CN109948625A - Definition of text images appraisal procedure and system, computer readable storage medium - Google Patents
Definition of text images appraisal procedure and system, computer readable storage medium Download PDFInfo
- Publication number
- CN109948625A CN109948625A CN201910172031.8A CN201910172031A CN109948625A CN 109948625 A CN109948625 A CN 109948625A CN 201910172031 A CN201910172031 A CN 201910172031A CN 109948625 A CN109948625 A CN 109948625A
- Authority
- CN
- China
- Prior art keywords
- assessed
- text image
- prospect
- image
- text
- 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
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000003860 storage Methods 0.000 title claims abstract description 17
- 238000012545 processing Methods 0.000 claims abstract description 59
- 239000000284 extract Substances 0.000 claims abstract description 50
- 238000011156 evaluation Methods 0.000 claims abstract description 42
- 238000013210 evaluation model Methods 0.000 claims abstract description 32
- 238000001914 filtration Methods 0.000 claims abstract description 28
- 230000009466 transformation Effects 0.000 claims abstract description 24
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 23
- 230000011218 segmentation Effects 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims description 41
- 238000000605 extraction Methods 0.000 claims description 27
- 239000013598 vector Substances 0.000 claims description 26
- 238000009826 distribution Methods 0.000 claims description 11
- 238000007477 logistic regression Methods 0.000 claims description 11
- 238000001228 spectrum Methods 0.000 claims description 10
- 239000000203 mixture Substances 0.000 claims description 8
- 238000002407 reforming Methods 0.000 claims description 6
- 230000001052 transient effect Effects 0.000 claims 1
- 230000006870 function Effects 0.000 description 11
- 230000008878 coupling Effects 0.000 description 6
- 238000010168 coupling process Methods 0.000 description 6
- 238000005859 coupling reaction Methods 0.000 description 6
- 238000005259 measurement Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 239000003086 colorant Substances 0.000 description 2
- 238000005315 distribution function Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011067 equilibration Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000003780 insertion Methods 0.000 description 1
- 230000037431 insertion Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A kind of definition of text images appraisal procedure and system, computer readable storage medium, the definition of text images appraisal procedure, comprising: marginalisation processing and contours extract are carried out to the prospect of the text image to be assessed；Pixel segmentation, filtering are carried out to the prospect of the text image to be assessed after marginalisation processing and contours extract, position the circumference position of the prospect of the text image to be assessed；According to the circumference position, affine transformation is carried out to the text image to be assessed, the text image to be assessed after being reformed；Text image to be assessed after the reformation is input to preset intelligibility evaluation model；Export the intelligibility evaluation result of the text image to be assessed.Using the above scheme, it can be improved the accuracy of definition of text images identification.
Description
Technical field
The present embodiments relate to technical field of image processing more particularly to a kind of definition of text images appraisal procedure and
System, computer readable storage medium.
Background technique
Image definition evaluation and test be in the modern artificial intelligence field such as image procossing, pattern-recognition it is very important it is a kind of before
Processing links.In the application such as Taking Photographic, text identification, certificate identification, image classification, industrial production, security protection, image
Clarity is related to image quality, accuracy of identification and the execution efficiency of subsequent processing task.
Since the source of image and type are varied, the noise feature showed is also different and same.Currently, right
The accuracy of image definition assessment is lower.
Summary of the invention
The technical issues of embodiment of the present invention solves is the accuracy for how improving definition of text images identification.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of definition of text images appraisal procedure, comprising: right
The prospect of the text image to be assessed carries out marginalisation processing and contours extract；To after marginalisation processing and contours extract
The prospect of text image to be assessed carry out pixel segmentation, filtering, position the periphery of the prospect of the text image to be assessed
Outline position；According to the circumference position, affine transformation is carried out to the text image to be assessed, after being reformed to
Assess text image；Text image to be assessed after the reformation is input to preset intelligibility evaluation model；Described in output
The intelligibility evaluation result of text image to be assessed.
Optionally, the prospect of the text image to be assessed carries out marginalisation processing and contours extract, comprising: to it is described to
The prospect of assessment text image is handled, and the edge feature figure of the text image to be assessed is obtained；To the text to be assessed
The edge feature figure of this image carries out polygonal profile extraction, obtains the polygon set being made of multiple polygons.
Optionally, it is described to the prospect of text image to be assessed after marginalisation processing and contours extract carry out picture
Vegetarian refreshments segmentation, filtering, position the circumference position of the prospect of the text image to be assessed, comprising: for each polygon
Profile calculates the within angle on adjacent both sides according to the apex coordinate on adjacent both sides；When the within angle on adjacent both sides is less than default threshold
When value, the within angle is filtered, and iterate to calculate and obtain new side to replace the adjacent edge of the within angle；From filtered polygon
All quadrangles are selected in shape set, using the corresponding profile of the maximum quadrangle of area as the text image to be assessed
The circumference position of prospect.
Optionally, before the prospect for extracting the text image to be assessed, further includes: to the text image to be assessed
It is pre-processed, being equalized image；The equalization image is changed into the channel LAB from primary display channels；Based on changing into LAB
Equalization image after channel, extracts the feature of each pixel, in which: the feature of each pixel includes following at least one
Kind: the feature vector set and the corresponding measurement of each feature vector of the feature vector composition of each pixel.
Optionally, described that the text image to be assessed is pre-processed, being equalized image, comprising: to described
Text image to be assessed carries out histogram equalization processing；The primary display channels of the text image to be assessed are transformed into HSV
Channel；It takes the channel V to carry out brightness processed, obtains the new channel V；The new channel V is merged with the channel H and channel S, is gone back to
Primary display channels obtain the image of the equalization.
Optionally, the prospect for extracting text image to be assessed, comprising: be based on described eigenvector set, carry out nothing
Two taxonomic clusterings of supervision isolate the prospect and background of the text image to be assessed, and extract the text diagram to be assessed
The prospect of picture.
Optionally, after the prospect and background for isolating the text image to be assessed, further includes: execute to it is described to
The prospect for assessing text image highlights and background darkens operation.
Optionally, training obtains the intelligibility evaluation model in the following way: obtaining sample text image；To described
Sample text image carries out image procossing, extracts the prospect of the sample text image；To the prospect of the sample text image
Carry out marginalisation processing and contours extract；To the prospect of the sample text image after marginalisation processing and contours extract
Pixel segmentation, filtering are carried out, the circumference position of the prospect of the sample text image is positioned；According to the circumference
Position carries out affine transformation to the sample text image, the sample text image after being reformed；For the sample after reforming
The corresponding pixel value of each Color Channel of pixel in text image, the sharpness information for obtaining the sample text image refer to
Mark, in which: the sharpness information index of the sample text image comprises at least one of the following: point sharpness information index, high-order are empty
Domain evaluation index, spectrum information evaluation index and statistical distribution evaluation index；The sharpness information of obtained each pixel is referred to
Mark carries out logistic regression training respectively, obtains corresponding training parameter when the minimum value that training obtains, obtains the clarity and comment
Estimate model.
