WO2015096535A1 - 残缺或变形的四边形图像的校正方法 - Google Patents

残缺或变形的四边形图像的校正方法 Download PDF

Info

Publication number
WO2015096535A1
WO2015096535A1 PCT/CN2014/088484 CN2014088484W WO2015096535A1 WO 2015096535 A1 WO2015096535 A1 WO 2015096535A1 CN 2014088484 W CN2014088484 W CN 2014088484W WO 2015096535 A1 WO2015096535 A1 WO 2015096535A1
Authority
WO
WIPO (PCT)
Prior art keywords
point
index
edge
pointsset
image
Prior art date
Application number
PCT/CN2014/088484
Other languages
English (en)
French (fr)
Inventor
王倩文
游晶
徐靖
谢德浩
Original Assignee
广州广电运通信息科技有限公司
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 广州广电运通信息科技有限公司 filed Critical 广州广电运通信息科技有限公司
Priority to EP14875817.0A priority Critical patent/EP3089103B1/en
Priority to US15/104,334 priority patent/US9773299B2/en
Priority to AU2014373249A priority patent/AU2014373249B2/en
Publication of WO2015096535A1 publication Critical patent/WO2015096535A1/zh
Priority to ZA2016/04301A priority patent/ZA201604301B/en

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/80Geometric correction
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00Geometric image transformations in the plane of the image
    • G06T3/18Image warping, e.g. rearranging pixels individually
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/12Edge-based segmentation

