CN108345881A - A kind of document quality detection method based on computer vision - Google Patents

A kind of document quality detection method based on computer vision Download PDF

Info

Publication number
CN108345881A
CN108345881A CN201810101325.7A CN201810101325A CN108345881A CN 108345881 A CN108345881 A CN 108345881A CN 201810101325 A CN201810101325 A CN 201810101325A CN 108345881 A CN108345881 A CN 108345881A
Authority
CN
China
Prior art keywords
image
point
picture
file
document
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
Application number
CN201810101325.7A
Other languages
Chinese (zh)
Other versions
CN108345881B (en
Inventor
郭文忠
张融
柯逍
陈羽中
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fuzhou University
Original Assignee
Fuzhou University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Fuzhou University filed Critical Fuzhou University
Priority to CN201810101325.7A priority Critical patent/CN108345881B/en
Publication of CN108345881A publication Critical patent/CN108345881A/en
Application granted granted Critical
Publication of CN108345881B publication Critical patent/CN108345881B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/40—Document-oriented image-based pattern recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/24—Aligning, centring, orientation detection or correction of the image
    • G06V10/243—Aligning, centring, orientation detection or correction of the image by compensating for image skew or non-uniform image deformations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/34—Smoothing or thinning of the pattern; Morphological operations; Skeletonisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/48—Extraction of image or video features by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/98—Detection or correction of errors, e.g. by rescanning the pattern or by human intervention; Evaluation of the quality of the acquired patterns
    • G06V10/993—Evaluation of the quality of the acquired pattern

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Artificial Intelligence (AREA)
  • Image Analysis (AREA)

Abstract

The present invention relates to a kind of document quality detection methods based on computer vision.Difficult for traditional visual inspection, efficiency is low, poor reliability, and the problem of being affected by subjectivity proposes document quality detection method based on computer vision.Accurately to detect document quality, this method obtains first prints video progress frozen frozen mass extraction by rational method to High Speed Document；Secondly, preprocess method based on computer vision carries out suitable image preprocessing to file and picture；Again, good that file and picture carries out accurate contour detecting and extraction to pre-processing；And then, Slant Rectify is carried out to the document profile image extracted so that lopsided image becomes the document image to be detected normally to tile；Finally, PSNR and MSE quality evaluations are carried out to image to be detected, is compared with template, obtains the testing result of document quality.This method possesses the features such as high efficiency, reliability, continuity, flexibility etc., has stronger practical application.