The embodiment of the present invention also provides a kind of definition of text images assessment system, comprising: foreground extraction unit, suitable for mentioning
Take the prospect of text image to be assessed；First processing units carry out marginalisation suitable for the prospect to the text image to be assessed
Processing and contours extract；Positioning unit is filtered, suitable for the text image to be assessed after marginalisation processing and contours extract
Prospect carry out pixel segmentation, filtering, position the circumference position of the prospect of the text image to be assessed；Affine transformation
Unit, is suitable for according to the circumference position, carries out affine transformation to the text image to be assessed, after being reformed to
Assess text image；Input unit, suitable for the text image to be assessed after the reformation is input to preset intelligibility evaluation
Model；Output unit, suitable for exporting the intelligibility evaluation result of the text image to be assessed.
Optionally, the first processing units handle suitable for the prospect to the text image to be assessed, obtain institute
The edge feature figure for stating text image to be assessed carries out polygonal profile to the edge feature figure of the text image to be assessed and mentions
It takes, obtains the polygon set being made of multiple polygons.
Optionally, the filtering positioning unit, suitable for the profile for each polygon, according to the apex coordinate on adjacent both sides
Calculate the within angle on adjacent both sides；When the within angle on adjacent both sides is less than preset threshold, the within angle, and iteration meter are filtered
Calculation obtains new side to replace the adjacent edge of the within angle；All quadrangles are selected from filtered polygon set, it will
Circumference position of the corresponding profile of the maximum quadrangle of area as the prospect of the text image to be assessed.
Optionally, the definition of text images assessment system further include: the second processing unit and feature extraction unit,
In: described the second processing unit is right suitable for before the prospect that the foreground extraction unit extracts the text image to be assessed
The text image to be assessed is pre-processed, being equalized image；The feature extraction unit is suitable for the equalization
Image changes into the channel LAB from primary display channels；Based on the equalization image changed into after the channel LAB, the spy of each pixel is extracted
Sign, in which: the feature of each pixel comprises at least one of the following: the set of eigenvectors of the feature vector composition of each pixel
Conjunction and the corresponding measurement of each feature vector.
Optionally, described the second processing unit is suitable for carrying out histogram equalization processing to the text image to be assessed；
The primary display channels of the text image to be assessed are transformed into the channel HSV；It takes the channel V to carry out brightness processed, it is logical to obtain new V
Road；The new channel V is merged with the channel H and channel S, primary display channels is gone back to, obtains the image of the equalization.
Optionally, the foreground extraction unit is suitable for being based on described eigenvector set, and it is poly- to carry out two unsupervised classification
Class isolates the prospect and background of the text image to be assessed, and extracts the prospect of the text image to be assessed.
Optionally, the foreground extraction unit is further adapted in the prospect and background for isolating the text image to be assessed
Later, it executes and background highlighted to the prospect of the text image to be assessed and darkens operation.
Optionally, the definition of text images assessment system further include: model training unit is suitable in the following way
Training obtains the intelligibility evaluation model: obtaining sample text image；Image procossing is carried out to the sample text image, is mentioned
Take the prospect of the sample text image；Marginalisation processing and contours extract are carried out to the prospect of the sample text image；It is right
The prospect of the sample text image after marginalisation processing and contours extract carries out pixel segmentation, filtering, positions institute
State the circumference position of the prospect of sample text image；According to the circumference position, to the sample text image into
Row affine transformation, the sample text image after being reformed；For each of the pixel in the sample text image after reforming
The corresponding pixel value of Color Channel obtains the sharpness information index of the sample text image using following at least one mode:
Point sharpness information index, high-order airspace evaluation index, spectrum information evaluation index and statistical distribution evaluation index；To obtained
The sharpness information index of each pixel carries out logistic regression training respectively, obtains corresponding training ginseng when the minimum value that training obtains
Number, obtains the intelligibility evaluation model.
The embodiment of the present invention also provides a kind of definition of text images assessment system, including memory and processor, described
The computer instruction that can be run on the processor is stored on memory, when the processor runs the computer instruction
The step of executing any of the above-described kind of definition of text images appraisal procedure.
The embodiment of the present invention also provides a kind of computer readable storage medium, and computer readable storage medium is non-volatile
Storage medium or non-transitory storage media, are stored thereon with computer instruction, and the computer instruction executes above-mentioned when running
A kind of the step of definition of text images appraisal procedure.
Compared with prior art, the technical solution of the embodiment of the present invention has the advantages that
After the prospect for extracting text image to be assessed, marginalisation processing is carried out to the prospect of image to be assessed and profile mentions
It takes, the circumference position of the prospect of text image to be assessed is hereafter positioned using pixel segmentation, filtering.Using pixel
Segmentation, filtering mode position the circumference position of the prospect of text image to be assessed, the standard of foreground extraction can be improved
The accuracy of the locations of contours of true property and prospect, so as to improve the accuracy of follow-up text image definition assessment.
Further, in training intelligibility evaluation model, a variety of sharpness information indexs are considered, so as to further increase
The accuracy of definition of text images assessment.
Detailed description of the invention
Fig. 1 is a kind of flow chart of definition of text images appraisal procedure in the embodiment of the present invention；
Fig. 2 is a kind of training flow chart of intelligibility evaluation model in the embodiment of the present invention；
Fig. 3 is a kind of flow chart of another kind definition of text images appraisal procedure in the embodiment of the present invention；
Fig. 4 is a kind of structural schematic diagram of definition of text images assessment system in the embodiment of the present invention.
Specific embodiment
As noted previously, as the source of image and type are varied, the noise feature showed it is also different and
Together.Currently, lower to the accuracy of image definition assessment.
In the embodiment of the present invention, after the prospect for extracting text image to be assessed, side is carried out to the prospect of image to be assessed
Edge processing and contours extract, the periphery that the prospect of text image to be assessed is hereafter positioned using pixel segmentation, filtering are taken turns
Wide position.The circumference position that the prospect of text image to be assessed is positioned by the way of pixel segmentation, filtering, can be with
The accuracy of the accuracy of foreground extraction and the locations of contours of prospect is improved, so as to improve follow-up text image definition
The accuracy of assessment.