Definitions

  • the invention relates to a technology for identifying documents of value documents, bills and certificates, in particular to a document for price documents such as banknotes in the financial field, a ticket for a ticket, and an edge detection and correction method for a quadrilateral image such as an ID card.
  • an identification device such as a value document, a ticket, a document, or the like
  • a positional inclination, a motion deformation, or the like is often caused by a process such as placement or high-speed movement, and thus the generated image is often an image that is inclined and deformed. Therefore, when performing its recognition, the first step is often to perform edge detection and tilt correction on the acquired image, thereby identifying the tilt-corrected image content. Therefore, edge detection and tilt correction are an important issue in the design of identification systems such as value documents, tickets, and documents.
  • Valuable documents, notes, and ID images have the following characteristics during edge detection and tilt correction:
  • the shape is quadrilateral, but there may be damage on the four edges, and we need to obtain the edge of the value document, the ticket, and the document without damage.
  • the identification device has higher requirements for recognition time and storage space.
  • Such as ordinary deposit and withdrawal machines the speed of processing valuable media is very fast.
  • a plurality of identification items are included, such as identification of valuable document types, identification of valuable documents, identification of valuable documents, identification of valuable documents, identification, etc., and therefore The edge detection and tilt correction time is short.
  • more valuable media are implemented on embedded hardware platforms, and strict requirements are imposed on storage space.
  • Valuable documents During the movement of the identification device, there may be factors such as inconsistent friction coefficient, and the resulting image is deformed. Or there is an angle problem in the document scanning, and the resulting ID image has a ladder. Shape deformation, etc.
  • the commonly used edge detection algorithm is Hough transform.
  • Hough transform for edge detection, the edge points in the image are calculated, the corresponding points (r, ⁇ ) in the corresponding polar coordinate transformation domain are calculated, and the corresponding points in the transform domain are accumulated to obtain the points of maximum distribution. The point on the line where the edge is to be detected is then obtained and the point not on the edge is removed. Since the cosine and one sine calculation are needed for each point in the mapping process, the calculation amount is large, and the calculation is floating point, and the calculation time is long.
  • Another common edge detection algorithm is Canny edge detection.
  • the Canny operator is based on the edge detection operator of the optimization algorithm and has good signal-to-noise ratio and detection accuracy.
  • Image denoising is first performed using Gaussian filtering.
  • the magnitude and direction of the gradient are then calculated using the finite difference of the first-order partial derivative.
  • Non-maximum suppression is then applied to the gradient magnitude.
  • the edge is detected and connected using a double threshold algorithm. It has a large amount of calculation and a long calculation time.
  • the present invention provides a method for quickly performing edge detection and tilt correction for a valuable document, a ticket, and a document image with defects and deformations.
  • the method for correcting the incomplete or deformed quadrilateral image comprises: step one, edge point detection, using the difference of the gray value between the image area and the background area in the collected image, quickly detecting the edge point, and using the value document
  • the quadrilateral feature of the image the edge is a straight line, the edge point of each edge is equally spaced ⁇ W in the X direction, the finite edge point is obtained, and the edge fitting is performed to obtain the straight line equation;
  • the second step is to eliminate the abnormal edge point and eliminate the defect due to the defect. Abnormal points detected by wrinkles, etc.
  • step 3 the straight line is fitted, and the least square fitting is performed on the edge point point set after the abnormal point is removed to obtain the edge line equation; step 4, the vertex calculation is solved according to the first three steps.
  • step five image correction, in the bilinear space, using the proportional relationship, obtain the correspondence between the points before and after the correction, and obtain the gray-scale interpolation A tilt corrected image of a value document image.
  • the step of removing the abnormal point includes: two (1), slope calculation: assuming that two adjacent upper edge points are PointsSet_Up[index n-1 ] and PointsSet_Up[index n ], and the slope is:
  • Ks PointsSet_Up[index n ].y-PointsSet_Up[index n-1 ].y,
  • slope distribution statistics is performed with a quantization standard of 1. Generally, the angles of the upper, lower, left, and right edges are less than 90°. It is assumed that the inclination angle of the four sides in the system should be less than ⁇ , therefore, the slope calculated in the second (1) Ks is an integer, and its maximum value ks max , minimum value ks min , then:
  • the method for obtaining the correspondence between the points before and after the correction in step 5 is:
  • the points on the original image are shifted by X displacement and Y direction to obtain the corresponding points on the calibrated image. It is assumed that through the calculation of step four, four vertices A, B, C, and D are obtained. After the tilt correction, the corresponding points are A', B', C', D', and a point X' (x', y') in the obliquely corrected image is calculated and the corresponding point X (x, in the original image corresponding thereto is calculated. y) the relationship, including the steps:
  • the displacement of the X' point in the y direction in the corrected image is y', and the y' displacement in the y direction is also the E' point and the F' point, and the points corresponding to the original picture are respectively X points. , point E and point F;
  • the displacement of E' in the y direction corresponds to the image before the tilt correction, that is, the movement of the E point on the AC line, and the movement of the E' point on the A'C' line Proportional
  • the displacement of the F' point in the y direction corresponds to the image before the tilt correction, that is, the movement of the F point on the BD line, and the point of the F' point on the B'D' line Move proportionally,
  • the displacement of the X' point in the corrected image in the x direction is x', and the point X' moves in the x direction, that is, the movement on the E'F' line, and the movement of the X point on the EF line becomes proportion,
  • the coordinates of the X point obtained from the coordinates of E and F points are:
  • step 5 when the corresponding relationship between the points before and after the correction is obtained in step 5, if x' traverses from 0 to Width-1, y' traverses from 0 to Height-1, and the corrected image of the entire value document is obtained; when x' Only the partial values in [0, Width] and/or y' are taken only when the partial values in [0, Height] are taken, and the corrected image of the local region of interest on the image of the value document is obtained.
  • the algorithm for gray interpolation in step 5 includes nearest neighbor interpolation, bilinear interpolation or high order interpolation.
  • the method for correcting a defective or deformed quadrilateral image provided by the present invention is applied to a value document, a ticket
  • the difference between the background and the foreground part in the acquired image is used to quickly detect the edge point and avoid using various gradient operators.
  • the complicated calculation caused by the reduction of the edge point detection time, and the equal-point edge point detection in the image can reduce the edge point detection time.
  • the information of the adjacent previous edge point can be used to narrow the detection range of the edge point and reduce the edge point detection. time.
  • the method for correcting a defective or deformed quadrilateral image provided by the present invention can quickly perform edge detection and tilt correction on a value document, a ticket or a document image having a damaged or deformed shape.
  • FIG. 1 is a flow chart of a method for correcting a broken or deformed quadrilateral image according to an embodiment of the present invention
  • Figure 2 is a schematic diagram of an image with a missing edge
  • Figure 3 is a schematic diagram of the result of edge fitting without abnormal point culling
  • FIG. 5 is a schematic diagram of an image tilt correction map.
  • WIDTH The width of the entire image.
  • HEIGHT The height of the entire image.
  • Width Image width after tilt correction
  • ⁇ W step interval in the x direction when detecting the upper and lower edge points
  • ⁇ y y-direction floating range when detecting upper and lower edge points
  • ⁇ H y-direction step interval when detecting left and right edge points
  • ⁇ x the x-direction floating range when detecting the left and right edge points
  • a flowchart of a method for correcting a defective or deformed quadrilateral image includes five steps of edge point detection, abnormal edge point culling, line fitting, vertex calculation, and image correction. The following steps are described in detail:
  • Step 1 Edge point detection.
  • the value documents, bills, and documents are quadrilateral-shaped, and the edges are straight lines. Therefore, it is not necessary to detect all the edge points, and each edge only needs to detect a limited edge point, and the edge fitting can be performed to obtain a straight line equation. Therefore, equally spaced edge point detection is performed in the image to reduce the edge point detection time.
  • the information of the adjacent previous edge point can be used to narrow the detection range of the edge point and reduce the edge point detection. time.
  • the detection of the above edge points is as an example:
  • the detected upper edge point set is: (PointsSet_Up[index -lm ], PointsSet_Up[index -lm+1 ], ...PointsSet_Up[index -1 ], PointsSet_Up[index 0 ], PointsSet_Up[index 1 ], ...PointsSet_Up[index rm-1 ], PointsSet_Up[index rm ].
  • step two the abnormal edge points are removed.
  • the edge points detected on the image include the missing edge points, as shown in Figure 2, resulting in the detected edge point sets not being in a straight line, which affects The fitting accuracy of the subsequent edge line. Therefore, anomalous point culling of edge points is required before fitting.
  • the edge of the value document, the ticket, and the image of the certificate is less than the non-damaged portion, and the slope distribution of the adjacent edge points is counted, and the statistical slope distribution is the largest to be fitted.
  • the slope of the line is culled for edge points that are not in the maximum slope range.
  • the edge points are detected at equal intervals ⁇ W in the x direction, and the adjacent two upper edge points are PointsSet_Up[index n-1 ] and PointsSet_Up[index n ], and the slope thereof is
  • Ks PointsSet_Up[index n ].y-PointsSet_Up[index n-1 ].y,
  • the slope statistics can be designed according to the accuracy of the slope solution. That is, the slope value can be calculated by using the slope value as 1 as the quantization standard, or the slope value can be calculated by using the slope value as 2 as the quantization standard.
  • the quantization criterion is 2, it means that the distribution of the two slopes adjacent to the slope merges the cumulative statistics.
  • the angles of the upper, lower, left and right edges are less than 90°. It is assumed that the inclination angle of the four sides in the system should be less than ⁇ . Therefore, the slope ks calculated in the first step above is an integer, and the maximum value ksmax is set. The minimum value ksmin, then:
  • Ks min -[tan( ⁇ )* ⁇ W], where [] denotes rounding because ks is an integer.
  • the edge point detected by the damaged portion is not in line with the actual edge, and the slope thereof is different from the actual edge slope.
  • the most distributed ks calculated by the step (2) can be used, and the points corresponding to the ks that are not the most distributed are eliminated from the edge point set. Defines the upper edge point set after the exception point is removed as PointsSet_Up_New.
  • Step 3 Straight line fitting, performing least square fitting on the point set after the abnormal point is removed, and obtaining a straight line equation.
  • the corresponding edge line equations are respectively determined using the above steps.
  • the edge fitting effect before and after the abnormal point culling is different. After the abnormal point is removed, the edge fitting effect is closer to the edge of the image without the defect.
  • Step 4 the vertex calculation, using the four-edge edge straight line solution to obtain four vertices.
  • the intersection is obtained, that is, the vertices of the quadrilateral.
  • Step 5 Correct the image to obtain the image content after the tilt correction.
  • ID card identification it may only be necessary to identify the ID number.
  • ticket identification only the two-dimensional code in the lower right corner may be identified.
  • the tilt correction proposed in this section can be applied not only to the tilt correction of the full image of the value document, the ticket and the document image, but also to the tilt correction of the local interest area of the value document, the ticket and the document image.
  • the calibration process is divided into two steps:
  • the points on the original image are shifted by X displacement and Y direction to obtain corresponding points on the calibrated image.
  • step four Assume that through the calculation of step four, four vertices A, B, C, and D are obtained.
  • the corresponding points after the tilt correction are A', B', C', D', respectively, as shown in FIG.
  • the correlation between the point X' (x', y') in the obliquely corrected image and the corresponding point X (x, y) in the original image (the image before the tilt correction is not performed) is calculated as follows.
  • the displacement of the X' point in the y direction in the corrected image is y', and also the y' displacement in the y direction is also the E' point and the F' point.
  • the points corresponding to the original image are X point, E point and F point, respectively.
  • the displacement of E' in the y direction that is, the displacement of the A'C' line.
  • the movement of the E point on the AC straight line which is proportional to the movement of the E' point on the A'C' line.
  • the displacement in the y direction of the point F' that is, the displacement of the B'D' line.
  • the movement of the F point on the BD line which is proportional to the movement of the F' point on the B'D' line.
  • the displacement of the X' point in the x direction in the corrected image is x'.
  • the point X' moves in the x direction, that is, on the E'F' line, which is proportional to the movement of the X point on the EF line.
  • the coordinates of the X point obtained from the coordinates of E and F points are:
  • x' traverses from 0 to Width-1
  • y' traverses from 0 to Height-1
  • a corrected image of the entire value document, ticket, and certificate image can be obtained.
  • x' only takes a partial value in [0, Width]
  • y' only takes a partial value in [0, Height]
  • the corrected image of the local interest region on the value document, the ticket, and the document image is obtained.
  • the range of values of x', y' can be set according to the needs of the application.
  • an algorithm of gray level interpolation is needed to obtain the final corrected image, so that the corrected image maintains good continuity and consistency.
  • Specific interpolation methods include nearest neighbor interpolation, bilinear interpolation, and high order interpolation.
  • the nearest neighbor interpolation is the shortest in the calculation time, the bilinear interpolation is the second, and the high-order interpolation time is the longest.
  • the degree of smoothing after interpolation is the best for high-order interpolation, followed by bilinear interpolation, and the nearest neighbor is the worst.
  • the corresponding interpolation method can be selected according to the identification needs.
  • the incomplete or deformed quadrilateral image correction method provided by the embodiment is applied to the identification technology of documents such as value documents, tickets and documents, and the recognition speed and precision are improved.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)
  • Geometry (AREA)
  • Inspection Of Paper Currency And Valuable Securities (AREA)