Description

A kind of document quality detection method based on computer vision
Technical field
The present invention relates to computer visions and digital image arts, and are applied in document quality detection, especially A kind of document quality detection method based on computer vision.
Background technology
Fast-developing with the society of 21st century, the fast development and progress of industry and service trade are printed in modernization Brush has industrially carried out supermatic production, since the competition between each enterprise is more and more fierce in service trade, So the requirement for the appearance printing quality of document is higher and higher, and it is more and more for the demand of document, it needs to huge The document of quantity carry out taxonomic revision also at headache the problem of.In Modern Press enterprise by the pressure shadow of time and cost Under sound, traditional visually observing can only qualitatively describe print quality, and cannot it is a large amount of, rapidly print quality is quantified Description can not accurately classify quickly and according to certain tagsorts document.Promote people in these cases Urgently seek the document detection of efficiently and accurately and the intelligence system of classification.
And in recent years, computer vision technique is fast-developing, and various detection techniques based on computer vision are in many realities It has been widely used and has developed in the application system on border.Computer vision is to replace the mankind with industrial camera and computer, In the case of no human intervention by computer computer come target is identified, tracing detection.Generally speaking, it exactly uses Machine replaces human eye to handle and analyze image.The features such as possessing high efficiency, reliability, continuity, flexibility etc..In modern chemical industry In Masses of Document printing in industry and service trade, probably due to not reach requirement, printing material inadequate for printing press precision, with And the complicated technology in printing technology also has environment etc. factor.The document printed may have various quality and lack Fall into, for example insurance policy exists as an example, printing defects have that ink stain, pin hole, corrugation, word be fuzzy, skip printing of characters, the flaw Defect, scratch, ghost image, cut, dislocation, register trouble, chromatography error etc..These mass defects can become the development bottle of industrial enterprise Neck, for service trade, the case where causing to be less competitive than others.Meanwhile because manual quality's detection efficiency well below machine The efficiency of computer, these situations force document quickly to detect with classification as inevitable development trend.
Invention content
The purpose of the present invention is to provide a kind of document quality detection methods based on computer vision, to overcome existing people The existing defect of work detection, and solve to be directed to document quality detection method based on computer vision.
To achieve the above object, the technical scheme is that：A kind of document quality detection side based on computer vision Method is realized in accordance with the following steps：
Step S1：Frozen frozen mass is extracted to the high-speed image video sequence of input and is detected processing to get to needs File and picture；
Step S2：The frozen frozen mass extracted is pre-processed, noise reduction and Edge Enhancement are carried out, excludes environmental disturbances item；
Step S3：Image after being pre-processed to step S2 carries out document profile detection, then presses contours extract file and picture Part；
Step S4：Slant Rectify is carried out to file and picture on spatial alternation, makes the vertical vertical view document of tiling Image；
Step S5：The file and picture that step S4 identifications navigate to is compared by quality evaluation algorithm with template, is examined Measure the quality level of document.
In an embodiment of the present invention, in the step S1, file and picture to be detected is extracted as follows：
Step S11：Frame number I is taken to the high-speed video under printer of input, image is extracted every I frame numbers, chooses The video streaming of input is image stream by predetermined interval frame number；
Step S12：Frozen frozen mass processing is carried out to image stream, two images before and after in image stream are subjected to image algorithm phases Subtract, it is pixel 255 to make same section, and it is 0 to differ part；
Step S13：The image after phase reducing, 0 pixel number N are counted, if N is less than the threshold value Y preset, when Previous frame is frozen frozen mass, otherwise, abandons this frame, continues step S12.
In an embodiment of the present invention, in the step S2, frozen frozen mass is pre-processed as follows：
Step S21：Image to be detected is subjected to picture smooth treatment, Mean by Mean Shift mean shift algorithms Shift algorithm steps are：
S211：First a pixel centered on point is randomly selected in picture；
S212：The offset mean value M (x) of central point is calculated,Wherein, x indicates that central point is horizontal Coordinate y indicates the ordinate of central point, ShIt is the higher-dimension ball region that a radius is h：K expressions have k in this n sample point Fall into ball ShIn；
S213：The mobile point deviates mean value, X to itt+1=Mt+xt, MtFor the offset mean value acquired under t states；xtFor t shapes Center under state, Xt+1For xtCenter under next state；Then it as new starting point, continues to move to, until meeting one Fixed condition terminates；