It is understandable to enable the above-mentioned purpose, feature and beneficial effect of the embodiment of the present invention to become apparent, below with reference to attached
Figure is described in detail specific embodiments of the present invention.
Referring to Fig.1, a kind of flow chart of definition of text images appraisal procedure in the embodiment of the present invention is given, can wrap
Include following steps.
Step 11, the prospect of text image to be assessed is extracted.
In specific implementation, text image to be assessed can be collected by all kinds of mobile terminals.For example, being clapped by high
The equipment with image collecting function such as instrument, smart phone or insertion terminal shoot to obtain.Getting the text to be assessed
After image, the prospect of the text image to be assessed can be extracted.
It in specific implementation, can also be to the text to be assessed before the prospect for extracting the text image to be assessed
This image is pre-processed, being equalized image.
Specifically, histogram equalization processing is carried out to the text image to be assessed, by the text diagram to be assessed
The primary display channels of picture are transformed into the channel HSV, wherein H indicates that tone, S indicate saturation degree, and V indicates lightness.The channel V is taken to carry out
Brightness processed obtains the new channel V.The new channel V is merged with the channel H and channel S, primary display channels RGB is gone back to, obtains
To the equalization image.
After being equalized image, the equalization image can also be changed into the channel LAB from primary display channels.L
Indicate brightness, A and B indicate relevant color, and A indicates the range from carmetta to green, and B indicates the model from yellow to blue
It encloses.Based on the equalization image changed into after the channel LAB, the feature of each pixel is extracted.The feature of each pixel includes
Following at least one: the feature vector set and the corresponding measurement of each feature vector of the feature vector composition of each pixel.
Each pixel point value can regard one group of five dimensional vector, including three-dimensional channel and two-dimensional position information as.Based on each
The feature vector set of pixel carries out unsupervised two taxonomic clustering.In two unsupervised taxonomic clustering, k- can be used
Any one of means cluster, Gaussian Mixture cluster, laminar cluster, the DBSCAN based on density is clustered etc..It is real in the present invention one
It applies in example, is clustered using convergence rate and the preferable laminar of Clustering Effect.In two unsupervised taxonomic clustering, using each
The corresponding measurement of feature vector.
In specific implementation, it is based on unsupervised two taxonomic clustering, the prospect and back of text image to be assessed can be isolated
Scape, and extract the prospect of text image to be assessed.
In inventing an embodiment, in order to significantly distinguish the prospect and background of text image to be assessed, to prospect and
Background is marked, and based on the prospect and background after label, executes highlighted operation to prospect, executes to background and darken operation.
Step 12, marginalisation processing and contours extract are carried out to the prospect of the text image to be assessed.
In the present invention one is implemented, the prospect of text image to be assessed can be handled using Sobel operator, be obtained
Picture edge characteristic figure.Polygonal profile extraction is carried out on picture edge characteristic figure, obtains being made of multiple polygons more
Side shape set.
Step 13, pixel point is carried out to the prospect of the text image to be assessed after marginalisation processing and contours extract
It cuts, filter, position the circumference position of the prospect of the text image to be assessed.
In specific implementation, picture is carried out to the prospect of the text image to be assessed after marginalisation processing and contours extract
Vegetarian refreshments segmentation, filtering, by the pixel to the circumference position for forming text image prospect to be assessed carry out screening adjustment,
It deletes, filtering positioning obtains the circumference position of the prospect of the text image to be assessed.
In an embodiment of the present invention, the within angle that adjacent both sides are calculated according to the apex coordinate on adjacent both sides judges institute
The relationship between the within angle on adjacent both sides and the preset threshold of setting being calculated, when the within angle on adjacent both sides is less than in advance
If when threshold value, filtering the within angle, and iterate to calculate and obtain new side to replace the adjacent edge of the within angle.
Step 14, according to the circumference position, affine transformation is carried out to the text image to be assessed, is reformed
Text image to be assessed afterwards.
In specific implementation, according to the circumference position, affine transformation is carried out to text image to be assessed.It is described imitative
Penetrating transformation may include that translation, scaling, rotation or mistake at least one of are cut, by imitating the text image to be assessed
Penetrate the text image to be assessed after converting available reformation.
For example, text image to be assessed is invoice image, the hair in invoice Image Acquisition can be corrected by affine transformation
Ticket anamorphose obtains the invoice image of the rectangular shape to match with invoice true form.
Step 15, the text image to be assessed after the reformation is input to preset intelligibility evaluation model.
In specific implementation, the text image to be assessed after reformation can be inputted in preset intelligibility evaluation model,
Carry out the intelligibility evaluation of text image to be assessed.
In an embodiment of the present invention, described clear in order to improve the accuracy to definition of text images to be assessed assessment
Clear degree assessment models can comprehensively consider a variety of sharpness information indexs in training, such as consider that point sharpness information index, high-order are empty
At least one of domain evaluation index, spectrum information evaluation index and statistical distribution evaluation index.
Step 16, the intelligibility evaluation result of the text image to be assessed is exported.
By above scheme it is found that after extracting the prospect of text image to be assessed, side is carried out to the prospect of image to be assessed
Edge processing and contours extract, the periphery that the prospect of text image to be assessed is hereafter positioned using pixel segmentation, filtering are taken turns
Wide position.The circumference position that the prospect of text image to be assessed is positioned by the way of pixel segmentation, filtering, can be with
The accuracy of the accuracy of foreground extraction and the locations of contours of prospect is improved, so as to improve follow-up text image definition
The accuracy of assessment.
In an embodiment of the present invention, it in order to improve the accuracy to definition of text images to be assessed assessment, can adopt
With the intelligibility evaluation model as described in obtaining under type training.Referring to Fig. 2, a kind of clarity is commented in the embodiment of the present invention provided
Estimate the training flow chart of model, the training process of the intelligibility evaluation model may comprise steps of.
Step 21, sample text image is obtained.
In specific implementation, text image to be assessed can be invoice image.
Step 22, image procossing is carried out to the sample text image, extracts the prospect of the sample text image.
In specific implementation, the quantity of the sample text image can be multiple.In principle, used sample text
The number of image is bigger, and the precision of intelligibility evaluation model that training obtains is also higher, but the number of sample text image compared with
When more, arithmetic speed can be relatively slow, therefore, in practical applications, can determine according to actual needs used by sample text
The number of image.
A certain proportion of sample text image is chosen from sample text image carries out clarity mark.For example, can adopt
Manually identification method is labeled clarity, according to the clarity situation of every sample text image labeled as clear or
It is fuzzy.By another part in sample text image, image procossing is carried out, separates prospect and background, and extracts the sample text
The prospect of this image.
Step 23, marginalisation processing and contours extract are carried out to the prospect of the sample text image.
Step 24, pixel is carried out to the prospect of the sample text image after marginalisation processing and contours extract
Segmentation, filtering, position the circumference position of the prospect of the sample text image.
Step 25, according to the circumference position, affine transformation is carried out to the sample text image, after obtaining reformation
Sample text image.
Step 26, the sharpness information index of the sample text image is obtained.
For the corresponding pixel value of each Color Channel of the pixel in the sample text image after reforming, described in acquisition
The sharpness information index of sample text image.The sharpness information index of the sample text image comprises at least one of the following: point
Sharpness information index, high-order airspace evaluation index, spectrum information evaluation index and statistical distribution evaluation index.
Step 27, logistic regression training is carried out to the sharpness information index of obtained each pixel respectively, obtains training
Corresponding training parameter when obtained minimum value obtains the intelligibility evaluation model.
In specific implementation, logistic regression training is carried out respectively to the sharpness information index of obtained each pixel.Example
Such as, sharpness information index includes: a sharpness information index, high-order airspace evaluation index, spectrum information evaluation index and statistical
Cloth evaluation index.A sharpness information index, high-order airspace evaluation index, spectrum information evaluation index and statistical distribution are commented respectively
Valence index carries out logistic regression training.Corresponding training parameter when by objective function minimum value, as intelligibility evaluation model
Parameter, to obtain the intelligibility evaluation model.Furthermore, it is possible to using the sample text image pair for having carried out clarity mark
The precision of the intelligibility evaluation model is verified.
The present invention is better understood and realized for the ease of those skilled in the art, the present invention provided below with reference to Fig. 3
The flow chart of another definition of text images appraisal procedure is illustrated in embodiment.
Step 31, text image to be assessed is pre-processed, being equalized image.
In specific implementation, after getting text image to be assessed, the text image to be assessed can be carried out
Pretreatment.Specifically, carrying out histogram equalization processing to text image to be assessed.By the three primary colors of text image to be assessed
Channel becomes the channel HSV, and wherein the channel H refers to Color Channel, and channel S refers to that saturation degree channel, the channel V refer to luminance channel.Using public affairs
Formula (1), takes the channel V to be handled.
fu,v=F (fu,v)*(fmax-fmin)+fmin； (1)
Wherein, F (i) is the original probability distribution function of pixel (u, v) in text image to be assessed；fmaxIt is to be assessed
The maximum value in the channel V of pixel in text image；fminFor the minimum value in the channel V of pixel in text image to be assessed；
fu,vFor the probability-distribution function in the new channel V.In specific implementation, it is converted after the new channel V being merged with the channel H and channel S
Return primary display channels, the image for being equalized.
Step 32, feature extraction.
In specific implementation, the image of equalization is changed into the channel LAB from three primary colors.To the image after conversion, own
Pixel point value regards one group of five dimensional vector set as, includes three-dimensional channel and two-dimensional position information in five dimensional vector set.Five
Two unsupervised taxonomic clusterings are carried out in dimensional vector set.Common unsupervised clustering has k-means cluster, Gaussian Mixture
Cluster, laminar cluster, DBSCAN cluster based on density etc..In an embodiment of the present invention, it is imitated using convergence rate and cluster
The preferable laminar clustering method of fruit.Based on used laminar Unsupervised clustering, using following metric form: five dimensional vectors of note
Point is x, y and corresponding component xi,yi(i=1 ..., 5), note collection are combined into A, then measure d accordingly using following formula (2) and public
Formula (3) obtains.
Wherein, μiFor weight；Min is to be minimized.
In specific implementation, due to the physical quantity of three-dimensional pixel point value information and rear two-dimensional space information preceding in feature vector
Guiding principle is inconsistent, therefore can introduce above-mentioned weight mui。μiSpecific value can be obtained by Training, can also be preparatory
Setting.In an embodiment of the present invention, μ1=μ2=μ3=0.65, μ4=μ5=0.35.It is understood that in practical application
In, μ1、μ2、μ3、μ4And μ5There may also be other values.
Step 33, the prospect of the text image to be assessed is extracted.
The metric function based on determined by formula (2) and formula (3) carries out unsupervised laminar cluster, using formula (4)
The prospect and background for isolating text image, are denoted as L1 for the label of prospect, and the label of background is denoted as L2.Based on prospect and back
The label of scape text image to be assessed carries out the highlighted operation of prospect, and background darkens operation.
Wherein, χ is indicator function.
Step 34, polygon filtering positions the circumference position of the text image prospect to be assessed.
After the prospect and background of isolating text image to be assessed, the prospect of text image to be assessed can be extracted.
Marginalisation processing is carried out using to be assessed text image of the methods of the Sobel operator to the prospect that separated and background and profile mentions
It takes, obtains the polygonal profile set being made of multiple polygons.To obtained polygonal profile set, according to adjacent edge
The within angle of apex coordinate calculating adjacent edge.The within angle being calculated is compared with the threshold value of setting, in the present invention one
In embodiment, the threshold value set is 120 °, by the inner clip angular filter less than 120 °, to filter excessively straight within angle.In reality
In the application of border, set threshold value may be other values.
Two adjacent edge l1And l2Respectively (x1, out,y1, out,xin,yin), and (x2, out,y2, out,xin,yin), wherein (xin,
yin) when within angle apex coordinate, (x1, out,y1, out), and (x2, out,y2, out) it is outer contact coordinate.
It can be using following formula (5), (6), (7), (8) to the circumference position of the text image prospect to be assessed
It is positioned.
(y2, out-y1, out)(x-xnew)+(x1, out-x2, out)(y-ynew)=0 (8)
Formula (8) is to work as l1And l2Within angle be less than threshold value when, by l1And l2Merge the side of the l of obtained new straight line
Formula.
Straight line l is replaced into the l in former polygon1And l2, and update other crosspoints.
In polygon set after filtration, all quadrangles are selected, and carry out according to the size of quadrangle
Sequence, using the maximum quadrangle of area as the circumference position of text image to be assessed.
Step 35, text image prospect affine transformation to be assessed.
According to the circumference position of identified text image to be assessed, using following formula (9) to text to be assessed
Image after image equilibration carries out affine transformation, the text image to be assessed after being reformed.