Abstract

本发明涉及一种残缺或变形的四边形图像的校正方法,该方法包括边缘点检测、异常边缘点剔除、直线拟合、顶点计算以及图像校正五个步骤。该残缺或变形的四边形图像的校正方法应用于有价文件、票据、身份证等证件的识别方法核识别系统中,能对存在残损、变形的有价文件、票据或证件图像,快速地进行边缘检测和倾斜校正。

Description

残缺或变形的四边形图像的校正方法
本申请要求于2013年12月25日提交中国专利局、申请号为201310733877.7、发明名称为“残缺或变形的四边形图像的校正方法”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本发明涉及有价文件、票据、证件图像的识别技术,尤其是涉及金融领域的钞票等有价文件、车票等票据以及如身份证等证件四边形图像的边缘检测及校正方法。
背景技术
有价文件、票据、证件等识别装置中,常常由于放置或高速运动等过程发生位置倾斜,运动变形等,因此产生的图像多为倾斜、存在变形的图像。因此在进行其识别时,第一步常常是对所采集到的图像进行边缘检测和倾斜校正,进而对倾斜校正后的图像内容进行识别。因此,边缘检测、倾斜校正是有价文件、票据、证件等识别系统设计中的一个重要课题。
有价文件、票据、证件图像在边缘检测、倾斜校正过程中,具有以下特点:
1.外形为四边形,但在四个边缘上常常可能存在残损,而我们需要获取的是有价文件、票据、证件没有残损情况下的边缘。
2.识别装置对于识别时间、存储空间要求较高。如普通的存取款机,其处理有价介质的速度很快。而在每个有价介质的处理中,包含了多项识别项目,例如有价文件类型识别、有价文件冠字号识别、有价文件真伪识别、有价文件清分识别等,因此要求其边缘检测、倾斜校正的时间短。而且,较多的有价介质在嵌入式等硬件平台上实现,对于存储空间也提出了严格要求。
3.有价文件在识别装置运动过程中,可能存在摩擦系数不一致等因素,产生的图像存在变形。或者在证件扫描中存在角度问题,产生的证件图像存在梯 形变形等。
常用的边缘检测算法有Hough变换,Hough变换是从将直角坐标系映射到极坐标系,如一条直线在直角坐标系中表示为y=kx+b,在极坐标系下表示为r=x cos(θ)+y sin(θ),直角坐标系中任何一条直线对应极坐标系的一个点。利用Hough变换进行边缘检测时,是将图像中的边缘点,计算其对应极坐标变换域中对应的点(r,θ),对变换域中的对应点进行累计,获取其最大分布的点,进而得到所要检测边缘的直线上的点,并剔除不在该边缘上的点。由于映射过程中,对于每一个点均需要进行一次余弦、一次正弦计算,计算量较大,且为浮点计算,计算时间较长。
另一种常见的边缘检测算法还有Canny边缘检测。Canny算子是基于最优化算法边缘检测算子,具有良好的信噪比和检测精度。首先利用高斯滤波进行图像去噪。接着用一阶偏导的有限差分计算梯度的幅值和方向。然后通过对梯度幅值应用非极大值抑制。最后用双阈值算法检测并且连接边缘。其计算量较大,计算时间较长。
因此,需要提出一种能对存在残损、变形的有价文件、票据或证件图像,快速进行边缘检测和倾斜校正的方法。
发明内容
为了解决现有技术中边缘检测算法计算时间较长的问题,本发明提供一种针对存在残缺、变形的有价文件、票据、证件图像,快速进行边缘检测和倾斜校正的方法。
该残缺或变形的四边形图像的校正方法,包括:步骤一、边缘点检测,利用采集到的图像中图像区域与背景区域的灰度值差异,快速地进行边缘点的检测,且利用有价文件图像的四边形特征,边缘为直线,对每条边在X方向进行等间隔ΔW的边缘点检测,获取有限的边缘点,进行边缘拟合获取直线方程;步骤二、异常边缘点剔除,剔除由于残缺、褶皱等情况而检测到的异常点,提 高直线拟合精度;利用有价文件图像的边缘残损的部分小于非残损部分的特点,统计其相邻边缘点的斜率分布,统计斜率分布最大的作为待拟合直线的斜率,对于不在最大斜率范围的边缘点进行剔除;步骤三、直线拟合,对剔除异常点后的边缘点点集进行最小二乘拟合,获得其边缘直线方程;步骤四、顶点计算,根据前三个步骤求解得到的四条边缘直线,利用直线交点求解获得四边形的四个顶点;以及步骤五、图像校正,在双线性空间内,利用比例关系,获取校正前后各点的对应关系,并通过灰度插值,从而获得有价文件图像的倾斜校正图像。
具体的,步骤一中,检测上边缘点的方法包括:一(1),在x=WIDTH/2直线上自上而下搜索上边缘点,得到的边缘点为PointsSet_Up[index0]=(x_up_index0,y_up_index0);一(2),在x=x_up_index0-ΔW直线上检测边缘点,其y的搜索范围是[y_up_index0-Δy,y_up_index0+Δy],搜索到的边缘点为PointsSet_Up[index-1]=(x_up_index-1,y_up_index-1);一(3),以PointsSet_Up[index-1]为原点,重复进行一(2)的操作,进行边缘点检测,直到在设定的搜索范围内检测不到边缘点为止;一(4),在x=x_up_index0+ΔW直线上检测边缘点,其y的搜索范围是[y_up_index0-Δy,y_up_index0+Δy],搜索到的边缘点为PointsSet_Up[index1]=(x_up_index1,y_up_index1);一(5),以PointsSet_Up[index1]为原点,重复进行一(4)的操作,进行边缘点检测,直到在设定的搜索范围内检测不到边缘点为止;以及一(6),检测到的上边缘点集为:(PointsSet_Up[index-lm],PointsSet_Up[index-lm+1],……PointsSet_Up[index-1],PointsSet_Up[index0],PointsSet_Up[index1],……PointsSet_Up[indexrm-1],PointsSet_Up[indexrm]。
具体的,步骤二中,剔除异常点的步骤包括:二(1),斜率计算:假设相邻的两个上边缘点为PointsSet_Up[indexn-1]和PointsSet_Up[indexn],其斜率为:
Figure PCTCN2014088484-appb-000001
由于
PointsSet_Up[indexn].x-PointsSet_Up[indexn-1].x=ΔW,
因此,
Figure PCTCN2014088484-appb-000002
由于ΔW为常量,因此评估斜率时,直接利用ks表示k,
ks=PointsSet_Up[indexn].y-PointsSet_Up[indexn-1].y,
k~=ks;
二(2),斜率分布统计:
以量化标准为1进行斜率分布统计示例,一般而言,上、下、左、右边缘的角度小于90°,现假设系统中四边倾斜角度应小于θ,因此,二(1)中计算的斜率ks为整数,设其最大值ksmax,最小值ksmin,则:
Figure PCTCN2014088484-appb-000003
Figure PCTCN2014088484-appb-000004
因此,
ksmax=[tan(θ)*ΔW],
ksmin=-[tan(θ)*ΔW],其中[]表示取整;
对[-[tan(θ)*ΔW],[tan(θ)*ΔW]]范围内的ks进行累计统计,得到最多的斜率分布;以及
二(3),边缘点异常点剔除,根据二(2)步计算得到的最多分布的ks,将 不是最多分布的ks对应的点从边缘点集中剔除,定义剔除了异常点后的上边缘点集为PointsSet_Up_New。
具体的,步骤三获得边缘直线方程的方法为:假设求解的上边缘直线方程为y=kx+b,剔除了异常点后的上边缘点集为PointsSet_Up_New,X表示PointsSet_Up_New.x,Y表示PointsSet_Up_New.y,n为PointsSet_Up_New的点数个数,则
Figure PCTCN2014088484-appb-000005