Step S22：Image to be detected is subjected to morphological operation opening operation, i.e., first corrodes the process expanded afterwards to image, Its mathematic(al) representation is as follows：
Dst=open (src, element)=dilate (erode (src, element))
Wherein, src is input picture, and element is the core of definition, and erode is etching operation, and dilate is expansion behaviour Make；
Step S23：Image to be detected is subjected to binary conversion treatment, a threshold value is chosen and image is switched into gray level image.
In an embodiment of the present invention, in the step S3, the image after being pre-processed to step S2 carries out document wheel Steps are as follows for the specific implementation of exterior feature detection：
Step S31：Contour detecting is carried out to image to be detected using Canny operators, the convolution operator of Canny isX and y is reference axis X and the variable of Y-axis, SxIt is that its x Directional partial derivative calculates masterplate, SyIts y Directional partial derivative calculates masterplate, x-axis direction, y-axis direction first-order partial derivative matrix, gradient magnitude and gradient side To mathematic(al) representation be：
P [i, j]=(f [i, j+1]-f [i, j]+f [i+1, j+1]-f [i+1, j])/2
Q [i, j]=(f [i, j]-f [i+1, j]+f [i, j+1]-f [i+1, j+1])/2
θ[i：J]=arCtan (Q [i：j]P[i：j])
Wherein, i and j is array index, and wherein f is gray value of image, and P represents X-direction gradient magnitude, and Q represents Y-axis side To gradient magnitude, M is the amplitude, and θ is gradient direction, that is, angle；
Non-maxima suppression is carried out further according to calculated gradient magnitude, dual threashold value-based algorithm is and then used to detect and connect side Edge；
Step S32：Function is searched by profile, and all contour edges are extracted from binary map again by writing a deletion Algorithm：The area of all connected domains is calculated, the connected domain that those areas are less than predetermined threshold value is then deleted, leaves behind the wheel of document Exterior feature figure；
Step S33：The straightway in profile diagram is detected, using accumulated probability Hough transformation, steps are as follows：
S331：Image coordinate is mapped on image polar coordinates, straight line expression formula is
Wherein, r is distance of the origin to straight line,It is X axis coordinate, β is Y axis coordinate, and λ is the inclination angle of straight line and X-axis；
S332：A characteristic point in image, i.e. marginal point are randomly selected, if the point has been demarcated as being a certain item Point on straight line then continues to randomly select a marginal point in remaining marginal point, be over until all marginal points all extract Until；
S333：Hough transformation is carried out to the point, and is added up and is calculated；
S334：It is chosen at the maximum point of value in hough space and carries out step S335 if the point is more than threshold value, it is no Then return to step S332；
S335：The maximum value obtained according to Hough transformation, from this point, along the direction displacement of straight line, to find Two endpoints of straight line；
S336：The length of straight line is calculated, if it is greater than some threshold value, then the edge line segment record for being considered as gets off to put In Lines arrays, step S332 is returned to.
In an embodiment of the present invention, in the step S4, Slant Rectify is carried out to file and picture on spatial alternation Specific implementation steps are as follows：
Step S51：File and picture edge is a quadrangle, is divided into four sides, respectively A, B, C, D up and down；One Vertical element is determined by two endpoints；All straightways in Lines arrays are substantially pressed into A, B, C, D by the two point coordinates Four side classification, file and picture upper left corner a points take the point near the upper left corner in the left end point of A and the upper extreme point of C；Document map As upper right corner b points, the point near the upper right corner in the right endpoint of A and the upper extreme point of D is taken；File and picture lower left corner c points, take C's Near the point in the lower left corner in the left end point of lower extreme point and B；File and picture lower right corner d points, take the lower extreme point of D and the right endpoint of B In near the lower right corner point, take out a, b, c, d file and picture four angle points；
Step S52：Then, it is converted using perspective space and Slant Rectify, first root is carried out to the lopsided image of file and picture Lopsided figure is obtained by perspective transformation matrix algorithmic function according to four angle points found in step S51 and changes to rectangular perspective change Matrix is changed, then converting algorithmic function by perspective space carries out spatial alternation completion Slant Rectify.
In an embodiment of the present invention, steps are as follows for the specific implementation of the step S5：
Step S61：The quality level of file and picture is weighed using PSNR Y-PSNRs；PSNR is applied into text to be detected Shelves image and template image, PSNR values are bigger between 2 images, then more similar；A threshold value ps is pre-set as measurement base It is accurate；PSNR formula are as follows：
Wherein, MAX indicates that the greatest measure of color of image, 8 sampled points are expressed as 255, therefore MAX2=255*255；
MSE indicates present image P1With reference picture P2Mean square error, MSE formula are：
Wherein, m and n indicates that the height and width of picture frame, I indicate currently in I row, and J indicates currently to arrange in J, K (I, J) Indicate present image P1I row J row pixel, L (I, J) indicates reference picture P2I row J row pixel；
