method for features of intelligent olfaction spectrum used for characterization of variation of honey |
| CN103489003A (en) * | 2013-09-29 | 2014-01-01 | 华南理工大学 | Mobile phone image annotation method based on cloud computing |
| CN104320661A (en) * | 2014-10-29 | 2015-01-28 | 武汉大学 | Image coding quality predicting method based on difference entropy and structural similarity |
| CN104392446A (en) * | 2014-11-20 | 2015-03-04 | 江南大学 | Improved PSNR (Peak Signal to Noise Ratio)-based DCT (Discrete Cosine Transformation) domain non-reference blurred image quality evaluation method |
| CN104537385A (en) * | 2015-01-23 | 2015-04-22 | 湖南大学 | Method for evaluating DAGSVM classification accuracy |
| CN105118516A (en) * | 2015-09-29 | 2015-12-02 | 浙江图维电力科技有限公司 | Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) |
| CN105469384A (en) * | 2015-10-13 | 2016-04-06 | 南京邮电大学 | Comprehensive evaluation methods of license plate image quality |
| CN105608700A (en) * | 2015-12-24 | 2016-05-25 | 广州视源电子科技股份有限公司 | Photo screening method and system |
| CN105631457A (en) * | 2015-12-17 | 2016-06-01 | 小米科技有限责任公司 | Method and device for selecting picture |
| CN105718848A (en) * | 2015-10-21 | 2016-06-29 | 深圳芯启航科技有限公司 | Quality evaluation method and apparatus of fingerprint images |
| CN106529590A (en) * | 2016-10-28 | 2017-03-22 | 中国水产科学研究院渔业机械仪器研究所 | Moving shrimp seed recognition method based on improved PCA (Principal Component Analysis) and SVM (Support Vector Machine) algorithms |
| CN106651834A (en) * | 2016-10-20 | 2017-05-10 | 国网山东省电力公司电力科学研究院 | Method and device for evaluating quality of substation equipment infrared thermal image with no reference image |
| CN107016244A (en) * | 2017-04-10 | 2017-08-04 | 厦门波耐模型设计有限责任公司 | A kind of beauty and shaping effect evaluation system and implementation method |
| CN108062341A (en) * | 2016-11-08 | 2018-05-22 | 中国移动通信有限公司研究院 | The automatic marking method and device of data |
| CN108288027A (en) * | 2017-12-28 | 2018-07-17 | 新智数字科技有限公司 | A kind of detection method of picture quality, device and equipment |
| WO2018195891A1 (en) * | 2017-04-28 | 2018-11-01 | 深圳大学 | Method and apparatus for evaluating quality of non-reference image |
| CN109035217A (en) * | 2018-07-06 | 2018-12-18 | 同济大学 | A kind of soil mass crack quantitative evaluation method based on two-dimensional image entropy |
| CN109189767A (en) * | 2018-08-01 | 2019-01-11 | 北京三快在线科技有限公司 | Data processing method, device, electronic equipment and storage medium |
| CN109377465A (en) * | 2018-10-26 | 2019-02-22 | 北京布本智能科技有限公司 | A kind of image quality verification method based on image information entropy |
| CN109495772A (en) * | 2017-09-11 | 2019-03-19 | 优酷网络技术（北京）有限公司 | A kind of sort method and system of video quality |
| US10350438B2 (en) | 2015-06-30 | 2019-07-16 | Elekta Ltd. | System and method for target tracking using a quality indicator during radiation therapy |
| CN110084317A (en) * | 2019-05-06 | 2019-08-02 | 北京字节跳动网络技术有限公司 | The method and apparatus of image for identification |
| CN110514656A (en) * | 2019-09-03 | 2019-11-29 | 华中科技大学 | Method for analyzing long-life active constituent in Low Temperature Plasma Treating solution |
| CN110717892A (en) * | 2019-09-18 | 2020-01-21 | 宁波大学 | Tone mapping image quality evaluation method |
| CN111275683A (en) * | 2020-01-20 | 2020-06-12 | 上海云从汇临人工智能科技有限公司 | Image quality grading processing method, system, device and medium |
| CN111680582A (en) * | 2020-05-23 | 2020-09-18 | 安徽天宇羽绒有限公司 | Down feather detection method and system based on support vector machine |
| CN112329636A (en) * | 2020-11-06 | 2021-02-05 | 华侨大学 | Video quality assessment method and system combining support vector machine and fuzzy inference |
| CN112560870A (en) * | 2020-12-15 | 2021-03-26 | 哈尔滨工程大学 | Image target identification method used in underwater complex environment |
| CN112906910A (en) * | 2019-12-04 | 2021-06-04 | 北京沃东天骏信息技术有限公司 | Test method, system, device and electronic equipment |
| CN113288452A (en) * | 2021-04-23 | 2021-08-24 | 北京大学 | Operation quality detection method and device |
| CN113553971A (en) * | 2021-07-29 | 2021-10-26 | 青岛以萨数据技术有限公司 | Method and device for extracting optimal frame of face sequence and storage medium |