Figure PCTCN2014088484-appb-000006
对下、左、右边缘,相似地分别求出相应的边缘直线方程。
具体的,步骤五中获取校正前后各点的对应关系的方法为:
在双线性变换空间中,原图像上的点经过X位移、Y方向位移,得到校准后图像上的对应点,假设通过步骤四的计算,得到四个顶点A、B、C、D,其通过倾斜校正后对应点分别为A’,B’,C’,D’,计算倾斜校正后图像中的一点X’(x’,y’)与其对应的原图像中的对应点X(x,y)的相互关系,包括步骤:
五(1),计算y方向上产生的位移:
X’点在校正后图像中的在y方向上的位移为y’,同样在y方向上有y’位移的还有E’点和F’点,其对应原图的点分别为为X点、E点和F点;
E点坐标:
E’在y方向的位移,即A’C’直线的位移,对应于倾斜校正前的图像,即为E点在AC直线上的移动,其与E’点在A’C’直线上的移动成比例;
Figure PCTCN2014088484-appb-000007
Figure PCTCN2014088484-appb-000008
也即【式1】:
Figure PCTCN2014088484-appb-000009
Figure PCTCN2014088484-appb-000010
F点坐标:
F’点在y方向的位移,即B’D’直线的位移,对应于倾斜校正前的图像,即为F点在BD直线上的移动,其与F’点在B’D’直线上的移动成比例,
Figure PCTCN2014088484-appb-000011
Figure PCTCN2014088484-appb-000012
利用校正后图像为四边形,
yD'=yC',yB'=yA',yF'=yE'
因此,
Figure PCTCN2014088484-appb-000013
Figure PCTCN2014088484-appb-000014
也即【式2】:
Figure PCTCN2014088484-appb-000015
Figure PCTCN2014088484-appb-000016
五(2),计算x方向上的位移:
X’点在校正后图像中的在x方向上的位移为x’,X’点在x方向移动,也就是在E’F’直线上的移动,其与X点在EF直线上的移动成比例,
由E、F点坐标得到X点坐标为:
Figure PCTCN2014088484-appb-000017
Figure PCTCN2014088484-appb-000018
也即【式3】:
Figure PCTCN2014088484-appb-000019
Figure PCTCN2014088484-appb-000020
通过【式1】【式2】【式3】,得到校正后图像上任何一点(x’,y’)其对应原图像上的点(x,y)的对应关系。
具体的,步骤五中获取校正前后各点的对应关系时,若x’从0遍历到Width-1,y’从0遍历到Height-1,得到整幅有价文件的校正图像;当x’只取[0,Width]中的部分值和/或y’只取[0,Height]中的部分值时,得到有价文件图像上局部感兴趣区域的校正图像。
优选的,步骤五中灰度插值的算法包括最近邻插值法、双线性插值法或高阶插值法。
本发明提供的残缺或变形的四边形图像的校正方法应用于有价文件、票 据、身份证等证件的识别方法核识别系统中,该校正方法的步骤一中利用采集到的图像中的背景与前景部分的差异,快速地进行边缘点的检测,避免利用各种梯度算子造成的复杂计算,减少边缘点检测时间,而且,在图像中进行等间隔的边缘点检测,可减少边缘点检测时间。此外,在同一条边缘的检测过程中,利用相邻边缘点坐标相近的特点,检测下一边缘点时可以借助相邻的上一边缘点的信息,缩小其边缘点检测范围,减少边缘点检测时间。在步骤二中进行斜率计算时提出简化计算方法,将带浮点除法的斜率计算,转换为整数型的减法计算,降低计算复杂度和减少计算时间。因此本发明提供的残缺或变形的四边形图像的校正方法能对存在残损、变形的有价文件、票据或证件图像,快速地进行边缘检测和倾斜校正。
附图说明
图1是本发明一实施例提供的残缺或变形的四边形图像的校正方法流程图;
图2是边缘存在残缺的图像示意图;
图3是没有经过异常点剔除的边缘拟合结果示意图;
图4是经过异常点剔除后的边缘拟合结果示意图;
图5是图像倾斜校正映射示意图。
参数说明:
WIDTH:整幅图像的宽度。
HEIGHT:整幅图像的高度。
Width:倾斜校正后图像宽度
Height:倾斜校正后图像高度
x:图像的水平方向变量
y:图像的竖直方向变量
上边缘点集PointsSet_Up
下边缘点集PointsSet_Down
左边缘点集PointsSet_Left
右边缘点集PointsSet_Right
上边缘直线Line_Up
下边缘直线Line_Down
左边缘直线Line_Left
右边缘直线Line_Right
ΔW:检测上下边缘点时的x方向步进间隔
Δy:检测上下边缘点时的y方向浮动范围
ΔH:检测左右边缘点时的y方向步进间隔
Δx:检测左右边缘点时的x方向浮动范围
具体实施方式
为进一步阐述本发明所提供的残缺或变形的四边形图像的校正方法,以下结合本发明的一个优选实施例的图示做进一步的详细介绍。
参阅图1,该残缺或变形的四边形图像的校正方法流程图,该方法包括边缘点检测,异常边缘点剔除,直线拟合,顶点计算以及图像校正五个步骤,以下按步骤进行详细介绍:
步骤一:边缘点检测。
有价文件、票据、证件成像过程中,图像区域和背景区域的灰度值存在差异。我们可以利用该差异,以不在非背景灰度范围,快速寻找到各边的边缘点,避免利用各种梯度算子造成的复杂计算,减少边缘点检测时间。
另外,利用有价文件、票据、证件为类四边形,边缘为直线,因此不需要检测所有的边缘点,每条边只需检测有限的边缘点,即可进行边缘拟合获取直线方程。因此,在图像中进行等间隔的边缘点检测,减少边缘点检测时间。
此外,在同一条边缘的检测过程中,利用相邻边缘点坐标相近的特点,检测下一边缘点时可以借助相邻的上一边缘点的信息,缩小其边缘点检测范围,减少边缘点检测时间。
具体地,以上边缘点的检测为例:
(1)在x=WIDTH/2直线上自上而下搜索上边缘点,得到的边缘点为PointsSet_Up[index0]=(x_up_index0,y_up_index0)。
(2)在x=x_up_index0-ΔW直线上检测边缘点,其y的搜索范围是[y_up_index0-Δy,y_up_index0+Δy]。搜索到的边缘点为PointsSet_Up[index-1]=(x_up_index-1,y_up_index-1)。
(3)以PointsSet_Up[index-1]为原点,重复进行(2)的操作,进行边缘点检测,直到在设定的搜索范围内检测不到边缘点为止。
(4)在x=x_up_index0+ΔW直线上检测边缘点,其y的搜索范围是[y_up_index0-Δy,y_up_index0+Δy]。搜索到的边缘点为PointsSet_Up[index1]=(x_up_index1,y_up_index1)。
(5)以PointsSet_Up[index1]为原点,重复进行(4)的操作,进行边缘点检测,直到在设定的搜索范围内检测不到边缘点为止。
(6)检测到的上边缘点集为:(PointsSet_Up[index-lm],PointsSet_Up[index-lm+1],……PointsSet_Up[index-1],PointsSet_Up[index0],PointsSet_Up[index1],……PointsSet_Up[indexrm-1],PointsSet_Up[indexrm]。
步骤二,异常边缘点剔除。
有价文件、票据、证件由于存在残损、褶皱等情况,图像上检测到的边缘点包括了残缺的边缘点,如图2所示,导致检测到的边缘点集不全在一条直线上,这影响了后续的边缘直线的拟合精度。因此,在拟合之前,需要进行边缘点的异常点剔除。