Step S62：It is substituted into image to be detected array as K (I, J), template image array is substituted into as L (I, J)；Meter The PSNR value pr for obtaining file and picture to be detected are calculated, if pr is more than threshold value ps, document quality level to be detected is eligible, no Then document quality level to be detected is unqualified.
Compared to the prior art, the invention has the advantages that：Caused by the present invention has considered environmental factor Interference deletes the small connected domain of area to improve the accuracy and reliability of document positioning by preprocessing noise reduction and algorithm.Together When, function pair document profile is searched by Canny operators and profile and is detected and extracts.Furthermore pass through accumulated probability Hough Transformation carries out line segment extraction, using the algorithm of design, four angular coordinates of locating documents.Then, by perspective space Slant correction is converted, lopsided file and picture is drawn into normal file and picture.Finally, with PSNR and MSE appraisal procedures pair Document quality is detected.This patent uses and optimizes the algorithm in computer vision, and identification document can be precisely located And quality testing is carried out to document.This method is simple, realizes that flexibly practicability is stronger.
Description of the drawings
Fig. 1 is that the present invention is based on the flow charts of the document quality detection method of computer vision.
Specific implementation mode
Below in conjunction with the accompanying drawings, technical scheme of the present invention is specifically described.
As shown in Figure 1, the present invention provides a kind of document quality detection method based on computer vision.To traditional artificial Naked eyes detection difficult, efficiency is low, poor reliability, and the problem of being affected by subjectivity proposes text based on computer vision Shelves quality determining method.Accurately to detect document quality, this method obtains first regards High Speed Document printing by rational method Frequency carries out frozen frozen mass extraction；Secondly, preprocess method based on computer vision locates the suitable image of file and picture progress in advance Reason；Again, good that file and picture carries out accurate contour detecting and extraction to pre-processing；And then, the document wheel to extracting Wide image carries out Slant Rectify so that lopsided image becomes the document image to be detected normally to tile；Finally, to image to be detected PSNR and MSE quality evaluations are carried out, is compared with template, obtains the testing result of document quality.It is as follows：
Step S1：Frozen frozen mass is extracted to the high-speed image video sequence of input and is detected processing to get to needs File and picture；
Step S2：The frozen frozen mass extracted is pre-processed, noise reduction and Edge Enhancement are carried out, excludes environmental disturbances item；
Step S3：Image after being pre-processed to step S2 carries out document profile detection, then presses contours extract file and picture Part；
Step S4：Slant Rectify is carried out to file and picture on spatial alternation, makes the vertical vertical view document of tiling Image；
Step S5：The file and picture that step S4 identifications navigate to is compared by quality evaluation algorithm with template, is examined Measure the quality level of document.
In the step S1, file and picture to be detected is extracted as follows：
Step S11：Suitable frame number I is taken to the high-speed video under printer of input, extracts and schemes every I frame numbers Picture, it is image stream to choose predetermined interval frame number by the video streaming of input；
Step S12：Frozen frozen mass processing is carried out to image stream, two images before and after in image stream are subjected to image algorithm phases Subtract, it is pixel 255 (white pixel) to make same section, and it is 0 (black picture element) to differ part；
Step S13：The image after phase reducing, 0 (black) pixel number N are counted, if N is less than the threshold value Y preset, Then present frame is frozen frozen mass, otherwise, abandons this frame, continues step S12.
In the step S2, frozen frozen mass is pre-processed as follows：
Step S21：Image to be detected is subjected to picture smooth treatment, Mean by Mean Shift mean shift algorithms Shift algorithm steps are：
S211：First a pixel centered on point is randomly selected in picture；
S212：The offset mean value M (x) of central point is calculated,Wherein, x indicates that central point is horizontal Coordinate, y indicate the ordinate of central point, ShIt is the higher-dimension ball region that a radius is h：K expressions have k in this n sample point It is a to fall into ball ShIn；
S213：The mobile point deviates mean value, X to itt+1=Mt+xt, MtFor the offset mean value acquired under t states；xtFor t shapes Center under state, Xt+1For xtCenter under next state；Then it as new starting point, continues to move to, until meeting one Fixed condition terminates；
Step S22：Image to be detected is subjected to morphological operation opening operation, i.e., first corrodes the process expanded afterwards to image, Its mathematic(al) representation is as follows：
Dst=open (src, element)=dilate (erode (src, element))
Wherein, src is input picture, and element is the core of definition, and erode is etching operation, and dilate is expansion behaviour Make；
Step S23：Image to be detected is subjected to binary conversion treatment, a threshold value is chosen and image is switched into gray level image.