| CN116381374A (en) * | 2023-03-03 | 2023-07-04 | 国网四川省电力公司营销服务中心 | Transformer core operation state diagnosis method based on ultrasonic signal voiceprint recognition |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105224960B (en) * | 2015-11-04 | 2018-07-20 | 江南大学 | Corn seed classification hyperspectral imagery identification model update method based on clustering algorithm |
-
2009
- 2009-04-21 CN CN2009100826082A patent/CN101540048B/en not_active Expired - Fee Related
Cited By (65)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101695035B (en) * | 2009-10-21 | 2012-07-04 | 成都市华为赛门铁克科技有限公司 | Flow rate identification method and device thereof |
| CN102156871A (en) * | 2010-02-12 | 2011-08-17 | 中国科学院自动化研究所 | Image classification method based on category correlated codebook and classifier voting strategy |
| CN102156871B (en) * | 2010-02-12 | 2012-12-12 | 中国科学院自动化研究所 | Image classification method based on category correlated codebook and classifier voting strategy |
| CN101887523B (en) * | 2010-06-21 | 2013-04-10 | 南京邮电大学 | Method for detecting image spam email by picture character and local invariant feature |
| CN101887523A (en) * | 2010-06-21 | 2010-11-17 | 南京邮电大学 | Method for detecting image spam email by picture character and local invariant feature |
| CN101847210B (en) * | 2010-06-25 | 2012-10-03 | 哈尔滨工业大学 | Multi-group image classification method based on two-dimensional empirical modal decomposition and wavelet denoising |
| CN101847210A (en) * | 2010-06-25 | 2010-09-29 | 哈尔滨工业大学 | Multi-group image classification method based on two-dimensional empirical modal decomposition and wavelet denoising |
| CN101853401A (en) * | 2010-06-25 | 2010-10-06 | 哈尔滨工业大学 | Multi-packet image classification method based on two-dimensional empirical mode decomposition |
| CN102749585B (en) * | 2011-04-21 | 2014-11-05 | 李昌 | Multi-layer SVM (support vector machine) based storage battery on-line monitoring method |
| CN102749585A (en) * | 2011-04-21 | 2012-10-24 | 李昌 | Multi-layer SVM (support vector machine) based storage battery on-line monitoring method |
| CN102170130A (en) * | 2011-04-26 | 2011-08-31 | 华北电力大学 | Short-term wind power prediction method |
| CN102170130B (en) * | 2011-04-26 | 2013-03-06 | 华北电力大学 | Short-term wind power prediction method |
| CN102833085B (en) * | 2011-06-16 | 2015-09-16 | 北京亿赞普网络技术有限公司 | Based on communication network message categorizing system and the method for mass users behavioral data |
| CN102833085A (en) * | 2011-06-16 | 2012-12-19 | 北京亿赞普网络技术有限公司 | System and method for classifying communication network messages based on mass user behavior data |
| CN103065118A (en) * | 2011-10-21 | 2013-04-24 | 索尼公司 | Image blurring detection method and device |
| CN102750695A (en) * | 2012-06-04 | 2012-10-24 | 清华大学 | Machine learning-based stereoscopic image quality objective assessment method |
| CN102750695B (en) * | 2012-06-04 | 2015-04-15 | 清华大学 | Machine learning-based stereoscopic image quality objective assessment method |
| CN102799669B (en) * | 2012-07-17 | 2015-06-17 | 杭州淘淘搜科技有限公司 | Automatic grading method for commodity image vision quality |
| CN102799669A (en) * | 2012-07-17 | 2012-11-28 | 杭州淘淘搜科技有限公司 | Automatic grading method for commodity image vision quality |
| CN103487464B (en) * | 2013-07-30 | 2016-06-08 | 中国标准化研究院 | A kind of sign Mel diversity intelligence olfactory sensation TuPu method extracting method based on individual event diagnostic method |
| CN103487463A (en) * | 2013-07-30 | 2014-01-01 | 中国标准化研究院 | Honey detection method using grid optimization-based selection of parameters of support vector machine classifier |
| CN103487464A (en) * | 2013-07-30 | 2014-01-01 | 中国标准化研究院 | Single-discrimination-based extraction method for features of intelligent olfaction spectrum used for characterization of variation of honey |
| CN103487463B (en) * | 2013-07-30 | 2015-12-23 | 中国标准化研究院 | A kind of honey detection method of the support vector machine classifier Selecting parameter based on grid optimization |
| CN103489003A (en) * | 2013-09-29 | 2014-01-01 | 华南理工大学 | Mobile phone image annotation method based on cloud computing |
| CN103489003B (en) * | 2013-09-29 | 2017-04-19 | 华南理工大学 | Mobile phone image annotation method based on cloud computing |
| CN104320661A (en) * | 2014-10-29 | 2015-01-28 | 武汉大学 | Image coding quality predicting method based on difference entropy and structural similarity |
| CN104320661B (en) * | 2014-10-29 | 2017-04-26 | 武汉大学 | Image coding quality predicting method based on difference entropy and structural similarity |
| CN104392446A (en) * | 2014-11-20 | 2015-03-04 | 江南大学 | Improved PSNR (Peak Signal to Noise Ratio)-based DCT (Discrete Cosine Transformation) domain non-reference blurred image quality evaluation method |
| CN104537385A (en) * | 2015-01-23 | 2015-04-22 | 湖南大学 | Method for evaluating DAGSVM classification accuracy |
| RU2696428C2 (en) * | 2015-06-30 | 2019-08-01 | Электа Лтд. | Target object tracking system and method using a quality indicator during radiation therapy |
| US10350438B2 (en) | 2015-06-30 | 2019-07-16 | Elekta Ltd. | System and method for target tracking using a quality indicator during radiation therapy |
| CN105118516A (en) * | 2015-09-29 | 2015-12-02 | 浙江图维电力科技有限公司 | Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) |
| CN105469384A (en) * | 2015-10-13 | 2016-04-06 | 南京邮电大学 | Comprehensive evaluation methods of license plate image quality |
| CN105469384B (en) * | 2015-10-13 | 2018-09-25 | 南京邮电大学 | The integrated evaluating method of license plate image quality |
| CN105718848A (en) * | 2015-10-21 | 2016-06-29 | 深圳芯启航科技有限公司 | Quality evaluation method and apparatus of fingerprint images |
| CN105718848B (en) * | 2015-10-21 | 2020-08-07 | 深圳芯启航科技有限公司 | Quality evaluation method and device for fingerprint image |
| CN105631457A (en) * | 2015-12-17 | 2016-06-01 | 小米科技有限责任公司 | Method and device for selecting picture |
| CN105608700A (en) * | 2015-12-24 | 2016-05-25 | 广州视源电子科技股份有限公司 | Photo screening method and system |
| CN105608700B (en) * | 2015-12-24 | 2019-12-17 | 广州视源电子科技股份有限公司 | Photo screening method and system |
| CN106651834A (en) * | 2016-10-20 | 2017-05-10 | 国网山东省电力公司电力科学研究院 | Method and device for evaluating quality of substation equipment infrared thermal image with no reference image |
| CN106529590A (en) * | 2016-10-28 | 2017-03-22 | 中国水产科学研究院渔业机械仪器研究所 | Moving shrimp seed recognition method based on improved PCA (Principal Component Analysis) and SVM (Support Vector Machine) algorithms |
| CN108062341A (en) * | 2016-11-08 | 2018-05-22 | 中国移动通信有限公司研究院 | The automatic marking method and device of data |
| CN107016244A (en) * | 2017-04-10 | 2017-08-04 | 厦门波耐模型设计有限责任公司 | A kind of beauty and shaping effect evaluation system and implementation method |
| WO2018195891A1 (en) * | 2017-04-28 | 2018-11-01 | 深圳大学 | Method and apparatus for evaluating quality of non-reference image |
| CN109495772B (en) * | 2017-09-11 | 2021-10-15 | 阿里巴巴（中国）有限公司 | Video quality sequencing method and system |
| CN109495772A (en) * | 2017-09-11 | 2019-03-19 | 优酷网络技术（北京）有限公司 | A kind of sort method and system of video quality |
| CN108288027A (en) * | 2017-12-28 | 2018-07-17 | 新智数字科技有限公司 | A kind of detection method of picture quality, device and equipment |
| CN109035217A (en) * | 2018-07-06 | 2018-12-18 | 同济大学 | A kind of soil mass crack quantitative evaluation method based on two-dimensional image entropy |
| CN109189767A (en) * | 2018-08-01 | 2019-01-11 | 北京三快在线科技有限公司 | Data processing method, device, electronic equipment and storage medium |
| CN109189767B (en) * | 2018-08-01 | 2021-07-23 | 北京三快在线科技有限公司 | Data processing method and device, electronic equipment and storage medium |
| CN109377465A (en) * | 2018-10-26 | 2019-02-22 | 北京布本智能科技有限公司 | A kind of image quality verification method based on image information entropy |
| CN110084317A (en) * | 2019-05-06 | 2019-08-02 | 北京字节跳动网络技术有限公司 | The method and apparatus of image for identification |
| CN110514656A (en) * | 2019-09-03 | 2019-11-29 | 华中科技大学 | Method for analyzing long-life active constituent in Low Temperature Plasma Treating solution |
| CN110717892A (en) * | 2019-09-18 | 2020-01-21 | 宁波大学 | Tone mapping image quality evaluation method |
| CN110717892B (en) * | 2019-09-18 | 2022-06-28 | 宁波大学 | Tone mapping image quality evaluation method |
| CN112906910A (en) * | 2019-12-04 | 2021-06-04 | 北京沃东天骏信息技术有限公司 | Test method, system, device and electronic equipment |
| CN111275683A (en) * | 2020-01-20 | 2020-06-12 | 上海云从汇临人工智能科技有限公司 | Image quality grading processing method, system, device and medium |
| CN111680582A (en) * | 2020-05-23 | 2020-09-18 | 安徽天宇羽绒有限公司 | Down feather detection method and system based on support vector machine |
| CN112329636A (en) * | 2020-11-06 | 2021-02-05 | 华侨大学 | Video quality assessment method and system combining support vector machine and fuzzy inference |
| CN112329636B (en) * | 2020-11-06 | 2023-06-16 | 华侨大学 | Video quality assessment method and system combining support vector machine and fuzzy reasoning |
| CN112560870A (en) * | 2020-12-15 | 2021-03-26 | 哈尔滨工程大学 | Image target identification method used in underwater complex environment |
| CN112560870B (en) * | 2020-12-15 | 2022-04-29 | 哈尔滨工程大学 | Image target identification method used in underwater complex environment |
| CN113288452A (en) * | 2021-04-23 | 2021-08-24 | 北京大学 | Operation quality detection method and device |
| CN113553971A (en) * | 2021-07-29 | 2021-10-26 | 青岛以萨数据技术有限公司 | Method and device for extracting optimal frame of face sequence and storage medium |
| CN116381374A (en) * | 2023-03-03 | 2023-07-04 | 国网四川省电力公司营销服务中心 | Transformer core operation state diagnosis method based on ultrasonic signal voiceprint recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101540048B (en) | 2010-08-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101540048B (en) | Image quality evaluating method based on support vector machine | |
| CN107633513B (en) | 3D image quality measuring method based on deep learning | |
| CN105975916B (en) | Age estimation method based on multi output convolutional neural networks and ordinal regression | |
| CN109801256B (en) | Image aesthetic quality assessment method based on region of interest and global features | |
| CN111126258B (en) | Image recognition method and related device | |
| CN109035188B (en) | Intelligent image fusion method based on target feature driving | |
| CN103632132B (en) | Face detection and recognition method based on skin color segmentation and template matching | |
| CN110084294A (en) | A kind of Remote Image Classification based on multiple dimensioned depth characteristic | |
| CN110119728A (en) | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network | |
| CN108389220B (en) | Remote sensing video image motion target real-time intelligent cognitive method and its device | |
| CN108564049A (en) | A kind of fast face detection recognition method based on deep learning | |
| CN109740588A (en) | The X-ray picture contraband localization method reassigned based on the response of Weakly supervised and depth | |
| CN103971106B (en) | Various visual angles facial image gender identification method and device | |
| CN109902590A (en) | Pedestrian's recognition methods again of depth multiple view characteristic distance study | |
| CN106446872A (en) | Detection and recognition method of human face in video under low-light conditions | |
| CN106228142A (en) | Face verification method based on convolutional neural networks and Bayesian decision | |
| CN110781829A (en) | Light-weight deep learning intelligent business hall face recognition method | |
| CN103778435A (en) | Pedestrian fast detection method based on videos | |
| CN101739555A (en) | Method and system for detecting false face, and method and system for training false face model | |
| CN105427309A (en) | Multiscale hierarchical processing method for extracting object-oriented high-spatial resolution remote sensing information | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| CN107145836A (en) | Hyperspectral image classification method based on stack boundary discrimination self-encoding encoder | |
| CN104636755A (en) | Face beauty evaluation method based on deep learning | |
| CN106022254A (en) | Image recognition technology | |
| CN103208012A (en) | Light face recognition method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20100811 Termination date: 20120421 |