Wherein,It is the matrix coefficient of affine transformation；b1、b2It is displacement；m11、m12、m21、m22It is rotation
Turn, scaling or mistake cut the coefficient that equiaffine converts.
Step 36, it is input to clarity evaluation index model.
In specific implementation, size normalized can be carried out to text image to be assessed, so that described to be assessed
The size of text image meets preset size requirement.It, can for the corresponding pixel value of each Color Channel of pixel (u, v)
To obtain the sharpness information of image using the different method of following four.
(1) sharpness information index is putIt can be calculated using formula (10).
(2) high-order airspace evaluation index, using 8 difference indexs of Laplace operatorFormula can specifically be used
(11) it is calculated.
(3) spectrum information evaluation indexIt can be calculated using following formula (12).
Wherein, | | | | indicate that the modulus of complex number is long；M and N refers to frequency space range；Frequency values when u and v；X and y is space bit
Value is set, f is pixel value.
(4) statistical distribution evaluation indexComentropy is used in one embodiment of the invention, can use following formula
It is calculated.
Wherein, piIt is the statistical distribution estimation of each pixel point value,It is distribution corresponding to the pixel point value of point (u, v)
Information.
Using above four kinds different index clusters as four groups of characteristic index parametric familiesIt willUnbalanced input
It couples discriminator (14).
Wherein,Input feature vector for above-mentioned four kinds of evaluation index values, as Logic Regression Models；It is every a sub- Logic Regression Models to training burden model parameter.
It, will using Logic Regression ModelsIt is separately input into corresponding sub- Logic Regression Models, it is available
The model of four sub- logistic regressions, the corresponding objective function of model (15) of every sub- logistic regression.
Based on the available coupling model of aforementioned four Logic Regression Models and the corresponding objective function of coupling model
(16)。
Coupling function is trained, the corresponding objective function of intelligibility evaluation model (17) is obtained.
Wherein, yLIt is the manual tag that whether each function obscures in training set,It is to training parameter.Using mesh
Scalar functions (17) carry out logistic regression training, and when by L value minimum, corresponding parameter is as the parameter in intelligibility evaluation model
Value obtains intelligibility evaluation model.
In order to enhance the stability and accuracy rate of intelligibility evaluation model, to the parameter to be estimated in coupling modelIt introduces soft-constraint condition (18):
ωk> 0, λk∈ [0,1]； (18)
It using above-mentioned coupling function, by nonlinear coupled modes, can inhibit to identify lack confidence model, activate
Identify the strong model of confidence.In specific implementation, it can determine that the identification confidence of model is strong according to the recognition result that model exports
Weak, for example, the intelligibility evaluation result of model one of for same text image output is 0.5, the other three model is defeated
Intelligibility evaluation result out is 0.7,0.7 and 0.6, then it is assumed that the identification for the model that assessment result is 0.5 is lack confidence.It can be with
According to each model export as a result, adjust weight shared by each model, it is clear to enhance clarity identification model and text image
The stability and accuracy rate of clear degree assessment.
In specific implementation, can using carried out the sample text image of clarity mark to intelligibility evaluation model into
Performing check examines the accuracy of intelligibility evaluation model.
Step 37, intelligibility evaluation result is exported.
From the foregoing, it will be observed that the superpixel segmentation method based on Unsupervised clustering, using the polygonal wheel of the prospect of this paper image
Wide extraction positioning and filter method, avoids that conventional foreground extracting method, which requires manual intervention, text edges segmentation is crude makes
At drawbacks such as subsequent affine transformation distortion, it is quasi- that the prospect and background separation of text image to be assessed under noisy environment can be improved
True property.In addition, using a kind of a variety of methods of coupling (point acutance discriminance, high-order airspace calculus of finite differences, Fourier domain analytic approach)
Supervised image definition decision model can solve conventional method and use model generalization scarce capacity brought by fixed threshold
The problem of, and used non-linear multi-model coupled modes are based on, the differentiation sensitivity of training pattern can be improved.
To better understand and realizing that the embodiment of the present invention, the embodiment of the present invention also provide one convenient for those skilled in the art
Kind definition of text images assessment system, is illustrated below with reference to Fig. 4.
The definition of text images assessment system 40 may include foreground extraction unit 41, first processing units 42, mistake
Filter positioning unit 43, affine transformation unit 44, input unit 45 and output unit 46.
The foreground extraction unit 41, suitable for extracting the prospect of text image to be assessed；
The first processing units 42 carry out marginalisation processing and profile suitable for the prospect to the text image to be assessed
It extracts；
The filtering positioning unit 43, suitable for the text image to be assessed after marginalisation processing and contours extract
Prospect carries out pixel segmentation, filtering, positions the circumference position of the prospect of the text image to be assessed；
The affine transformation unit 44 is suitable for carrying out the text image to be assessed according to the circumference position
Affine transformation, the text image to be assessed after being reformed；
The input unit 45, suitable for the text image to be assessed after the reformation is input to preset intelligibility evaluation
Model；
The output unit 46, suitable for exporting the intelligibility evaluation result of the text image to be assessed.
In specific implementation, the first processing units 42, can be at the prospect to the text image to be assessed
Reason, obtains the edge feature figure of the text image to be assessed, carries out to the edge feature figure of the text image to be assessed more
Side shape contours extract, obtains the polygon set being made of multiple polygons.
In specific implementation, the filtering positioning unit 43, may be adapted to the profile for each polygon, according to adjacent two
The apex coordinate on side calculates the within angle on adjacent both sides；When the within angle on adjacent both sides is less than preset threshold, filter in described
Angle, and iterate to calculate and obtain new side to replace the adjacent edge of the within angle；Institute is selected from filtered polygon set
Some quadrangles, using the corresponding profile of the maximum quadrangle of area as the circumference of the prospect of the text image to be assessed
Position.
In specific implementation, the definition of text images assessment system 40 can also include: the second processing unit (in figure
It is not shown) and feature extraction unit (not shown), in which: described the second processing unit may be adapted to mention in the prospect
Before taking unit to extract the prospect of the text image to be assessed, the text image to be assessed is pre-processed, is obtained
Weighing apparatusization image；The feature extraction unit may be adapted to the equalization image changing into the channel LAB from primary display channels；Base
Equalization image after changing into the channel LAB, extracts the feature of each pixel, in which: the feature of each pixel includes
Following at least one: the feature vector set and the corresponding measurement of each feature vector of the feature vector composition of each pixel.
In specific implementation, described the second processing unit may be adapted to carry out histogram to the text image to be assessed
Equalization processing；The primary display channels of the text image to be assessed are transformed into the channel HSV；The channel V is taken to carry out brightness processed,
Obtain the new channel V；The new channel V is merged with the channel H and channel S, primary display channels is gone back to, obtains the equalization
Image.
In specific implementation, the foreground extraction unit 41 may be adapted to carry out based on described eigenvector set without prison
Two taxonomic clusterings superintended and directed isolate the prospect and background of the text image to be assessed, and extract the text image to be assessed
Prospect.
In specific implementation, the foreground extraction unit 41 can be further adapted for isolating the text image to be assessed
Prospect and background after, execute and background highlighted to the prospect of the text image to be assessed and darken operation.
In specific implementation, the definition of text images assessment system 40 can also include: model training unit 47, fit
The intelligibility evaluation model is obtained in training in the following way: obtaining sample text image；To the sample text image
Image procossing is carried out, the prospect of the sample text image is extracted；The prospect of the sample text image is carried out at marginalisation
Reason and contours extract；Pixel point is carried out to the prospect of the sample text image after marginalisation processing and contours extract
It cuts, filter, position the circumference position of the prospect of the sample text image；According to the circumference position, to described
Sample text image carries out affine transformation, the sample text image after being reformed；For in the sample text image after reforming
Pixel the corresponding pixel value of each Color Channel, the sample text image is obtained using following at least one mode
Sharpness information index: point sharpness information index, high-order airspace evaluation index, spectrum information evaluation index and statistical distribution evaluation refer to
Mark；Logistic regression training is carried out to the sharpness information index of obtained each pixel respectively, obtains the minimum value that training obtains
When corresponding training parameter, obtain the intelligibility evaluation model.
In specific implementation, the working principle and workflow of the definition of text images assessment system, can refer to
The description in any definition of text images appraisal procedure provided in the above embodiment of the present invention, is not repeated herein.
The embodiment of the present invention also provides a kind of definition of text images assessment system, including memory and processor, described
The computer instruction that can be run on the processor is stored on memory, when the processor runs the computer instruction
The step of executing any of the above-described kind of definition of text images appraisal procedure provided in an embodiment of the present invention.
The embodiment of the present invention also provides a kind of text computer readable storage medium storing program for executing, and computer readable storage medium is non-easy
The property lost storage medium or non-transitory storage media, are stored thereon with computer instruction, and the computer instruction executes sheet when running
The step of any of the above-described kind of definition of text images appraisal procedure that inventive embodiments provide.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can
It is completed with indicating relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium may include: ROM, RAM, disk or CD etc..
Although present disclosure is as above, present invention is not limited to this.Anyone skilled in the art are not departing from this
It in the spirit and scope of invention, can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
Subject to the range of restriction.
Claims (18)
1. a kind of definition of text images appraisal procedure characterized by comprising
Extract the prospect of text image to be assessed；
Marginalisation processing and contours extract are carried out to the prospect of the text image to be assessed；
Pixel segmentation, filtering are carried out to the prospect of the text image to be assessed after marginalisation processing and contours extract, it is fixed
The circumference position of the prospect of the position text image to be assessed；
According to the circumference position, affine transformation is carried out to the text image to be assessed, it is to be assessed after being reformed
Text image；
Text image to be assessed after the reformation is input to preset intelligibility evaluation model；
Export the intelligibility evaluation result of the text image to be assessed.
2. definition of text images appraisal procedure according to claim 1, which is characterized in that the text image to be assessed
Prospect carry out marginalisation processing and contours extract, comprising:
The prospect of the text image to be assessed is handled, the edge feature figure of the text image to be assessed is obtained；
Polygonal profile extraction is carried out to the edge feature figure of the text image to be assessed, obtains being made of multiple polygons
Polygon set.
3. definition of text images appraisal procedure according to claim 2, which is characterized in that described to passed through marginalisation
The prospect of text image to be assessed after processing and contours extract carries out pixel segmentation, filtering, positions the text to be assessed
The circumference position of the prospect of image, comprising:
For the profile of each polygon, the within angle on adjacent both sides is calculated according to the apex coordinate on adjacent both sides；
When the within angle on adjacent both sides is less than preset threshold, the within angle is filtered, and iterate to calculate and obtain new side to replace
The adjacent edge of the within angle；
All quadrangles are selected from filtered polygon set, using the corresponding profile of the maximum quadrangle of area as institute
State the circumference position of the prospect of text image to be assessed.
4. definition of text images appraisal procedure according to any one of claims 1 to 3, which is characterized in that extracting institute
Before the prospect for stating text image to be assessed, further includes:
The text image to be assessed is pre-processed, being equalized image；
The equalization image is changed into the channel LAB from primary display channels；
Based on the equalization image changed into after the channel LAB, the feature of each pixel is extracted, in which: the spy of each pixel
Sign comprises at least one of the following: the feature vector set and the corresponding degree of each feature vector of the feature vector composition of each pixel
Amount.
5. definition of text images appraisal procedure according to claim 4, which is characterized in that described to the text to be assessed
This image is pre-processed, being equalized image, comprising:
Histogram equalization processing is carried out to the text image to be assessed；
The primary display channels of the text image to be assessed are transformed into the channel HSV；
It takes the channel V to carry out brightness processed, obtains the new channel V；
The new channel V is merged with the channel H and channel S, primary display channels is gone back to, obtains the image of the equalization.
6. definition of text images appraisal procedure according to claim 4, which is characterized in that described to extract text to be assessed
The prospect of image, comprising:
Based on described eigenvector set, the prospect that two unsupervised taxonomic clusterings isolate the text image to be assessed is carried out
And background, and extract the prospect of the text image to be assessed.
7. definition of text images appraisal procedure according to claim 6, which is characterized in that described to be assessed isolating
After the prospect and background of text image, further includes:
It executes and background highlighted to the prospect of the text image to be assessed and darkens operation.
8. definition of text images appraisal procedure according to claim 1, which is characterized in that trained in the following way
To the intelligibility evaluation model:
Obtain sample text image；
Image procossing is carried out to the sample text image, extracts the prospect of the sample text image；
Marginalisation processing and contours extract are carried out to the prospect of the sample text image；
Pixel segmentation, filtering are carried out to the prospect of the sample text image after marginalisation processing and contours extract,
Position the circumference position of the prospect of the sample text image；
According to the circumference position, affine transformation is carried out to the sample text image, the sample text after being reformed
Image；
For the corresponding pixel value of each Color Channel of the pixel in the sample text image after reforming, the sample is obtained
The sharpness information index of text image, in which: the sharpness information index of the sample text image comprises at least one of the following: point
Sharpness information index, high-order airspace evaluation index, spectrum information evaluation index and statistical distribution evaluation index；
Logistic regression training is carried out to the sharpness information index of obtained each pixel respectively, obtains the minimum value that training obtains
When corresponding training parameter, obtain the intelligibility evaluation model.
9. a kind of definition of text images assessment system characterized by comprising
Foreground extraction unit, suitable for extracting the prospect of text image to be assessed；
First processing units carry out marginalisation processing and contours extract suitable for the prospect to the text image to be assessed；
Positioning unit is filtered, carries out picture suitable for the prospect to the text image to be assessed after marginalisation processing and contours extract
Vegetarian refreshments segmentation, filtering, position the circumference position of the prospect of the text image to be assessed；
Affine transformation unit, is suitable for according to the circumference position, carries out affine transformation to the text image to be assessed, obtains
Text image to be assessed after to reformation；
Input unit, suitable for the text image to be assessed after the reformation is input to preset intelligibility evaluation model；
Output unit, suitable for exporting the intelligibility evaluation result of the text image to be assessed.
10. definition of text images assessment system according to claim 9, which is characterized in that the first processing units,
It is handled suitable for the prospect to the text image to be assessed, obtains the edge feature figure of the text image to be assessed, it is right
The edge feature figure of the text image to be assessed carries out polygonal profile extraction, obtains the polygon being made of multiple polygons
Set.
11. definition of text images assessment system according to claim 10, which is characterized in that the filtering positioning is single
Member calculates the within angle on adjacent both sides according to the apex coordinate on adjacent both sides suitable for the profile for each polygon；When adjacent two
When the within angle on side is less than preset threshold, the within angle is filtered, and iterate to calculate and obtain new side to replace the within angle
Adjacent edge；All quadrangles are selected from filtered polygon set, and the corresponding profile of the maximum quadrangle of area is made
For the circumference position of the prospect of the text image to be assessed.
12. according to the described in any item definition of text images assessment systems of claim 9 to 11, which is characterized in that further include:
The second processing unit and feature extraction unit, in which:
Described the second processing unit, suitable for before the prospect that the foreground extraction unit extracts the text image to be assessed,
The text image to be assessed is pre-processed, being equalized image；
The feature extraction unit, suitable for the equalization image is changed into the channel LAB from primary display channels；
Based on the equalization image changed into after the channel LAB, the feature of each pixel is extracted, in which: the spy of each pixel
Sign comprises at least one of the following: the feature vector set and the corresponding degree of each feature vector of the feature vector composition of each pixel
Amount.
13. 2 described in any item definition of text images assessment systems according to claim 1, which is characterized in that at described second
Unit is managed, is suitable for carrying out histogram equalization processing to the text image to be assessed；By the three of the text image to be assessed
Primary channel is transformed into the channel HSV；It takes the channel V to carry out brightness processed, obtains the new channel V；The new channel V is led to H
Road and channel S merge, and go back to primary display channels, obtain the image of the equalization.
14. definition of text images assessment system according to claim 13, which is characterized in that the foreground extraction list
Member is suitable for being based on described eigenvector set, carries out two unsupervised taxonomic clusterings and isolate the text image to be assessed
Prospect and background, and extract the prospect of the text image to be assessed.
15. definition of text images assessment system according to claim 14, which is characterized in that the foreground extraction list
Member is further adapted for after the prospect and background for isolating the text image to be assessed, executes to the text image to be assessed
Prospect it is highlighted and background darkens operation.
16. definition of text images assessment system according to claim 9, which is characterized in that further include: model training list
Member obtains the intelligibility evaluation model suitable for training in the following way: obtaining sample text image；To the sample text
Image carries out image procossing, extracts the prospect of the sample text image；Edge is carried out to the prospect of the sample text image
Change processing and contours extract；Pixel is carried out to the prospect of the sample text image after marginalisation processing and contours extract
Point segmentation, filtering, position the circumference position of the prospect of the sample text image；It is right according to the circumference position
The sample text image carries out affine transformation, the sample text image after being reformed；For the sample text figure after reforming
The corresponding pixel value of each Color Channel of pixel as in obtains the sample text figure using following at least one mode
The sharpness information index of picture: point sharpness information index, high-order airspace evaluation index, spectrum information evaluation index and statistical distribution are commented
Valence index；Logistic regression training is carried out to the sharpness information index of obtained each pixel respectively, training is obtained and obtains most
Corresponding training parameter when small value obtains the intelligibility evaluation model.
17. a kind of definition of text images assessment system, including memory and processor, being stored on the memory can be in institute
State the computer instruction run on processor, which is characterized in that the perform claim when processor runs the computer instruction
It is required that the step of 1 to 8 described in any item definition of text images appraisal procedures.
18. a kind of computer readable storage medium, computer readable storage medium is non-volatile memory medium or non-transient deposits
Storage media is stored thereon with computer instruction, which is characterized in that perform claim requires 1 to 8 when the computer instruction is run
The step of definition of text images appraisal procedure described in one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910172031.8A CN109948625A (en) | 2019-03-07 | 2019-03-07 | Definition of text images appraisal procedure and system, computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910172031.8A CN109948625A (en) | 2019-03-07 | 2019-03-07 | Definition of text images appraisal procedure and system, computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109948625A true CN109948625A (en) | 2019-06-28 |
Family
ID=67008513
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910172031.8A Pending CN109948625A (en) | 2019-03-07 | 2019-03-07 | Definition of text images appraisal procedure and system, computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948625A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111461070A (en) * | 2020-04-29 | 2020-07-28 | Oppo广东移动通信有限公司 | Text recognition method and device, electronic equipment and storage medium |
| CN111553431A (en) * | 2020-04-30 | 2020-08-18 | 上海眼控科技股份有限公司 | Picture definition detection method and device, computer equipment and storage medium |
| CN112347899A (en) * | 2020-11-03 | 2021-02-09 | 广州杰赛科技股份有限公司 | Moving target image extraction method, device, equipment and storage medium |
| CN112396050A (en) * | 2020-12-02 | 2021-02-23 | 上海优扬新媒信息技术有限公司 | Image processing method, device and storage medium |
| CN112668640A (en) * | 2020-12-28 | 2021-04-16 | 泰康保险集团股份有限公司 | Text image quality evaluation method, device, equipment and medium |
| CN112767392A (en) * | 2021-03-02 | 2021-05-07 | 百果园技术(新加坡)有限公司 | Image definition determining method, device, equipment and storage medium |
| CN112967191A (en) * | 2021-02-19 | 2021-06-15 | 泰康保险集团股份有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| WO2021143233A1 (en) * | 2020-01-17 | 2021-07-22 | 苏州科达科技股份有限公司 | Image definition detection method, system, and device, and storage medium |
| CN113315995A (en) * | 2021-05-06 | 2021-08-27 | 北京智通东方软件科技有限公司 | Method and device for improving video quality, readable storage medium and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107481238A (en) * | 2017-09-20 | 2017-12-15 | 众安信息技术服务有限公司 | Image quality measure method and device |
| CN107679497A (en) * | 2017-10-11 | 2018-02-09 | 齐鲁工业大学 | Video face textures effect processing method and generation system |
| WO2018036276A1 (en) * | 2016-08-22 | 2018-03-01 | 平安科技（深圳）有限公司 | Image quality detection method, device, server and storage medium |
| CN107958455A (en) * | 2017-12-06 | 2018-04-24 | 百度在线网络技术（北京）有限公司 | Image definition appraisal procedure, device, computer equipment and storage medium |
-
2019
- 2019-03-07 CN CN201910172031.8A patent/CN109948625A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018036276A1 (en) * | 2016-08-22 | 2018-03-01 | 平安科技（深圳）有限公司 | Image quality detection method, device, server and storage medium |
| CN107481238A (en) * | 2017-09-20 | 2017-12-15 | 众安信息技术服务有限公司 | Image quality measure method and device |
| CN107679497A (en) * | 2017-10-11 | 2018-02-09 | 齐鲁工业大学 | Video face textures effect processing method and generation system |
| CN107958455A (en) * | 2017-12-06 | 2018-04-24 | 百度在线网络技术（北京）有限公司 | Image definition appraisal procedure, device, computer equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 薛萍: "基于超像素特征表示的图像前景背景分割算法" * |
| 薛萍: "基于超像素特征表示的图像前景背景分割算法", 西安科技大学学报, vol. 37, no. 5, pages 731 - 735 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021143233A1 (en) * | 2020-01-17 | 2021-07-22 | 苏州科达科技股份有限公司 | Image definition detection method, system, and device, and storage medium |
| CN111461070B (en) * | 2020-04-29 | 2023-12-08 | Oppo广东移动通信有限公司 | Text recognition method, device, electronic equipment and storage medium |
| CN111461070A (en) * | 2020-04-29 | 2020-07-28 | Oppo广东移动通信有限公司 | Text recognition method and device, electronic equipment and storage medium |
| CN111553431A (en) * | 2020-04-30 | 2020-08-18 | 上海眼控科技股份有限公司 | Picture definition detection method and device, computer equipment and storage medium |
| CN112347899A (en) * | 2020-11-03 | 2021-02-09 | 广州杰赛科技股份有限公司 | Moving target image extraction method, device, equipment and storage medium |
| CN112347899B (en) * | 2020-11-03 | 2023-09-19 | 广州杰赛科技股份有限公司 | Moving object image extraction method, device, equipment and storage medium |
| CN112396050B (en) * | 2020-12-02 | 2023-09-15 | 度小满科技(北京)有限公司 | Image processing method, device and storage medium |
| CN112396050A (en) * | 2020-12-02 | 2021-02-23 | 上海优扬新媒信息技术有限公司 | Image processing method, device and storage medium |
| CN112668640A (en) * | 2020-12-28 | 2021-04-16 | 泰康保险集团股份有限公司 | Text image quality evaluation method, device, equipment and medium |
| CN112668640B (en) * | 2020-12-28 | 2023-10-17 | 泰康保险集团股份有限公司 | Text image quality evaluation method, device, equipment and medium |
| CN112967191A (en) * | 2021-02-19 | 2021-06-15 | 泰康保险集团股份有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN112967191B (en) * | 2021-02-19 | 2023-12-05 | 泰康保险集团股份有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN112767392A (en) * | 2021-03-02 | 2021-05-07 | 百果园技术(新加坡)有限公司 | Image definition determining method, device, equipment and storage medium |
| CN112767392B (en) * | 2021-03-02 | 2024-04-09 | 百果园技术(新加坡)有限公司 | Image definition determining method, device, equipment and storage medium |
| CN113315995A (en) * | 2021-05-06 | 2021-08-27 | 北京智通东方软件科技有限公司 | Method and device for improving video quality, readable storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948625A (en) | Definition of text images appraisal procedure and system, computer readable storage medium | |
| CN109154978B (en) | System and method for detecting plant diseases | |
| Barron | Convolutional color constancy | |
| CN107833220B (en) | Fabric defect detection method based on deep convolutional neural network and visual saliency | |
| Kong et al. | Partitioning histopathological images: an integrated framework for supervised color-texture segmentation and cell splitting | |
| Paris et al. | A topological approach to hierarchical segmentation using mean shift | |
| CN108765465B (en) | Unsupervised SAR image change detection method | |
| CN105096259B (en) | The depth value restoration methods and system of depth image | |
| Tosta et al. | Segmentation methods of H&E-stained histological images of lymphoma: A review | |
| CN108537239B (en) | Method for detecting image saliency target | |
| CN107016664B (en) | A kind of bad needle flaw detection method of large circle machine | |
| CN107133651A (en) | The functional magnetic resonance imaging data classification method of subgraph is differentiated based on super-network | |
| CN104217221A (en) | Method for detecting calligraphy and paintings based on textural features | |
| CN110728302A (en) | Method for identifying color textile fabric tissue based on HSV (hue, saturation, value) and Lab (Lab) color spaces | |
| CN108710916A (en) | The method and device of picture classification | |
| CN110298817A (en) | Object statistical method, device, equipment and storage medium based on image procossing | |
| Kong et al. | Towards building computerized image analysis framework for nucleus discrimination in microscopy images of diffuse glioma | |
| CN112102928B (en) | Pathological image dyeing style normalization method and device | |
| Jonnalagedda et al. | [regular paper] mvpnets: Multi-viewing path deep learning neural networks for magnification invariant diagnosis in breast cancer | |
| CN106462773B (en) | Use the pattern recognition system and method for GABOR function | |
| CN113850792A (en) | Cell classification counting method and system based on computer vision | |
| CN108805838A (en) | A kind of image processing method, mobile terminal and computer readable storage medium | |
| CN113379785B (en) | Saliency target detection method integrating boundary priori and frequency domain information | |
| Zhang et al. | An empirical study on the artificial intelligence-aided quantitative design of art images | |
| CN114973131A (en) | Full-automatic fisheye opening and closing indicator state identification method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190628 |