In the step S3, the image after being pre-processed to step S2 carries out the specific implementation step of document profile detection It is as follows：
Step S31：Contour detecting is carried out to image to be detected using Canny operators, the convolution operator of Canny isX and y is reference axis X and the variable of Y-axis, SxIt is that its x Directional partial derivative calculates masterplate, SyIts y Directional partial derivative calculates masterplate, x-axis direction, y-axis direction first-order partial derivative matrix, gradient magnitude and gradient side To mathematic(al) representation be：
P [i, j]=(f [i, j+1]-f [i, j]+f [i+1, j+1]-f [i+1, j])/2
Q [i, j]=(f [i, j]-f [i+1, j]+f [i, j+1]-f [i+1, j+1])/2
θ [i, j]=arctan (Q [i, j] P [i：j])
Wherein, i and j is array index, and wherein f is gray value of image, and P represents X-direction gradient magnitude, and Q represents Y-axis side To gradient magnitude, M is the amplitude, and θ is gradient direction, that is, angle；
Non-maxima suppression is carried out further according to calculated gradient magnitude, dual threashold value-based algorithm is and then used to detect and connect side Edge；
Step S32：Function is searched by profile, and all contour edges are extracted from binary map again by writing a deletion Algorithm：The area of all connected domains is calculated, the connected domain that those areas are less than predetermined threshold value is then deleted, leaves behind the wheel of document Exterior feature figure；
Step S33：The straightway in profile diagram is detected, using accumulated probability Hough transformation, steps are as follows：
S331：Image coordinate is mapped on image polar coordinates, straight line expression formula is
Wherein, r is distance of the origin to straight line,It is X axis coordinate, β is Y axis coordinate, and λ is the inclination angle of straight line and X-axis；
S332：A characteristic point in image, i.e. marginal point are randomly selected, if the point has been demarcated as being a certain item Point on straight line then continues to randomly select a marginal point in remaining marginal point, be over until all marginal points all extract Until；
S333：Hough transformation is carried out to the point, and is added up and is calculated；
S334：It is chosen at the maximum point of value in hough space and carries out step S335 if the point is more than threshold value, it is no Then return to step S332；
S335：The maximum value obtained according to Hough transformation, from this point, along the direction displacement of straight line, to find Two endpoints of straight line；
S336：The length of straight line is calculated, if it is greater than some threshold value, then the edge line segment record for being considered as gets off to put In Lines arrays, step S332 is returned to.
In the step S4, to the specific implementation of file and picture progress Slant Rectify, steps are as follows on spatial alternation：
Step S51：File and picture edge is a quadrangle, is divided into four sides, respectively A, B, C, D up and down；One Vertical element is determined by two endpoints；All straightways in Lines arrays are substantially pressed into A, B, C, D by the two point coordinates Four side classification, file and picture upper left corner a points take the point near the upper left corner in the left end point of A and the upper extreme point of C；Document map As upper right corner b points, the point near the upper right corner in the right endpoint of A and the upper extreme point of D is taken；File and picture lower left corner c points, take C's Near the point in the lower left corner in the left end point of lower extreme point and B；File and picture lower right corner d points, take the lower extreme point of D and the right endpoint of B In near the lower right corner point, take out a, b, c, d file and picture four angle points；
Step S52：Then, it is converted using perspective space and Slant Rectify, first root is carried out to the lopsided image of file and picture Lopsided figure is obtained by perspective transformation matrix algorithmic function according to four angle points found in step S51 and changes to rectangular perspective change Matrix is changed, then converting algorithmic function by perspective space carries out spatial alternation completion Slant Rectify.
Steps are as follows for the specific implementation of the step S5：
Step S61：The quality level of file and picture is weighed using PSNR Y-PSNRs；PSNR is applied into text to be detected Shelves image and template image, PSNR values are bigger between 2 images, then more similar；A threshold value ps is pre-set as measurement base It is accurate；PSNR formula are as follows：
Wherein, MAX indicates that the greatest measure of color of image, 8 sampled points are expressed as 255, therefore MAX2=255*255；
Wherein, m and n indicates that the height and width of picture frame, I indicate currently in I row, and J indicates currently to arrange in J, K (I, J) Indicate present image P1I row J row pixel, L (I, J) indicates reference picture P2I row J row pixel；
Step S62：It is substituted into image to be detected array as K (I, J), template image array is substituted into as L (I, J)；Meter The PSNR value pr for obtaining file and picture to be detected are calculated, if pr is more than threshold value ps, document quality level to be detected is eligible, no Then document quality level to be detected is unqualified.
The above are preferred embodiments of the present invention, all any changes made according to the technical solution of the present invention, and generated function is made When with range without departing from technical solution of the present invention, all belong to the scope of protection of the present invention.