本实施例利用有价文件、票据、证件图像的边缘残损的部分要小于非残损部分的特点,统计其相邻边缘点的斜率分布,统计斜率分布最大的作为待拟合 直线的斜率,对于不在最大斜率范围的边缘点进行剔除。此处,特别地,精简了斜率的计算和斜率分布统计。
同样地,利用上边缘为例子,按以下步骤计算:
(1)斜率计算
在步骤一的边缘点检测中,在x方向是等间隔ΔW的进行边缘点的检测,假设相邻的两个上边缘点为PointsSet_Up[indexn-1]和PointsSet_Up[indexn],其斜率为
Figure PCTCN2014088484-appb-000021
由于PointsSet_Up[indexn].x-PointsSet_Up[indexn-1].x=ΔW,
因此,
Figure PCTCN2014088484-appb-000022
由于ΔW为常量,因此评估斜率时,我们直接利用ks表示k,
ks=PointsSet_Up[indexn].y-PointsSet_Up[indexn-1].y,
k~=ks。
这样,将斜率的计算从带浮点运算的除法,简化为整数的减法运算。
(2)斜率分布统计
对斜率分布进行量化统计。在实际应用中,可以根据斜率求解的精度,设计斜率的量化统计,即可以对斜率值以1作为量化标准进行斜率分布统计,也可以对斜率值以2作为量化标准进行斜率分布统计等。当量化标准为2时,意味着斜率相邻的2个斜率的分布合并累计统计。
以下以量化标准为1进行斜率分布统计的示例。
一般而言,上、下、左、右边缘的角度小于90°,现假设系统中四边倾斜角度应小于θ,因此,上面第一步计算的斜率ks为整数,设其最大值ksmax, 最小值ksmin,则:
Figure PCTCN2014088484-appb-000023
Figure PCTCN2014088484-appb-000024
因此,
ksmax=[tan(θ)*ΔW]
ksmin=-[tan(θ)*ΔW],其中[]表示取整,因为ks为整数。
因此,只需对[-[tan(θ)*ΔW],[tan(θ)*ΔW]]范围内的ks进行累计统计,得到最多的斜率分布。
(3)边缘点异常点剔除
如上所述,如遇有价文件、票据、证件图像的边缘出现残损等情况,残损部分检测到的边缘点与实际边缘不在一条直线上,其斜率与实际边缘斜率存在差异。而且假设残缺部分小于整体边缘的1/2,则可以通过第(2)步计算得到的最多分布的ks,将不是最多分布的ks对应的点从边缘点集中剔除。定义剔除了异常点后的上边缘点集为PointsSet_Up_New。
步骤三,直线拟合,对剔除异常点后的点集进行最小二乘拟合,得到直线方程。
假设求解的上边缘直线方程为y=kx+b,剔除了异常点后的上边缘点集为PointsSet_Up_New,X表示PointsSet_Up_New.x,Y表示PointsSet_Up_New.y,n为PointsSet_Up_New的点数个数。则
Figure PCTCN2014088484-appb-000025
Figure PCTCN2014088484-appb-000026
对下、左、右边缘,相似地,利用上述步骤分别求出相应的边缘直线方程。如图3和图4所示,异常点剔除前后边缘拟合效果是不同的,异常点剔除后,边缘拟合效果更接近未残缺变形的图像的边缘。
步骤四,顶点计算,利用四边边缘直线求解得到四个顶点。
利用两相交直线方程,求解得到交点,也即四边形的顶点。
对上边缘直线与左边缘直线求解其交点,得到左上顶点A,对上边缘直线与右边缘直线求解其交点,得到右上顶点B,对下边缘直线与左边缘直线求解其交点,得到左下顶点C,对下边缘直线与右边缘直线求解其交点,得到右下顶点D,如图5所示。
步骤五,校正图像,获取倾斜校正后的图像内容。
采集到有价文件、票据、证件图像后,应用中可能只存在识别有价文件、票据、证件中的部分内容的需求。如,在身份证识别中,可能只需要识别身份证号码。而票据识别中可能只识别其右下角的二维码等。
因此,本部分提出的倾斜校正不但可以适用于对有价文件、票据、证件图像进行的全图的倾斜校正,可适用于有价文件、票据、证件图像局部感兴趣区域的倾斜校正。
校正的过程分两步:
(1)计算校正前后的变换关系。
在双线性变换空间中,原图像上的点经过X位移、Y方向位移,得到校准后图像上的对应点。
假设通过步骤四的计算,得到四个顶点A、B、C、D。其通过倾斜校正后对应点分别为A’,B’,C’,D’,如图5所示。
以下计算倾斜校正后图像中的一点X’(x’,y’)与其对应的原图像(未进行倾斜校正之前图像)中的对应点X(x,y)的相互关系。
(1.1)首先计算y方向的产生的位移。
X’点在校正后图像中的在y方向上的位移为y’,同样在y方向上有y’位移的还有E’点和F’点。其对应原图的点分别为为X点、E点和F点。
E点坐标:
E’的在y方向的位移,即A’C’直线的位移。对应于倾斜校正前的图像,即为E点在AC直线上的移动,其与E’点在A’C’直线上的移动成比例。
Figure PCTCN2014088484-appb-000027
Figure PCTCN2014088484-appb-000028
也即【式1】:
Figure PCTCN2014088484-appb-000029
Figure PCTCN2014088484-appb-000030
F点坐标:
F’点的在y方向的位移,即B’D’直线的位移。对应于倾斜校正前的图像,即为F点在BD直线上的移动,其与F’点在B’D’直线上的移动成比例。
Figure PCTCN2014088484-appb-000031
Figure PCTCN2014088484-appb-000032
利用校正后图像为四边形,
yD'=yC',yB'=yA',yF'=yE'
因此,
Figure PCTCN2014088484-appb-000033
Figure PCTCN2014088484-appb-000034
也即【式2】:
Figure PCTCN2014088484-appb-000035
Figure PCTCN2014088484-appb-000036
(1.2)计算x方向上的位移。
X’点在校正后图像中的在x方向上的位移为x’。X’点在x方向移动,也就是在E’F’直线上的移动,其与X点在EF直线上的移动成比例。
由E、F点坐标得到X点坐标为:
Figure PCTCN2014088484-appb-000037
Figure PCTCN2014088484-appb-000038
也即【式3】:
Figure PCTCN2014088484-appb-000039
Figure PCTCN2014088484-appb-000040
这样通过【式1】【式2】【式3】,我们可以得到校正后图像上任何一点(x’,y’)其对应原图像上的点(x,y)的对应关系。
若x’从0遍历到Width-1,y’从0遍历到Height-1,可以得到整幅有价文件、票据、证件图像的校正图像。当x’只取[0,Width]中的部分值,或者y’只取[0,Height]中的部分值时,得到的是有价文件、票据、证件图像上局部感兴趣区域的校正图像。实际应用中可根据应用需要设定x’,y’的取值范围。
(2)灰度插值
在得到倾斜校正前后点的对应关系后,还需要灰度级插值的算法,来获取最终的校正后图像,使得校正后图像保持较好的连续性和连贯性。具体的插值方法有最近邻插值、双线性插值、高阶插值等。计算时间上最近邻插值最短,双线性插值次之,高阶插值时间最长。而插值后平滑程度是高阶插值最好,双线性插值次之,最近邻插值最差。实际应用中可根据识别需要选择对应的插值方式。
本实施例提供的残缺或变形的四边形图像校正方法应用于有价文件、票据以及证件等文件的识别技术中,提高识别速度和精度。
以上仅是本发明的优选实施方式,应当指出的是,上述优选实施方式不应视为对本发明的限制,本发明的保护范围应当以权利要求所限定的范围为准。对于本技术领域的普通技术人员来说,在不脱离本发明的精神和范围内,还可以做出若干改进和润饰,这些改进和润饰也应视为本发明的保护范围。

Claims (7)

  1. 一种残缺或变形的四边形图像的校正方法,包括:
    步骤一、边缘点检测,利用采集到的图像中图像区域与背景区域的灰度值差异,快速地进行边缘点的检测,且利用有价文件图像的四边形特征,边缘为直线,对每条边在X方向进行等间隔ΔW的边缘点检测,获取有限的边缘点,进行边缘拟合获取直线方程;
    步骤二、异常边缘点剔除,剔除由于残缺、褶皱等情况而检测到的异常点,提高直线拟合精度;利用有价文件图像的边缘残损的部分小于非残损部分的特点,统计其相邻边缘点的斜率分布,统计斜率分布最大的作为待拟合直线的斜率,对于不在最大斜率范围的边缘点进行剔除;
    步骤三、直线拟合,对剔除异常点后的边缘点点集进行最小二乘拟合,获得其边缘直线方程;
    步骤四、顶点计算,根据前三个步骤求解得到的四条边缘直线,利用直线交点求解获得四边形的四个顶点;
    步骤五、图像校正,在双线性空间内,利用比例关系,获取校正前后各点的对应关系,并通过灰度插值,从而获得有价文件图像的倾斜校正图像。
  2. 如权利要求1所述的残缺或变形的四边形图像的校正方法,其特征在于,步骤一中,检测上边缘点的方法包括:
    一(1),在x=WIDTH/2直线上自上而下搜索上边缘点,得到的边缘点为PointsSet_Up[index0]=(x_up_index0,y_up_index0);
    一(2),在x=x_up_index0-ΔW直线上检测边缘点,其y的搜索范围是[y_up_index0-Δy,y_up_index0+Δy],搜索到的边缘点为 PointsSet_Up[index-1]=(x_up_index-1,y_up_index-1);
    一(3),以PointsSet_Up[index-1]为原点,重复进行一(2)的操作,进行边缘点检测,直到在设定的搜索范围内检测不到边缘点为止;
    一(4),在x=x_up_index0+ΔW直线上检测边缘点,其y的搜索范围是[y_up_index0-Δy,y_up_index0+Δy],搜索到的边缘点为PointsSet_Up[index1]=(x_up_index1,y_up_index1);
    一(5),以PointsSet_Up[index1]为原点,重复进行一(4)的操作,进行边缘点检测,直到在设定的搜索范围内检测不到边缘点为止;以及
    一(6),检测到的上边缘点集为:(PointsSet_Up[index-lm],PointsSet_Up[index-lm+1],……PointsSet_Up[index-1],PointsSet_Up[index0],PointsSet_Up[index1],……PointsSet_Up[indexrm-1],PointsSet_Up[indexrm]。
  3. 如权利要求2所述的残缺或变形的四边形图像的校正方法,其特征在于,步骤二中,
    剔除异常点的步骤包括:
    二(1),斜率计算:
    假设相邻的两个上边缘点为PointsSet_Up[indexn-1]和PointsSet_Up[indexn],其斜率为:
    Figure PCTCN2014088484-appb-100001
    由于
    PointsSet_Up[indexn].x-PointsSet_Up[indexn-1].x=ΔW,
    因此,
    Figure PCTCN2014088484-appb-100002
    由于ΔW为常量,因此评估斜率时,直接利用ks表示k,
    ks=PointsSet_Up[indexn].y-PointsSet_Up[indexn-1].y
    k~=ks;
    二(2),斜率分布统计:
    以量化标准为1进行斜率分布统计示例,一般而言,上、下、左、右边缘的角度小于90°,现假设系统中四边倾斜角度应小于θ,因此,二(1)中计算的斜率ks为整数,设其最大值ksmax,最小值ksmin,则:
    Figure PCTCN2014088484-appb-100003
    Figure PCTCN2014088484-appb-100004
    因此,
    ksmax=[tan(θ)*ΔW],
    ksmin=-[tan(θ)*ΔW],其中[]表示取整;
    对[-[tan(θ)*ΔW],[tan(θ)*ΔW]]范围内的ks进行累计统计,得到最多的斜率分布;
    二(3),边缘点异常点剔除
    根据二(2)步计算得到的最多分布的ks,将不是最多分布的ks对应的点从边缘点集中剔除,定义剔除了异常点后的上边缘点集为PointsSet_Up_New。
  4. 如权利要求3所述的残缺或变形的四边形图像的校正方法,其特征在于,步骤三具体获得边缘直线方程的方法为:
    假设求解的上边缘直线方程为y=kx+b,剔除了异常点后的上边缘点集为PointsSet_Up_New,X表示PointsSet_Up_New.x,Y表示PointsSet_Up_New.y,n为PointsSet_Up_New的点数个数,则
    Figure PCTCN2014088484-appb-100005
    Figure PCTCN2014088484-appb-100006
    对下、左、右边缘,相似地分别求出相应的边缘直线方程。
  5. 如权利要求4所述的残缺或变形的四边形图像的校正方法,其特征在于,步骤五中获取校正前后各点的对应关系的方法为:
    在双线性变换空间中,原图像上的点经过X位移、Y方向位移,得到校准后图像上的对应点,假设通过步骤四的计算,得到四个顶点A、B、C、D,其通过倾斜校正后对应点分别为A’,B’,C’,D’,计算倾斜校正后图像中的一点X’(x’,y’)与其对应的原图像中的对应点X(x,y)的相互关系,包括步骤:
    五(1),计算y方向上产生的位移:
    X’点在校正后图像中的在y方向上的位移为y’,同样在y方向上有y’位移的还有E’点和F’点,其对应原图的点分别为为X点、E点和F点;
    E点坐标:
    E’在y方向的位移,即A’C’直线的位移,对应于倾斜校正前的图像,即为E点在AC直线上的移动,其与E’点在A’C’直线上的移动成比例;
    Figure PCTCN2014088484-appb-100007
    Figure PCTCN2014088484-appb-100008
    也即【式1】:
    Figure PCTCN2014088484-appb-100009
    Figure PCTCN2014088484-appb-100010
    F点坐标:
    F’点在y方向的位移,即B’D’直线的位移,对应于倾斜校正前的图像,即为F点在BD直线上的移动,其与F’点在B’D’直线上的移动成比例,
    Figure PCTCN2014088484-appb-100011
    Figure PCTCN2014088484-appb-100012
    利用校正后图像为四边形,
    yD'=yC',yB'=yA',yF'=yE'
    因此,
    Figure PCTCN2014088484-appb-100013
    Figure PCTCN2014088484-appb-100014
    也即【式2】:
    Figure PCTCN2014088484-appb-100015
    Figure PCTCN2014088484-appb-100016
    五(2),计算x方向上的位移:
    X’点在校正后图像中的在x方向上的位移为x’,X’点在x方向移动,也就是在E’F’直线上的移动,其与X点在EF直线上的移动成比例,
    由E、F点坐标得到X点坐标为:
    Figure PCTCN2014088484-appb-100017
    Figure PCTCN2014088484-appb-100018
    也即【式3】:
    Figure PCTCN2014088484-appb-100019
    Figure PCTCN2014088484-appb-100020
    通过【式1】【式2】【式3】,得到校正后图像上任何一点(x’,y’)其对应原图像上的点(x,y)的对应关系。
  6. 如权利要求5所述的残缺或变形的四边形图像的校正方法,其特征在于,步骤五中获取校正前后各点的对应关系时,若x’从0遍历到Width-1,y’从0遍历到Height-1,得到整幅有价文件的校正图像;当x’只取[0,Width]中的部分值和/或y’只取[0,Height]中的部分值时,得到有价文件图像上局部感兴趣区域的校正图像。
  7. 如权利要求4~6中任意一项所述的残缺或变形的四边形图像的校正方法,其特征在于,步骤五中灰度插值的算法包括最近邻插值法、双线性插值法或高阶插值法。
PCT/CN2014/088484 2013-12-25 2014-10-13 残缺或变形的四边形图像的校正方法 WO2015096535A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
EP14875817.0A EP3089103B1 (en) 2013-12-25 2014-10-13 Method for correcting fragmentary or deformed quadrangular image
US15/104,334 US9773299B2 (en) 2013-12-25 2014-10-13 Method for correcting fragmentary or deformed quadrangular image
AU2014373249A AU2014373249B2 (en) 2013-12-25 2014-10-13 Method for correcting fragmentary or deformed quadrangular image
ZA2016/04301A ZA201604301B (en) 2013-12-25 2016-06-24 Method for correcting fragmentary or deformed quadrangular image

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201310733877.7 2013-12-25
CN201310733877.7A CN103679638A (zh) 2013-12-25 2013-12-25 残缺或变形的四边形图像的校正方法

Publications (1)

Publication Number Publication Date
WO2015096535A1 true WO2015096535A1 (zh) 2015-07-02

Family

ID=50317095

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/088484 WO2015096535A1 (zh) 2013-12-25 2014-10-13 残缺或变形的四边形图像的校正方法

Country Status (8)

Country Link
US (1) US9773299B2 (zh)
EP (1) EP3089103B1 (zh)
CN (1) CN103679638A (zh)
AU (1) AU2014373249B2 (zh)
CL (1) CL2016001485A1 (zh)
TR (1) TR201819357T4 (zh)
WO (1) WO2015096535A1 (zh)
ZA (1) ZA201604301B (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106874818A (zh) * 2016-08-30 2017-06-20 阿里巴巴集团控股有限公司 一种数字对象唯一标识符doi识别方法与装置
CN112085708A (zh) * 2020-08-19 2020-12-15 浙江华睿科技有限公司 产品外轮廓中的直线边缘的缺陷检测方法及设备
CN112633275A (zh) * 2020-12-22 2021-04-09 航天信息股份有限公司 一种基于深度学习的多票据混拍图像校正方法及系统
CN113052896A (zh) * 2019-12-27 2021-06-29 大族激光科技产业集团股份有限公司 视觉定位方法及装置

Families Citing this family (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679638A (zh) 2013-12-25 2014-03-26 广州广电运通金融电子股份有限公司 残缺或变形的四边形图像的校正方法
CN104361672B (zh) * 2014-10-14 2017-03-15 深圳怡化电脑股份有限公司 一种对纸币折角进行检测的方法
KR102279026B1 (ko) * 2014-11-07 2021-07-19 삼성전자주식회사 적어도 하나의 객체를 포함하는 영상에서 객체를 추출하여 보정한 영상을 제공하는 장치 및 방법
CN106845508B (zh) 2015-12-07 2019-05-17 腾讯科技(深圳)有限公司 一种检测图像中信息卡的方法、装置和系统
CN106023136B (zh) * 2016-04-29 2019-02-12 北京小米移动软件有限公司 边界优化方法及装置
JP6714477B2 (ja) * 2016-09-09 2020-06-24 株式会社アドテックエンジニアリング 基板角位置特定方法
CN106898083A (zh) * 2017-03-01 2017-06-27 深圳怡化电脑股份有限公司 一种用于纸币的图像处理方法及装置
CN106960184B (zh) * 2017-03-08 2020-03-31 南昌航空大学 一种书本边框的图像定位方法
US10140693B2 (en) * 2017-03-23 2018-11-27 Intergraph Corporation Motion imagery corner point sequencer
CN107169493A (zh) * 2017-05-31 2017-09-15 北京小米移动软件有限公司 信息识别方法及装置
CN109035266B (zh) * 2017-06-08 2022-11-15 杭州睿沃科技有限公司 一种利用普通摄像完成身份证便携扫描的方法
CN107451569A (zh) * 2017-08-04 2017-12-08 深圳易嘉恩科技有限公司 一种自动识别并裁切扫描件中票据的方法
CN107563330B (zh) * 2017-09-04 2020-10-16 南京邮电大学 一种监控视频中的水平倾斜车牌矫正方法
EP3786844A4 (en) * 2018-07-06 2021-05-05 Rakuten, Inc. IMAGE PROCESSING SYSTEM, IMAGE PROCESSING METHOD AND PROGRAM
US11881043B2 (en) 2018-07-06 2024-01-23 Rakuten Group, Inc. Image processing system, image processing method, and program
CN109034050B (zh) * 2018-07-23 2022-05-03 顺丰科技有限公司 基于深度学习的身份证图像文本识别方法及装置
CN109413330A (zh) * 2018-11-07 2019-03-01 深圳市博纳思信息技术有限公司 一种证件照片智能换背景方法
CN109493383B (zh) * 2018-11-23 2022-02-11 深圳市威尔德医疗电子有限公司 超声图像中内中膜厚度的测量方法、服务器及存储介质
CN110490886B (zh) * 2019-07-31 2022-04-01 武汉大学 一种针对倾斜视角下证件图像的自动纠正方法及系统
CN111091505B (zh) * 2019-11-26 2022-06-24 浙江大学 基于古典概型的影印图像倾斜纠正方法
CN113255579B (zh) * 2021-06-18 2021-09-24 上海建工集团股份有限公司 一种施工监测异常采集数据自动识别与处理的方法
CN114022661B (zh) * 2021-11-08 2024-10-15 广东电网有限责任公司 应用于票据影像字符识别的数字图像处理系统及方法
CN114240983A (zh) * 2021-12-21 2022-03-25 哈尔滨工业大学芜湖机器人产业技术研究院 一种圆形物体检测方法及系统
CN114648542A (zh) * 2022-03-11 2022-06-21 联宝(合肥)电子科技有限公司 一种目标物提取方法、装置、设备及可读存储介质
CN115082935A (zh) * 2022-07-04 2022-09-20 网易有道信息技术(北京)有限公司 用于对文档图像进行矫正的方法、设备及存储介质
CN116030120B (zh) * 2022-09-09 2023-11-24 北京市计算中心有限公司 一种识别并校正六边形的方法
CN115861357B (zh) * 2023-02-27 2023-06-20 常州微亿智造科技有限公司 基于k均值聚类与点位规划的工件过渡边缘检测方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6111996A (en) * 1998-03-13 2000-08-29 Northern Telecom Limited Optical multiplexer/demultiplexer
CN102222229A (zh) * 2011-07-28 2011-10-19 陈庆武 手指静脉图像预处理方法
CN103679638A (zh) * 2013-12-25 2014-03-26 广州广电运通金融电子股份有限公司 残缺或变形的四边形图像的校正方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH10208056A (ja) * 1997-01-16 1998-08-07 Honda Motor Co Ltd 直線検出方法
US6433896B1 (en) * 1997-06-10 2002-08-13 Minolta Co., Ltd. Image processing apparatus
JP4508553B2 (ja) * 2003-06-02 2010-07-21 カシオ計算機株式会社 撮影画像投影装置、及び撮影画像の補正方法
JP2005316755A (ja) * 2004-04-28 2005-11-10 Nec Electronics Corp 2次元矩形コードシンボル読み取り装置及び2次元矩形コードシンボル読み取り方法
JP4289414B2 (ja) * 2007-03-27 2009-07-01 セイコーエプソン株式会社 画像変形のための画像処理
CN102521806B (zh) * 2011-12-04 2014-10-22 山东大学 基于消失线的长方体表面透视变形的矫正方法
US8855375B2 (en) * 2012-01-12 2014-10-07 Kofax, Inc. Systems and methods for mobile image capture and processing

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6111996A (en) * 1998-03-13 2000-08-29 Northern Telecom Limited Optical multiplexer/demultiplexer
CN102222229A (zh) * 2011-07-28 2011-10-19 陈庆武 手指静脉图像预处理方法
CN103679638A (zh) * 2013-12-25 2014-03-26 广州广电运通金融电子股份有限公司 残缺或变形的四边形图像的校正方法

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106874818A (zh) * 2016-08-30 2017-06-20 阿里巴巴集团控股有限公司 一种数字对象唯一标识符doi识别方法与装置
WO2018040948A1 (zh) * 2016-08-30 2018-03-08 阿里巴巴集团控股有限公司 一种数字对象唯一标识符doi识别方法与装置
CN106874818B (zh) * 2016-08-30 2019-11-22 阿里巴巴集团控股有限公司 一种数字对象唯一标识符doi识别方法与装置
US10664674B2 (en) 2016-08-30 2020-05-26 Alibaba Group Holding Limited Digital object unique identifier (DOI) recognition method and device
CN113052896A (zh) * 2019-12-27 2021-06-29 大族激光科技产业集团股份有限公司 视觉定位方法及装置
CN112085708A (zh) * 2020-08-19 2020-12-15 浙江华睿科技有限公司 产品外轮廓中的直线边缘的缺陷检测方法及设备
CN112085708B (zh) * 2020-08-19 2023-07-07 浙江华睿科技股份有限公司 产品外轮廓中的直线边缘的缺陷检测方法及设备
CN112633275A (zh) * 2020-12-22 2021-04-09 航天信息股份有限公司 一种基于深度学习的多票据混拍图像校正方法及系统
CN112633275B (zh) * 2020-12-22 2023-07-18 航天信息股份有限公司 一种基于深度学习的多票据混拍图像校正方法及系统

Also Published As

Publication number Publication date
US9773299B2 (en) 2017-09-26
EP3089103A4 (en) 2017-01-11
AU2014373249A1 (en) 2016-06-30
ZA201604301B (en) 2017-08-30
EP3089103B1 (en) 2018-12-05
TR201819357T4 (tr) 2019-01-21
US20160314563A1 (en) 2016-10-27
AU2014373249B2 (en) 2017-08-31
EP3089103A1 (en) 2016-11-02
CL2016001485A1 (es) 2016-12-30
CN103679638A (zh) 2014-03-26

Similar Documents

Publication Publication Date Title
WO2015096535A1 (zh) 残缺或变形的四边形图像的校正方法
US8509536B2 (en) Character recognition device and method and computer-readable medium controlling the same
JP5844783B2 (ja) テキスト領域を含むグレースケール文書画像を処理する方法、グレースケール文書画像の少なくともテキスト領域を二値化する方法、グレースケール文書画像においてグリッドを形成するテーブルの抽出方法及びプログラム
CN106934803B (zh) 电子器件表面缺陷的检测方法及装置
CN105447512B (zh) 一种精粗结合的光学表面缺陷的检测方法及装置
US20120294528A1 (en) Method of Detecting and Correcting Digital Images of Books in the Book Spine Area
US8805117B2 (en) Methods for improving image search in large-scale databases
CN107045634B (zh) 一种基于最大稳定极值区域与笔画宽度的文本定位方法
CN112233116B (zh) 基于邻域决策与灰度共生矩阵描述的凹凸痕视觉检测方法
US20120093434A1 (en) Edge detection
US10586321B2 (en) Automatic detection, counting, and measurement of lumber boards using a handheld device
CN109632808A (zh) 棱边缺陷检测方法、装置、电子设备及存储介质
EP2536123B1 (en) Image processing method and image processing apparatus
Jipeng et al. Skew correction for Chinese character using Hough transform
Pan et al. An efficient method for skew correction of license plate
Magnier et al. Ridges and valleys detection in images using difference of rotating half smoothing filters
Hadi et al. A novel approach of skew estimation and correction in persian manuscript text using radon transform
CN107092909B (zh) 基于三角形相似定理的角度检测算法
Yang et al. Grid-based modelling and correction of arbitrarily warped historical document images for large-scale digitisation
CN109643451B (zh) 线检测方法
CN116228601B (zh) 一种火车双向平煤的平煤效果视觉监控方法
CN117495799A (zh) 一种由边缘成像引起的缺陷检测方法、装置及电子设备
Li et al. A novel method of straight-line extraction based on Wallis filtering for the close-range building
Ding et al. A correction algorithm for document images based on edge contour
CN114219951A (zh) 形状匹配方法、装置、计算机设备以及存储装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 14875817

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 15104334

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

REEP Request for entry into the european phase

Ref document number: 2014875817

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2014875817

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2014373249

Country of ref document: AU

Date of ref document: 20141013

Kind code of ref document: A