Claims (6)

1. a kind of document quality detection method based on computer vision, which is characterized in that realize in accordance with the following steps：
Step S1：Frozen frozen mass is extracted to get to the text for needing to be detected processing to the high-speed image video sequence of input Shelves image；
Step S2：The frozen frozen mass extracted is pre-processed, noise reduction and Edge Enhancement are carried out, excludes environmental disturbances item；
Step S3：Image after being pre-processed to step S2 carries out document profile detection, then presses contours extract file and picture part；
Step S4：Slant Rectify is carried out to file and picture on spatial alternation, makes the vertical vertical view file and picture of tiling；
Step S5：The file and picture that step S4 identifications navigate to is compared by quality evaluation algorithm with template, is detected The quality level of document.
2. a kind of document quality detection method based on computer vision according to claim 1, which is characterized in that in institute It states in step S1, extracts file and picture to be detected as follows：
Step S11：Frame number I is taken to the high-speed video under printer of input, image is extracted every I frame numbers, is chosen default The video streaming of input is image stream by interval frame number；
Step S12：Frozen frozen mass processing is carried out to image stream, two images before and after in image stream, which are carried out image algorithms, to be subtracted each other, It is pixel 255 to make same section, and it is 0 to differ part；
Step S13：The image after phase reducing, 0 pixel number N are counted, if N is less than the threshold value Y preset, present frame This frame otherwise is abandoned for frozen frozen mass, continues step S12.
3. a kind of document quality detection method based on computer vision according to claim 1, which is characterized in that in institute It states in step S2, frozen frozen mass is pre-processed as follows：
Step S21：Image to be detected is subjected to picture smooth treatment, Mean Shift by Mean Shift mean shift algorithms Algorithm steps are：
S211：First a pixel centered on point is randomly selected in picture；
S212：The offset mean value M (x) of central point is calculated,Wherein, x indicates central point abscissa, Y indicates the ordinate of central point, ShIt is the higher-dimension ball region that a radius is h：K expressions have k to fall into this n sample point Ball ShIn；
S213：The mobile point deviates mean value, X to itt+1=Mt+xt, MtFor the offset mean value acquired under t states；xtFor under t states Center, Xt+1For xtCenter under next state；Then it as new starting point, continues to move to, until meeting centainly Condition terminates；
Step S22：Image to be detected is subjected to morphological operation opening operation, i.e., first corrodes the process expanded afterwards to image, is counted It is as follows to learn expression formula：
Dst=open (src, element)=dilate (erode (src, element))
Wherein, src is input picture, and element is the core of definition, and erode is etching operation, and dilate is expansive working；
Step S23：Image to be detected is subjected to binary conversion treatment, a threshold value is chosen and image is switched into gray level image.
4. a kind of document quality detection method based on computer vision according to claim 1, which is characterized in that in institute It states in step S3, steps are as follows for the specific implementation of the image progress document profile detection after being pre-processed to step S2：
Step S31：Contour detecting is carried out to image to be detected using Canny operators, the convolution operator of Canny isX and y is reference axis X and the variable of Y-axis, SxIt is that its x Directional partial derivative calculates masterplate, SyIts y Directional partial derivative calculates masterplate, x-axis direction, y-axis direction first-order partial derivative matrix, gradient magnitude and gradient side To mathematic(al) representation be：
P [i, j]=(f [i, j+1]-f [i, j]+f [i+1, j+1]-f [i+1, j])/2
Q [i, j]=(f [i, j]-f [i+1, j]+f [i, j+1]-f [i+1, j+1])/2
θ [i, j]=arctan (Q [i, j]/P [i, j])
Wherein, i and j is array index, and wherein f is gray value of image, and P represents X-direction gradient magnitude, and Q represents Y direction ladder Amplitude is spent, M is the amplitude, and θ is gradient direction, that is, angle；
Non-maxima suppression is carried out further according to calculated gradient magnitude, dual threashold value-based algorithm is and then used to detect and connect edge；
Step S32：Function is searched by profile, and all contour edges are extracted from binary map again by writing a deletion algorithm： The area of all connected domains is calculated, the connected domain that those areas are less than predetermined threshold value is then deleted, leaves behind the profile diagram of document；
Step S33：The straightway in profile diagram is detected, using accumulated probability Hough transformation, steps are as follows：
S331：Image coordinate is mapped on image polar coordinates, straight line expression formula is
Wherein, r is distance of the origin to straight line,It is X axis coordinate, β is Y axis coordinate, and λ is the inclination angle of straight line and X-axis；
S332：A characteristic point in image, i.e. marginal point are randomly selected, if the point has been demarcated as being certain straight line On point, then continue a marginal point is randomly selected in remaining marginal point, until all marginal points, which all extract, to be over；
S333：Hough transformation is carried out to the point, and is added up and is calculated；
S334：The maximum point of value in hough space is chosen to carry out step S335 if the point is more than threshold value, otherwise return To step S332；
S335：The maximum value obtained according to Hough transformation, from this point, along the direction displacement of straight line, to find straight line Two endpoints；
S336：The length of straight line is calculated, if it is greater than some threshold value, then the edge line segment record for being considered as gets off to be placed on In Lines arrays, step S332 is returned to.
5. a kind of document quality detection method based on computer vision according to claim 1, which is characterized in that in institute It states in step S4, to the specific implementation of file and picture progress Slant Rectify, steps are as follows on spatial alternation：
Step S51：File and picture edge is a quadrangle, is divided into four sides, respectively A, B, C, D up and down；One straight Lines are determined by two endpoints；All straightways in Lines arrays are substantially pressed into A, B, C, D tetra- by the two point coordinates Side is classified, and file and picture upper left corner a points take the point near the upper left corner in the left end point of A and the upper extreme point of C；File and picture is right Upper angle b points take the point near the upper right corner in the right endpoint of A and the upper extreme point of D；File and picture lower left corner c points, take the lower end of C Near the point in the lower left corner in the left end point of point and B；File and picture lower right corner d points, take in the lower extreme point of D and the right endpoint of B most Point close to the lower right corner takes out four angle points of a, b, c, d file and picture；
Step S52：Then, it is converted using perspective space and Slant Rectify is carried out to the lopsided image of file and picture, first according to step Four angle points found in rapid S51 obtain lopsided figure by perspective transformation matrix algorithmic function and change to rectangular perspective transform square Then battle array converts algorithmic function by perspective space and carries out spatial alternation completion Slant Rectify.
6. a kind of document quality detection method based on computer vision according to claim 1, which is characterized in that described Steps are as follows for the specific implementation of step S5：
Step S61：The quality level of file and picture is weighed using PSNR Y-PSNRs；PSNR is applied into document map to be detected As and template image, PSNR values are bigger between 2 images, then more similar；A threshold value ps is pre-set as benchmark； PSNR formula are as follows：
Wherein, MAX indicates that the greatest measure of color of image, 8 sampled points are expressed as 255, therefore MAX2=255*255；
MSE indicates present image P1With reference picture P2Mean square error, MSE formula are：
Wherein, m and n indicates that the height and width of picture frame, I indicate currently in I row, and J indicates currently to arrange in J, and K (I, J) is indicated Present image P1I row J row pixel, L (I, J) indicates reference picture P2I row J row pixel；
Step S62：It is substituted into image to be detected array as K (I, J), template image array is substituted into as L (I, J)；It calculates Go out the PSNR value pr of file and picture to be detected, if pr is more than threshold value ps, document quality level to be detected is eligible, otherwise waits for It is unqualified to detect document quality level.
CN201810101325.7A 2018-02-01 2018-02-01 Document quality detection method based on computer vision Active CN108345881B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810101325.7A CN108345881B (en) 2018-02-01 2018-02-01 Document quality detection method based on computer vision

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810101325.7A CN108345881B (en) 2018-02-01 2018-02-01 Document quality detection method based on computer vision

Publications (2)

Publication Number Publication Date
CN108345881A true CN108345881A (en) 2018-07-31
CN108345881B CN108345881B (en) 2021-12-21

Family

ID=62958831

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810101325.7A Active CN108345881B (en) 2018-02-01 2018-02-01 Document quality detection method based on computer vision

Country Status (1)

Country Link
CN (1) CN108345881B (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109242787A (en) * 2018-08-15 2019-01-18 南京光辉互动网络科技股份有限公司 It paints in a kind of assessment of middle and primary schools' art input method
CN109389121A (en) * 2018-10-30 2019-02-26 金现代信息产业股份有限公司 A kind of nameplate recognition methods and system based on deep learning
CN109785367A (en) * 2019-01-21 2019-05-21 视辰信息科技(上海)有限公司 Exterior point filtering method and device in threedimensional model tracking
CN109871844A (en) * 2019-01-09 2019-06-11 东南大学 A kind of correction of shooting receipt image text and extracting method
CN110033471A (en) * 2019-04-19 2019-07-19 福州大学 A kind of wire detection method based on connected domain analysis and morphological operation
CN110032989A (en) * 2019-04-23 2019-07-19 福州大学 A kind of form document image classification method based on wire feature and pixel distribution
CN111108515A (en) * 2019-12-27 2020-05-05 威创集团股份有限公司 Picture target point correcting method, device and equipment and storage medium
CN111435531A (en) * 2018-12-25 2020-07-21 Tcl集团股份有限公司 Image deflection angle detection method, intelligent device and storage medium
CN111488931A (en) * 2020-04-10 2020-08-04 腾讯科技（深圳）有限公司 Article quality evaluation method, article recommendation method and corresponding devices
CN112131841A (en) * 2020-08-27 2020-12-25 北京云动智效网络科技有限公司 Document quality evaluation method and system
CN112200789A (en) * 2020-10-16 2021-01-08 中国铁道科学研究院集团有限公司 Image identification method and device, electronic equipment and storage medium
CN112435218A (en) * 2020-11-04 2021-03-02 南京火眼锐视信息科技有限公司 Method and device for evaluating and screening deformability of document image
WO2021105902A1 (en) * 2019-11-27 2021-06-03 Cimpress Schweiz Gmbh Technologies for detecting crop marks in electronic documents

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2004109586A1 (en) * 2003-06-05 2004-12-16 Aware, Inc. Image quality control techniques
JP2011239066A (en) * 2010-05-07 2011-11-24 Sony Corp Device and method of processing image
CN102496018A (en) * 2011-12-08 2012-06-13 方正国际软件有限公司 Document skew detection method and system
CN103839058A (en) * 2012-11-21 2014-06-04 方正国际软件(北京)有限公司 Information locating method for document image based on standard template

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2004109586A1 (en) * 2003-06-05 2004-12-16 Aware, Inc. Image quality control techniques
JP2011239066A (en) * 2010-05-07 2011-11-24 Sony Corp Device and method of processing image
CN102496018A (en) * 2011-12-08 2012-06-13 方正国际软件有限公司 Document skew detection method and system
CN103839058A (en) * 2012-11-21 2014-06-04 方正国际软件(北京)有限公司 Information locating method for document image based on standard template

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
LINHUI LI ET AL.: "Vehicle detection method based on mean shift clustering", 《JOURNAL OF INTELLIGENT & FUZZY SYSTEMS》 *
ZHOU WANG ET AL.: "Mean Squared Error: Love It or Leave It?", 《IEEE SIGNAL PROCESSING MAGAZINE》 *
商萌萌: "运动物体相对位置的视觉检测算法的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
王景丹等: "基于智能检测车位信息的车辆识别算法", 《计算机科学与应用》 *
高晓波等: "基于边缘检测的OMR图像倾斜矫正", 《长春理工大学学报（自然科学版）》 *

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109242787A (en) * 2018-08-15 2019-01-18 南京光辉互动网络科技股份有限公司 It paints in a kind of assessment of middle and primary schools' art input method
CN109389121B (en) * 2018-10-30 2021-11-09 金现代信息产业股份有限公司 Nameplate identification method and system based on deep learning
CN109389121A (en) * 2018-10-30 2019-02-26 金现代信息产业股份有限公司 A kind of nameplate recognition methods and system based on deep learning
CN111435531B (en) * 2018-12-25 2023-08-08 Tcl科技集团股份有限公司 Image deflection angle detection method, intelligent device and storage medium
CN111435531A (en) * 2018-12-25 2020-07-21 Tcl集团股份有限公司 Image deflection angle detection method, intelligent device and storage medium
CN109871844A (en) * 2019-01-09 2019-06-11 东南大学 A kind of correction of shooting receipt image text and extracting method
CN109785367A (en) * 2019-01-21 2019-05-21 视辰信息科技(上海)有限公司 Exterior point filtering method and device in threedimensional model tracking
CN109785367B (en) * 2019-01-21 2019-12-20 视辰信息科技(上海)有限公司 Method and device for filtering foreign points in three-dimensional model tracking
CN110033471A (en) * 2019-04-19 2019-07-19 福州大学 A kind of wire detection method based on connected domain analysis and morphological operation
CN110033471B (en) * 2019-04-19 2022-09-13 福州大学 Frame line detection method based on connected domain analysis and morphological operation
CN110032989A (en) * 2019-04-23 2019-07-19 福州大学 A kind of form document image classification method based on wire feature and pixel distribution
CN110032989B (en) * 2019-04-23 2022-07-08 福州大学 Table document image classification method based on frame line characteristics and pixel distribution
WO2021105902A1 (en) * 2019-11-27 2021-06-03 Cimpress Schweiz Gmbh Technologies for detecting crop marks in electronic documents
US11145064B2 (en) 2019-11-27 2021-10-12 Cimpress Schweiz Gmbh Technologies for detecting crop marks in electronic documents
CN111108515B (en) * 2019-12-27 2021-11-12 威创集团股份有限公司 Picture target point correcting method, device and equipment and storage medium
CN111108515A (en) * 2019-12-27 2020-05-05 威创集团股份有限公司 Picture target point correcting method, device and equipment and storage medium
CN111488931A (en) * 2020-04-10 2020-08-04 腾讯科技（深圳）有限公司 Article quality evaluation method, article recommendation method and corresponding devices
CN111488931B (en) * 2020-04-10 2023-04-07 腾讯科技（深圳）有限公司 Article quality evaluation method, article recommendation method and corresponding devices
CN112131841A (en) * 2020-08-27 2020-12-25 北京云动智效网络科技有限公司 Document quality evaluation method and system
CN112200789A (en) * 2020-10-16 2021-01-08 中国铁道科学研究院集团有限公司 Image identification method and device, electronic equipment and storage medium
CN112200789B (en) * 2020-10-16 2023-11-21 中国铁道科学研究院集团有限公司 Image recognition method and device, electronic equipment and storage medium
CN112435218A (en) * 2020-11-04 2021-03-02 南京火眼锐视信息科技有限公司 Method and device for evaluating and screening deformability of document image

Also Published As

Publication number Publication date
CN108345881B (en) 2021-12-21

Similar Documents

Publication Publication Date Title
CN108345881A (en) A kind of document quality detection method based on computer vision
CN112508826B (en) Printed matter defect detection method
CN106683075B (en) Method for detecting bolt defects at cross arm of power transmission line tower
CN114943739B (en) Aluminum pipe quality detection method
CN108355987A (en) A kind of screen printing of battery quality determining method based on piecemeal template matches
CN110264445A (en) The screen printing of battery quality determining method of piecemeal template matching combining form processing
CN111127417B (en) Printing defect detection method based on SIFT feature matching and SSD algorithm improvement
CN112991305A (en) Visual inspection method for surface defects of paint spraying panel
CN112926563B (en) Fault diagnosis system for steel coil spray printing mark
CN107085850A (en) Masking gathers the marked body of foreign matter in image, recognizes the method and book scanning method of foreign matter marked body in image
CN110047063B (en) Material drop detection method, device, equipment and storage medium
CN115100206B (en) Printing defect identification method for textile with periodic pattern
CN111369516A (en) Transformer bushing heating defect detection method based on infrared image recognition
CN112419260A (en) PCB character area defect detection method
CN109993154A (en) The lithium sulfur type instrument intelligent identification Method of substation's simple pointer formula
CN114022439B (en) Flexible circuit board defect detection method based on morphological image processing
CN111753794A (en) Fruit quality classification method and device, electronic equipment and readable storage medium
CN111563896A (en) Image processing method for catenary anomaly detection
CN115619845A (en) Self-adaptive scanning document image inclination angle detection method
CN113723494A (en) Laser visual stripe classification and weld joint feature extraction method under uncertain interference source
CN115760696A (en) Surface scratch detection method
CN106530273B (en) High-precision FPC (Flexible printed Circuit) linear line detection and defect positioning method
CN113902894B (en) Automatic reading identification method for strip level based on image processing
CN109685791B (en) Corrugated board robust detection method based on multi-window autocorrelation texture detection
CN112418210B (en) Intelligent classification method for tower inspection information

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant