WO2018170937A1 - 遮蔽采集图像中异物的标记体、识别图像中异物标记体的方法以及书籍扫描方法 - Google Patents

遮蔽采集图像中异物的标记体、识别图像中异物标记体的方法以及书籍扫描方法 Download PDF

Info

Publication number
WO2018170937A1
WO2018170937A1 PCT/CN2017/078745 CN2017078745W WO2018170937A1 WO 2018170937 A1 WO2018170937 A1 WO 2018170937A1 CN 2017078745 W CN2017078745 W CN 2017078745W WO 2018170937 A1 WO2018170937 A1 WO 2018170937A1
Authority
WO
WIPO (PCT)
Prior art keywords
marker
image
straight line
foreign matter
length
Prior art date
Application number
PCT/CN2017/078745
Other languages
English (en)
French (fr)
Inventor
周康
Original Assignee
大连成者科技有限公司
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 大连成者科技有限公司 filed Critical 大连成者科技有限公司
Priority to US16/496,877 priority Critical patent/US10846549B2/en
Priority to EP17901740.5A priority patent/EP3605461A4/en
Publication of WO2018170937A1 publication Critical patent/WO2018170937A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
    • H04N1/46—Colour picture communication systems
    • H04N1/56—Processing of colour picture signals
    • H04N1/60—Colour correction or control
    • H04N1/62—Retouching, i.e. modification of isolated colours only or in isolated picture areas only
    • H04N1/626—Detection of non-electronic marks, e.g. fluorescent markers
    • 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
    • G06V10/225—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 based on a marking or identifier characterising the area
    • 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
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/107—Static hand or arm
    • G06V40/11—Hand-related biometrics; Hand pose recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/13—Type of disclosure document
    • G06V2201/131—Book

Definitions

  • the invention relates to a marker for masking a foreign object of a book page during an image acquisition type scanning process, an algorithm for identifying a marker body in an image, a book scanning method, and a corresponding image collection type book page turning scanning method.
  • G06 calculation; calculation; counting G06F electric digital data processing G06F9/00 program control device, for example, controller G06F9/06 application stored program, that is, the internal storage of the application processing device to receive the program and maintain the program G06F9/44 is used to implement specialized programs.
  • the scanner based on the video image acquisition collects the photo of the book page through the camera located above the scanned object, and the image is processed by the video algorithm, then the scanning can be completed, eliminating the traditional scanning method of manually pressing the printed matter to be scanned on the scanning surface. Brought a lot of work.
  • the skin color region is extracted using the elliptical skin color model to position the finger and remove the area near the skin color.
  • Elliptical skin color model transforms the skin color image from RGB space to YCrCb color space. In the two-dimensional space of CrCb, the sample area presents elliptical features, so people use a CrCb spatially approximated elliptical region as the basis for determining skin color.
  • the present invention is directed to the above problem, and a marker body for masking foreign objects in an image is developed, including:
  • a marking portion having a two-sided continuous pattern formed by at least one or more combination of primitives; a fixing portion that fixes the marking body to a foreign object appearing in the collection target, such as a finger of a book or a similar automatic flip book device
  • a foreign object appearing in the collection target such as a finger of a book or a similar automatic flip book device
  • the surface of the book turning mechanism appearing on the surface of the book page, so that the surface of the foreign object in the collected image is covered by the sign portion, which is convenient for algorithm identification and marking.
  • the primitive includes equal straight line segments parallel to each other and a quarter circle or a hollow circle.
  • the two-way continuous pattern is concentrated in a rectangular identification area located in the middle of the marker portion, and each of the isometric segments and the rectangular region is long.
  • the sides are vertical, and the elliptical focus (or the center point of the focal length) is connected to the long side of the rectangular area;
  • the edge variation gradient is enhanced, so that the primitive is more obvious under different light.
  • the color of the rectangular recognition region is the second.
  • each of the two consecutive patterns is composed of one
  • the picture element is composed; a plurality of two-party continuous patterns are parallel to the long sides of the rectangular area.
  • the multi-line two-sided continuous pattern includes at least a two-sided continuous pattern of the parallel straight line segments and a two-sided continuous pattern of four-circle/open circles.
  • a method for identifying a foreign object marker in an image includes the following steps:
  • a planar image including the marker is acquired; image preprocessing including at least binarization and denoising is performed.
  • edge detection is performed on the planar image to obtain an edge map in the planar image; and all contours in the edge map are extracted.
  • a certain number of candidate straight line segments are first determined by an algorithm, and the region is determined according to the position of each of the straight line segments, and a larger size is used to expand outward from the initial straight line segment position, and then The final area image is obtained by taking a circumscribed rectangle from each of the connected regions (possibly the overlapping regions caused by the overlap of the expansion of a plurality of straight segment regions).
  • a local edge map corresponding to the marker body is obtained, and the local contour in the local edge map is extracted, and the partial contour of all the ellipse is obtained as an candidate ellipse by screening the local contour; each ellipse is calculated The elliptical focus and length of the outline;
  • Verify the median length and angle of the straight line segments near each candidate circle (considering, sometimes the detected straight line segments on some non-finger-set patterns, such as books, have a higher probability of occurrence
  • the difference between the length and the angle of the straight line and the straight line segment in the marker body is relatively large. Calculating the average number is very likely to bring the error into it, so the median is used, by passing each of the straight line segments with the median of the angle and length. Compare, remove the optional line segments whose deviation exceeds the threshold range.
  • the parallel straight line segments are closer.
  • Center the pixel center C of the parallel straight line segment is selected as the center position of the marker body;
  • the average long axis length R of all the ellipse is calculated as an index for judging the distance of the marker from the lens;
  • the average length L of all the straight lines is calculated as a finger direction The basis for the downtilt size.
  • the image range of the marker is calculated based on the pixel center C, the average major axis length R, and the average length L.
  • the Canny edge detection is performed on the edge detection of the planar image; the straight line segment contour is eliminated by determining the contour enclosure area threshold range, the contour circumscribed minimum rectangle size, and the aspect ratio of the circumscribed rectangle.
  • the pixel path center of all the eligible straight line segments is calculated to have an alternative circle and an alternative straight line segment correspondence relationship screening step.
  • Each found line is compared to each candidate circle to find an associated line that satisfies the condition around each circle.
  • the farther the marker is from the lens the smaller the pixel distance between the point and the point; the greater the inclination of the marker, the smaller the pixel distance between the point and the point.
  • the distance between the near end of each line and the focus of each line satisfies the upper and lower threshold requirements, taking into account the height of the image of the book page image, the distance of the mark from the lens, the degree of tilt of the mark, and the resolution of the camera. Combined with the above factors, the following parameter thresholds are given: the lower limit is 8 pixels and the upper limit is 45 pixels. At the same time, the distance between the far end of each straight line and the focus meets the threshold requirement: the lower limit requirement: not less than the lower limit of 25 pixels, which can satisfy the conventional mainstream Resolution of 720p, 1080p and 2k or even 4k image acquisition accuracy and algorithm running speed.
  • the candidate straight line segment traverses the corresponding candidate circle before calculating all the final line segment pixel centers. If it passes, the candidate straight line segment is culled.
  • the threshold is used to eliminate the external interference circle. : Calculate whether the candidate circular focus deviates from the line of the candidate circular row before calculating the final line segment pixel center of all the eligible segments; if the deviation exceeds the threshold distance, the candidate circle is eliminated.
  • the associated line group is deleted and the corresponding candidate circle is deleted.
  • a book scanning method comprising the steps of: determining a range of images of a marker object for a two-dimensional image of a book page with a marker body masked; - expanding to an area of an approximate area above or below the marker body to The image range of the mark body is removed, and the mark body image of the current book page is removed, and the current book page is scanned.
  • FIG. 1 is a schematic diagram of a finger sleeve as a marker body according to an embodiment of the present invention
  • FIG. 2 is a schematic view showing a straight line pattern in an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of a quarter circle pattern in an embodiment of the present invention.
  • Figure 5 is a schematic diagram of a planar image of the present invention
  • Figure 6 is a schematic view of a partial image extracted by the present invention.
  • FIG. 7 is a diagram showing an image range mask of a marker body according to an embodiment of an algorithm of the present invention.
  • FIG. 9 is a detailed structural diagram of a mask in an embodiment of an algorithm of the present invention.
  • FIG. 10 is a schematic diagram of calculation of marker parameters in an algorithm embodiment of the present invention.
  • FIG. 11 is a schematic diagram of a scanned image after an algorithm removes a marker in an algorithm embodiment of the present invention
  • FIG. 12 is a schematic diagram of an application scenario of Embodiment 2 of an algorithm according to the present invention.
  • FIG. 13 is a schematic diagram of the recognition result in the algorithm embodiment 2 of the present invention.
  • FIG. 14 is a schematic diagram of an imaging principle in an algorithm embodiment of the present invention.
  • Figure 15 is a flow chart of the algorithm for scanning a book of the present invention
  • a finger sleeve solution for flipping a book with a finger is provided, and the fixing portion is a plastic/rubber finger sleeve whose circle is similar to a rotating body, and is convenient for the user to flip through the book.
  • the fixing portion is a plastic/rubber finger sleeve whose circle is similar to a rotating body, and is convenient for the user to flip through the book.
  • both plastic and paper pages have greater friction.
  • a rectangular identification area perpendicular to the central axis is provided, and two consecutive patterns of equal straight line segments parallel to each other are provided in the area, and two are formed by hollow circles. Two rows of continuous patterns are arranged, and two rows of hollow circles are alternately arranged. In the present embodiment, only a scheme of a circle which is a special case of an ellipse is considered.
  • the primitives ie, the parallel isometric segments and the open circles are white, the background color of the rectangular recognition region is white, the reverse color is black), and the finger sleeve is entirely yellow (as a preferred embodiment, Use other colors that differ from the paper quality of the book).
  • the inner surface of the finger/marker body is also provided with densely arranged rubber/plastic teeth, which cooperate with the elasticity of the rubber/plastic material to ensure a firm wearing without excessive pressure on the fingers.
  • the rubber tooth itself will deform and reduce the pressure on the finger.
  • the rubber/plastic tooth is placed along the finger in the direction of the finger sleeve, which is convenient for wearing and taking off the finger sleeve, and at the same time ensuring the firmness of the finger during the lateral movement of the book, especially when the hand points out the friction between the sweat and the rubber material. It may be less than the friction of the paper pages, causing the deflection of the finger cuffs.
  • Embodiment 1 is a book scanning application scenario. This embodiment mainly solves the influence of a finger on a book image during an image capture scanning process, as shown in FIG. 2-15.
  • the area enclosed by the contour must meet the upper and lower limits: the upper limit of the area is 10 pixels, and the upper limit of the area is 500 pixels;
  • the two end points of the corresponding straight line segment are obtained according to the circumscribed rectangle, and each straight line segment represented by two end points is stored as an alternative straight line segment.
  • the yellow line segment in this figure is the line that meets the criteria found in this step.
  • the area of the contour needs to meet the threshold range requirement: the lower limit is 200 pixels and the upper limit is 2500 pixels;
  • contours that meet the above two conditions are retained, and they are considered to be alternative circular patterns of the finger sleeve, and the elliptical focus, length and length of each contour are recorded.
  • the blue part of the figure is the alternative circle found.
  • the condition is that the distance between the near end of each line and the focus of each line satisfies the upper and lower limits: the lower limit is 8 pixels and the upper limit is 45 pixels; at the same time, the distance between the far end of each straight line and the focus of the focus meets the lower limit requirement: not less than the lower limit of 25 pixels ;
  • a) calculate the central straightness of all associated lines, that is, the average distance from the center of each line to the center line formed by the center of all associated lines;
  • the retained circle is the circle on the finalized finger sleeve, and the information of the associated line corresponding to the condition corresponding to each circle is also saved.
  • the typical mask range is composed of four parts: a rectangle (for limiting the width of the finger sleeve) and a middle portion of the finger.
  • the short axes of the three ellipses in the figure follow the direction of the finger A-B line, which corresponds to the length or length direction of the finger area.
  • the length of the short axes of the three ellipse depends on L: when the inclination of the finger is constant, and L is larger when the lens is closer to the lens, the length of the finger area is increased; when the distance between the finger and the lens is constant, the fingertip is inclined downward. The larger the L, the smaller the length of the finger area at this time.
  • the long axis of the three ellipse and the width of the rectangle are perpendicular to the AB line (ie, the direction of the central axis of the finger sleeve, the rubber/plastic finger sleeve is in a non-use state, and the cross section as a whole is a symmetrical image, similar to an ellipse).
  • This direction corresponds to the width of the finger area.
  • the length of the quantity is dependent on R, because R is the long axis of the ellipse, so it is independent of the tilt of the finger, only related to the distance of the finger from the lens. When the finger is closer to the lens, R becomes larger, and the finger width increases.
  • Rectangular centered on C, length 15L, width 9.5R; ellipse in the middle of the finger: centered on C, short axis radius 4.3L, long axis radius 7R; fingertip part ellipse: centered on A,
  • 3L, short axis radius 3L, long axis radius 3.5R; finger root ellipse: centered on B,
  • 5L, short axis radius 5L, long axis radius 6R.
  • the length of the corresponding parallel line in the image is proportional to the distance from the optical center of the camera; and the parallel of the same length in the physical object
  • the line, the corresponding length on the imaged image, is inversely proportional to its vertical distance from the optical center.
  • the marker body is used as a general identification application scenario for the differentiation of batch products.
  • a mark body is provided on the upper surface of the mouse.
  • the fixing portion of the marking body is preferably in the form of a sticker.
  • the marker body can be identified by employing the algorithm in Embodiment 1.
  • the marker recognition can be completed under the application scene of multiple color backgrounds, and the recognition accuracy is ensured.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Human Computer Interaction (AREA)
  • Image Analysis (AREA)

Abstract

一种遮蔽采集图像中异物的标记体，包括：标志部，表面具有至少由一种或者多种图元组合形成的二方连续图案；固定部，将标记体固定在采集目标中出现的异物，使得采集到的图像中异物表面被所述的标志部覆盖，便于算法识别和标记。一种识别图像中异物标记体的方法，包括如下步骤：对平面图像进行边缘检测，得到平面图像中的边缘图；提取该边缘图中的全部轮廓。通过算法确定一定数量的备选直线段，根据备选每条直线段的位置确定所述的区域，最后使用标记体上方或下方的近似面积的区域。一种书籍扫描方法，向所述标记体图像范围延伸，最终去除当前书页的标记体图像，完成当前书页的扫描。

Description

遮蔽采集图像中异物的标记体、识别图像中异物标记体的方法以及书籍扫描方法 技术领域
本发明涉及一种图像采集式扫描过程中遮蔽翻书页的异物的标记体，识别图像中标记体的算法以及书籍扫描方法，及相应的图像采集式的书籍翻页扫描方法。涉及专利分类号G06计算；推算；计数G06F电数字数据处理G06F9/00程序控制装置，例如，控制器G06F9/06应用存入的程序的，即应用处理设备的内部存储来接收程序并保持程序的G06F9/44用于执行专门程序的装置。
背景技术
基于视频图像采集的扫描仪通过位于扫描物上方的摄像头采集书页照片，通过视频是算法对图像进行处理后，即可完成扫描，免去了人工将待扫描的印刷品按压在扫描面的传统扫描方式带来的繁重工作量。
但是，此类扫描装置在扫描厚度较厚的书籍时或者在快速扫描的过程中需要手翻书页，并且在扫描过程中为了保持书页平整，需要用手/手指按压单侧或者双侧的书页，导致算法采集的书页图像不可避免的带有手指图像。
采用椭圆肤色模型提取肤色区域，从而定位手指，将肤色附近区域去除。椭圆肤色模型：将肤色图像从RGB空间变换到YCrCb颜色空间，在CrCb的二维空间上，样本区域呈现椭圆状特征，从而人们使用一个CrCb空间上近似的椭圆区域来作为判定肤色的依据。
但由于用于扫描的书籍内容多样，以及受光线变化的影响，单纯用颜色特征来识别手指区域很容易造成误检和漏检。
发明内容
本发明针对以上问题的提出，而研制的一种遮蔽采集图像中异物的标记体，包括：
标志部，表面具有至少由一种或者多种图元组合形成的二方连续图案；固定部，将标记体固定在采集目标中出现的异物，比如翻书的手指或者类似公开的自动翻书设备出现在书页表面的翻书机构的表面，使得采集到的图像中异物表面被所述的标志部覆盖，便于算法识别和标记。
为了便于算法识别，作为优选的实施方式，所述的图元包括相互平行的等长直线段以及四分圆或者空心圆。
为了便于算法定义标记体整体的图形范围，作为优选的实施方式，所述的二方连续图案集中于一位于标志部中部的矩形识别区域中，每个所述的等长线段与矩形区域的长边垂直，所述的椭圆焦点(或者焦距的中心点)连线与所述的矩形区域的长边平行；
更进一步的，为了加强图元与背景的对比度差异，增强边缘变化梯度，从而使图元在不同光线下都能较为明显，作为优选的实施方式，所述的矩形识别区域的颜色为所述二方连续图案中图元色彩的反色。
更进一步的，为了能够使的算进行精确的定位和识别(原理将在算法部分详述)，当所述的标志部包括多种图元时，每个所述的二方连续图案由一种图元组成；多个二方连续图案与矩形区域长边平行。
所述的多行二方连续图案中至少包括所述的平行直线段的二方连续图案和四分圆/空心圆的二方连续图案。
一种识别图像中异物标记体的方法，包括如下步骤：
首先，采集包括所述的标记体的平面图像；完成至少包括二值化和去噪的图像预处理。
然后，对平面图像进行边缘检测，得到平面图像中的边缘图；提取该边缘图中的全部轮廓。
通过对全部轮廓进行直线筛选，获得所述的标记体中的平行直线段的二方联系图案；
再后，在采集的平面图像中提取平行直线段所在的区域图像。
作为优选的实施方式，首先通过算法确定一定数量的备选直线段，根据备选每条直线段的位置确定所述的区域，采用一个较大的尺寸由初始的直线段位置向外扩张，然后将扩张后的每一个连通区域(可能是很多条直线段区域扩张的重叠导致的联通区域)取一个外接矩形这样的方法得到最终的区域图像。
通过边缘检测所述的区域图像，得到标记体对应的局部边缘图，提取该局部边缘图中的局部轮廓，通过对局部轮廓的筛选，得到全部椭圆的局部轮廓作为备选椭圆；计算每个椭圆轮廓的椭圆焦点和长短轴长度；
检验每个备选圆附近直线段的长度和角度的中位数(考虑到，有的时候检测到的旁边的某些非指套图案上的干扰直线段，比如书籍中较、大概率出现的直线与标志体中的直线段的长度和角度差异比较大，计算平均数，极有可能将误差带进去，故采用中位数)，通过将每条所述直线段与角度和长度中位数比较，去除偏差超出阈值范围的备选直线段。
分别考虑平行直线段和椭圆的像素中心的位置，选择更为接近标记体的图案的像素中心的二方连续图案，计算图案的像素中心，在后续描述的实施方式中，平行直线段更为接近中心，选择了平行直线段的像素中心C，作为标记体中心位置基准；计算所有椭圆的平均长轴长度R，作为判断标记体距离镜头远近的指标；计算所有直线的平均长度L，作为手指向下倾斜大小的依据。
根据所述的像素中心C、平均长轴长度R和平均长度L计算得出标记体的图像范围。
作为优选的实施方式，对平面图像进行边缘检测采用Canny边缘检测；通过判定轮廓包围面积阈值范围、轮廓外接最小矩形尺寸以及外接矩形的长宽比，剔除非直线段轮廓。
为了保证算法精度，作为优选的实施方式，计算最终所有符合条件的直线段像素中心之前还具有备选圆和备选直线段对应关系筛选步骤。
对每个找到的直线与每个备选圆进行比对，寻找每个圆周围满足条件的关联直线。通常标记体距离镜头越远，点与点间的图像像素距离越小；标记体倾斜程度越大，点与点间的图像像素距离越小。
每条直线相对焦点的近端与焦点的距离满足上下限阈值要求，综合考虑采集书页图像摄像头的高度、标记体距离镜头的距离以及标记体倾斜程度以及相机分辨率。结合上述因素，给出如下的参数阈值：下限8像素，上限45像素，同时，每条直线距焦点的远端与焦点的距离满足阈值要求：下限要求：不小于下限25像素，可以满足常规主流分辨率720p、1080p以及2k甚至4k的图像采集精度和算法运行速度。
按上述方法找到每个圆周围的关联直线，如果该圆周围找到的关联直线个数低于4条，则删除该备选圆。
作为优选的实施方式，计算最终所有符合条件的直线段像素中心之前判定备选直线段是否穿越对应的备选圆；若穿过，则剔除该备选直线段。
更进一步的，考虑到实际应用情况中，比如佩戴标记体的手指按压周围的 书籍内容中可能存在一定的圆(比如附图)，圆或者椭圆距离指套上的直线距离明显大于标记体中的椭圆/圆，故作为优选的实施方式，使用阈值来剔除掉这些外界干扰圆：计算最终所有符合条件的直线段像素中心之前，判定所述备选圆焦点是否偏离所在备选圆行的连线；若偏离超过阈值距离，则剔除该备选圆。
更进一步的，计算最终所有符合条件的直线段像素中心之前计算所有关联直线的中心直线度，即每条直线中心到所有关联直线的中心形成的中心线的平均距离；
若上述平均距离大于3像素，则关联直线组被删除，同时对应的备选圆被删除。
一种书籍扫描方法，包括如下步骤：—针对带有标记体遮蔽的标记体的书籍页面的二维图像，确定标记体的图像范围；—使用标记体上方或下方的近似面积的区域，扩展至所述的标记体的图像范围，去除完成当前书页的标记体图像，完成当前书页的扫描。
附图说明
为了更清楚的说明本发明的实施例或现有技术的技术方案，下面将对实施例或现有技术描述中所需要使用的附图做一简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例中作为标记体的指套示意图
图2为本发明实施例中直线段图案示意图
图3为本发明实施例中四分圆图案示意图
图4为本发明实施例中优选图案示意图
图5为本发明平面图像示意图
图6为本发明提取的局部图像示意图
图7为本发明算法实施例中标记体(指套)图像范围掩膜图
图8为本发明算法实施例中掩膜解析图
图9为本发明算法实施例中掩膜的细部结构图
图10为本发明算法实施例中标记体参数计算示意图
图11为本发明算法实施例中算法消除标记体后的扫描图像示意图
图12为本发明算法实施例2的应用场景示意图
图13为本发明算法实施例2中识别结果示意图
图14为本发明算法实施例中成像原理示意图
图15为本发明书籍扫描的算法流程图
具体实施方式
为使本发明的实施例的目的、技术方案和优点更加清楚，下面结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚完整的描述：
如图1所示，本实施例中给出的是针对采用手指翻书的指套方案，所述的固定部为圆近似于回转体的塑胶/橡胶指套，便于套接在使用者翻书手指上，同时塑胶与纸质书页有较大的摩擦力。
在如图1所示图形的中部，设有垂直于中轴线的矩形识别区域，在区域内设有自身相互平行的等长直线段够成的二方连续图案，以及由空心圆够成的两行二方连续图案，两行空心圆交错设置。在本实施例仅考虑作为椭圆的特例的圆的方案。
在颜色选择上，图元(即平行的等长线段和空心圆均为白色，矩形识别区域的背景颜色为白色的反色-黑色)，指套整体为黄色(作为优选的实施方式，也可采用与书籍纸质区别较大的其它颜色)。
为了方便佩戴，在指套/标记体的内表面还设有密集排列的橡胶/塑胶齿，与橡胶/塑胶材质的弹性共同作用，保证佩戴的牢固，同时不会对手指产生过大的压力而造成不适(当手指直径尺寸较大，橡胶齿自身会发生形变，减少对手指压力)。
同时，橡胶/塑胶齿沿手指进入指套方向设置，便于佩戴和脱下指套，同时保证手指横向运动翻书过程中的牢固性，尤其是在手指出汗后与橡胶材质之间的摩擦力可能小于纸质书页的摩擦力，导致指套偏转的情况发生。
实施例1，书籍扫描应用场景，本实施例主要解决图像采集式扫描过程中，手指等翻页过程中对书籍图像的影响，如图2-15所示：
(一)从图像中寻找合适大小的直线段：
1.将图像转为灰度图；
2.进行[5,5]大小的中值滤波去噪；
3.进行Canny边缘检测，梯度下限75，上限120，得到图像的边缘图；
4.从边缘图中提取轮廓；
5.分析每一个小轮廓是否符合指套直线的尺寸和形状要求，从而剔除非直线段的轮廓：
a)轮廓包围的面积需满足上下限要求：面积下限10像素，面积上限500像素；
b)求轮廓的最小外接矩形对应的宽和高，要求外接矩形的长边需要满足直线长度的上下限要求：长度下限：12像素，长度上限：70像素；
c)外接矩形的长宽比需大于3；
针对符合上述形状尺寸要求的轮廓，根据其外接矩形求得对应的直线段的两个端点，将每个用两个端点表示的直线段作为备选直线段进行存储。
6.对找到的备选直线进行重叠线的剔除：判断依据为两条直线如果其两个端点的距离都小于3像素，则认为它们重叠。
此图中的黄色线段为该步找到的符合条件的直线。
(二)根据直线段密集度来定位指套可能存在的局部区域
1.遍历每条直线段，计算得到该直线段的中心，斜率和长度；
2.将该直线段依次与其他每条直线段的特征相比较，看二者是否满足下列所有要求：
a)两条直线段的中心之间的距离满足上下限要求：下限4个像素，上限60像素；
b)两条直线段的斜率差异小于0.05；
c)两条直线段的长度差异小于其中任一条直线段长度的0.3倍。
如果某条直线段与该直线段的关系满足上面所有要求，则该直线段对应的相似直线段计数增加1.
3.对于相似直线累计数超过阈值5的区域进行提取，得到大概指套区域。
(三)在上述区域中寻找圆
1.将上面的局部定位图变换为灰度图；
2.对灰度图进行Canny边缘检测，梯度下限50，上限100，得到边缘图；
3.针对边缘图提取轮廓；
4.对每个轮廓进行分析，从而查找符合椭圆特征的轮廓：
a)轮廓的面积需满足阈值范围要求：下限200像素，上限2500像素；
b)根据轮廓的二维点集拟合椭圆，求拟合椭圆的面积与真实轮廓面积之差，二者面积差需满足：小于10像素；否则，认为该轮廓的椭圆度不足。
对符合上面两个条件的轮廓进行保留，认为它们是备选指套圆形图案，并记录每个轮廓的椭圆焦点、长短轴长度。
该图蓝色的部分为找到的备选圆。
(四)将圆与直线信息相结合，对备选圆和直线进行筛选
1.第一次筛选：
a)对每个找到的直线与每个备选圆进行比对，寻找每个圆周围满足条件的关联直线。条件是每条直线相对焦点的近端与焦点的距离满足上下限要求：下限8像素，上限45像素；同时，每条直线距焦点的远端与焦点的距离满足下限要求：不小于下限25像素；
b)按上述方法找到每个圆周围的关联直线，如果该圆周围找到的关联直线个数低于4条，则删除该备选圆；
c)将剩下满足条件的备选圆周围关联直线的角度、长度和中心存储起来。
2.第二次筛选：
a)计算每个备选圆附近直线的长度和角度的中位数，将每条直线的角度和长度与这两个中位数相比较，差异需满足上下限要求：每条直线的角度与中位角度差不能大于5度；每条线与中位长度的差异不能大于中位长度的5％；
b)不满足上下限要求的直线从关联直线队列中删除。
c)如果经过上步筛选，某个圆周围剩下的关联直线个数如果小于4条，则删除该备选圆。
3.第三次筛选：
a)每条直线的两个端点不能跨越与其关联的备选圆的两端，否则说明该直线穿越了该圆，而有直线穿越的圆将被删除。
4.第四次筛选：
a)计算备选圆的所有关联直线的中心点，利用这些中心点拟合出一条中心线；
b)计算每个备选圆的中心与上述中心线的距离，该距离不能小于该圆关联直线中位长度的0.8倍。不满足该条件的圆被删除。
5.第五次筛选：
a)计算所有关联直线的中心直线度，即每条直线中心到所有关联直线的中心形成的中心线的平均距离；
b)若上述平均距离大于3像素，则关联直线组被删除，同时对应的备选圆 被删除。
经上述共五轮筛选，被保留的圆即为最终确定的指套上的圆，同时也保存了每个圆对应的满足条件的关联直线的信息。
完成定位后，即可得出完整的指套图像掩膜，如图10所示，典型的掩膜范围是由如下四部分组成的：一个矩形(用于限制指套宽度)、手指中间部分的椭圆、手指指尖部分的椭圆以及手指指根部分的椭圆。
图中三个椭圆的短轴都顺着手指A-B线方向，该方向对应手指区域的长度或者长度方向。三个椭圆短轴的长度都依赖于L：当手指倾斜度一定，而距离镜头较近时，L变大，此时手指区域长度增加；当手指与镜头距离一定，手指指尖向下倾斜幅度越大，L越小，此时手指区域长度越短。
图示的三个椭圆的长轴以及矩形的宽都是垂直于A-B线(即指套的中心轴线方向，橡胶/塑胶指套在非使用状态下，横截面整体呈对称图像，类似椭圆)方向，这个方向对应手指区域的宽度。所述的量的长度都依赖于R，因为R因为是椭圆的长轴，因此与手指倾斜大小无关，仅与手指距离镜头的远近有关。当手指距离镜头较近时，R变大，此时手指宽度增加。
其中每个部分的尺寸和位置对应指套中心C、椭圆平均长轴长度R和直线平均长度L的关系如下：
矩形：以C为中心，长15L，宽9.5R；手指中间部分的椭圆：以C为中心，短轴半径4.3L，长轴半径7R；指尖部分的椭圆：以A为中心，|A-C|＝3L，短轴半径3L，长轴半径3.5R；指根部分的椭圆：以B为中心，|B-C|＝5L，短轴半径5L，长轴半径6R。
如图15所示：对于本发明所述的数理关系，一般的，在图像中相同长度的线，对应的实际平行直线的长度与其相对相机光心的距离成正比；而实物上相同长度的平行线，在成像图像上对应的长度与其相对光心的垂直距离成反比。
实施例2，标记体作为一般识别的应用场景，用于批量产品的区分。在鼠标上表面设置有标记体。在本实施例中，标记体的固定部优选为胶贴形式。通过采用实施例1中的算法，即可识别标记体。
由于标记体自身的图元特征以及反色特征，使得在多种颜色背景的应用场景下，都可完成标记体的识别，保证识别精度。
以上所述，仅为本发明较佳的具体实施方式，但本发明的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，根据本发明的技术方案及其发明构思加以等同替换或改变，都应涵盖在本发明的保护范围之内。

Claims (10)

  1. 一种遮蔽采集图像中异物的标记体，其特征在于包括：
    标志部，表面具有至少由一种或者多种图元组合形成的二方连续图案；
    固定部，将标记体固定在采集目标中出现的异物的表面，使得采集到的图像中异物表面被所述的标志部覆盖。
  2. 根据权利要求1所述的遮蔽采集图像中异物的标记体，其特征还在于所述的图元包括相互平行的等长直线段以及椭圆。
  3. 根据权利要求1或2所述的遮蔽采集图像中异物的标记体，其特征还在于所述的二方连续图案集中于一位于标志部中部的矩形识别区域中，每个所述的等长线段与矩形区域的长边垂直，所述的椭圆焦点或者椭圆焦距的中点连线与所述的矩形区域的长边平行；所述的矩形识别区域的颜色为所述二方连续图案中图元色彩的反色。
  4. 根据权利要求1-3任意一项权利要求所述的遮蔽采集图像中异物的标记体，其特征还在于：当所述的标志部包括多种图元时，每个所述的二方连续图案由一种图元组成；多个二方连续图案与矩形区域长边平行。
  5. 根据权利要求4所述的所述的遮蔽采集图像中异物的标记体，其特征还在于所述的多行二方连续图案中至少包括所述的平行直线段的二方连续图案和四分圆/空心圆的二方连续图案。
  6. 一种识别图像中异物标记体的方法，其特征在于包括如下步骤：
    —采集包括如权利要求5所述的标记体的平面二维图像；完成至少包括二值化和去噪的图像预处理；
    —对平面图像进行边缘检测，得到平面图像中的边缘图；提取该边缘图中的全部轮廓；
    —通过对全部轮廓进行直线筛选，获得所述备选平行直线段；
    —在采集的平面图像中提取平行直线段所在的区域图像；
    —通过二值化和边缘检测所述的区域图像，得到局部边缘图，提取该局部边缘图中的局部轮廓，通过对局部轮廓的筛选，得到全部椭圆的局部轮廓作为备选圆；计算每个椭圆轮廓的椭圆焦点和长短轴长度；
    —检验每个备选圆附近直线段的长度和角度的中位数，通过将每条所述直线段与角度和长度中位数比较，去除偏差超出阈值范围的备选直线段；
    —计算最终所有符合条件的直线段的像素中心C，作为标记体中心位置基准；计算所有椭圆的平均长轴长度R，作为判断标记体距离镜头远近的指标；计算所有直线的平均长度L，作为手指向下倾斜大小的依据；
    —根据所述的像素中心C、平均长轴长度R和平均长度L计算得出标记体的图像范围。
  7. 根据权利要求6所述的识别图像中异物标记体的方法，其特征还在于对平面图像进行边缘检测采用Canny边缘检测；通过判定轮廓包围面积阈值范围、轮廓外接最小矩形尺寸以及外接矩形的长宽比，剔除非直线段轮廓。
  8. 根据权利要求6所述的识别图像中异物标记体的方法，其特征还在于计算最终所有符合条件的直线段像素中心之前判定备选直线段是否穿越对应的备选圆；若穿过，则剔除该备选直线段；
    计算最终所有符合条件的直线段像素中心之前，判定所述备选圆焦点是否偏离所在备选圆行的连线；若偏离超过阈值距离，则剔除该备选圆。
  9. 根据权利要求6所述的识别图像中异物标记体的方法，其特征还在于计算最终所有符合条件的直线段像素中心之前计算所有关联直线的中心直线度，即每条直线中心到所有关联直线的中心形成的中心线的平均距离；
    若上述平均距离大于3像素，则关联直线组被删除，同时对应的备选圆被删除。
  10. 一种书籍扫描方法，其特征在于包括如下步骤：
    —针对带有标记体遮蔽的标记体的书籍页面的二维图像，采用如权利要求6-9任意一项权利要求所述的方法，确定标记体的图像范围；
    —使用标记体上方或下方的近似面积的区域，扩展至所述的标记体的图像范围，去除完成当前书页的标记体图像，完成当前书页的扫描。
PCT/CN2017/078745 2017-03-24 2017-03-30 遮蔽采集图像中异物的标记体、识别图像中异物标记体的方法以及书籍扫描方法 WO2018170937A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US16/496,877 US10846549B2 (en) 2017-03-24 2017-03-30 Marker for occluding foreign matter in acquired image, method for recognizing foreign matter marker in image and book scanning method
EP17901740.5A EP3605461A4 (en) 2017-03-24 2017-03-30 MARKER FOR FOREIGN BODY OCCLUSION IN AN ACQUIRED IMAGE, PROCESS FOR RECOGNIZING A FOREIGN BODY MARKER IN AN IMAGE AND PROCESS FOR BOOK SCAN

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710182739.2A CN107085850B (zh) 2017-03-24 2017-03-24 遮蔽采集图像中异物的标记体、识别图像中异物标记体的方法以及书籍扫描方法
CN201710182739.2 2017-03-24

Publications (1)

Publication Number Publication Date
WO2018170937A1 true WO2018170937A1 (zh) 2018-09-27

Family

ID=59614221

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/078745 WO2018170937A1 (zh) 2017-03-24 2017-03-30 遮蔽采集图像中异物的标记体、识别图像中异物标记体的方法以及书籍扫描方法

Country Status (4)

Country Link
US (1) US10846549B2 (zh)
EP (1) EP3605461A4 (zh)
CN (1) CN107085850B (zh)
WO (1) WO2018170937A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114612496A (zh) * 2022-03-18 2022-06-10 易思维（杭州）科技有限公司 一种直线段轮廓提取方法

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110910393B (zh) * 2018-09-18 2023-03-24 北京市商汤科技开发有限公司 数据处理方法及装置、电子设备及存储介质
CN109712132B (zh) * 2018-12-28 2021-03-19 中国科学院长春光学精密机械与物理研究所 一种使用电子器件计数装置自动筛选计数的方法
US20200320334A1 (en) * 2019-04-05 2020-10-08 Ent. Services Development Corporation Lp Systems and methods for digital image-based object authentication
CN111079494B (zh) * 2019-06-09 2023-08-25 广东小天才科技有限公司 一种学习内容推送方法及电子设备
CN111814686A (zh) * 2020-07-09 2020-10-23 国网山西省电力公司吕梁供电公司 一种基于视觉的输电线路识别及异物入侵在线检测方法
US20220327319A1 (en) * 2021-04-09 2022-10-13 Veoneer US LCC Tagging objects with fault override patterns during calibration of vehicle sensing systems
CN113361643A (zh) * 2021-07-02 2021-09-07 人民中科(济南)智能技术有限公司 基于深度学习的通用标志识别方法、系统、设备及存储介质
CN115147407B (zh) * 2022-08-29 2022-11-18 聊城市博源节能科技有限公司 基于计算机视觉的轴承质量检测方法
CN117066702B (zh) * 2023-08-25 2024-04-19 上海频准激光科技有限公司 一种基于激光器的激光打标控制系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW201023622A (en) * 2008-12-09 2010-06-16 Qisda Corp Scanning method and auxiliary apparatus of auto-executing appending scanning function
CN101915573A (zh) * 2010-08-04 2010-12-15 中国科学院自动化研究所 一种基于标记物的关键点检测的定位测量方法
US20130272602A1 (en) * 2012-04-16 2013-10-17 Fujitsu Limited Method and apparatus for processing scanned image
CN106127203A (zh) * 2016-06-29 2016-11-16 孟祥雨 一种对目标物定位和追踪的装置及图像识别的方法
CN106412570A (zh) * 2016-10-11 2017-02-15 惠州市德赛西威汽车电子股份有限公司 一种摄像系统去噪能力的测定方法及测定工具

Family Cites Families (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5065257A (en) * 1985-03-20 1991-11-12 Canon Kabushiki Kaisha Image processing apparatus with means for compensating for dust., etc., on an original
JP3436025B2 (ja) * 1995-12-27 2003-08-11 ミノルタ株式会社 読取り画像の修正方法及び画像読取り装置
US20090254994A1 (en) * 2002-02-18 2009-10-08 David Lynch Waterson Security methods and systems
JP2004357190A (ja) * 2003-05-30 2004-12-16 Ricoh Co Ltd 画像読取装置
US10635723B2 (en) * 2004-02-15 2020-04-28 Google Llc Search engines and systems with handheld document data capture devices
JP4385903B2 (ja) * 2004-09-21 2009-12-16 ブラザー工業株式会社 画像形成装置
JP2008227889A (ja) * 2007-03-13 2008-09-25 Ricoh Co Ltd 画像読取装置およびその制御方法
CN101295149B (zh) * 2007-04-24 2011-03-02 京瓷美达株式会社 图像形成装置
US8018631B2 (en) * 2007-07-04 2011-09-13 Ricoh Company, Limited Original reading device and image forming apparatus
JP2010283789A (ja) * 2009-06-08 2010-12-16 Canon Inc 画像読取装置
CN102034445B (zh) * 2009-09-28 2013-01-09 群康科技(深圳)有限公司 液晶显示器供电电路
KR101692407B1 (ko) * 2010-08-19 2017-01-04 삼성전자주식회사 라인 패턴 구조물의 형성 방법
CN102646193B (zh) * 2012-03-23 2013-09-25 河海大学 一种环形排布字符图像分割方法
US9726883B2 (en) * 2012-07-27 2017-08-08 Nissan Motor Co., Ltd Three-dimensional object detection device and foreign matter detection device
CA2849557A1 (en) * 2013-04-22 2014-10-22 Pierre Hamel Method and system using two parallel optical character recognition processes
JP6099686B2 (ja) * 2015-03-20 2017-03-22 株式会社Pfu 画像処理装置、領域検出方法及びコンピュータプログラム
CN106199750A (zh) * 2015-05-04 2016-12-07 中国科学院上海高等研究院 机场飞行区活动异物检测系统及其检测方法
CN105872291B (zh) * 2016-05-31 2018-11-27 大连成者科技有限公司 带有激光校正的智能互联网高清扫描仪
CN107291993A (zh) * 2017-05-27 2017-10-24 中国石油大学(华东) 一种多孔介质中预交联凝胶悬浮液微观流动的模拟方法
US10291805B1 (en) * 2018-03-20 2019-05-14 Kabushiki Kaisha Toshiba Image processing apparatus
WO2020076472A2 (en) * 2018-09-19 2020-04-16 Sana Rezgui Human-robots: the new specie

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW201023622A (en) * 2008-12-09 2010-06-16 Qisda Corp Scanning method and auxiliary apparatus of auto-executing appending scanning function
CN101915573A (zh) * 2010-08-04 2010-12-15 中国科学院自动化研究所 一种基于标记物的关键点检测的定位测量方法
US20130272602A1 (en) * 2012-04-16 2013-10-17 Fujitsu Limited Method and apparatus for processing scanned image
CN106127203A (zh) * 2016-06-29 2016-11-16 孟祥雨 一种对目标物定位和追踪的装置及图像识别的方法
CN106412570A (zh) * 2016-10-11 2017-02-15 惠州市德赛西威汽车电子股份有限公司 一种摄像系统去噪能力的测定方法及测定工具

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3605461A4 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114612496A (zh) * 2022-03-18 2022-06-10 易思维（杭州）科技有限公司 一种直线段轮廓提取方法
CN114612496B (zh) * 2022-03-18 2024-05-31 易思维(杭州)科技股份有限公司 一种直线段轮廓提取方法

Also Published As

Publication number Publication date
EP3605461A4 (en) 2021-03-24
EP3605461A1 (en) 2020-02-05
US20200104621A1 (en) 2020-04-02
CN107085850B (zh) 2019-10-01
US10846549B2 (en) 2020-11-24
CN107085850A (zh) 2017-08-22

Similar Documents

Publication Publication Date Title
WO2018170937A1 (zh) 遮蔽采集图像中异物的标记体、识别图像中异物标记体的方法以及书籍扫描方法
CN112508826B (zh) 一种印刷品缺陷检测方法
CN111243032B (zh) 一种棋盘格角点全自动检测方法
CN107491730A (zh) 一种基于图像处理的化验单识别方法
CN103914827B (zh) 汽车密封条轮廓缺陷的视觉检测方法
KR101809543B1 (ko) 비접촉식 지문 인식하는 방법 및 이를 수행하기 위한 전자 기기
CN107358628B (zh) 基于靶标的线阵图像处理方法
CN107392849B (zh) 基于图像细分的靶标识别与定位方法
CN106033535A (zh) 电子阅卷方法
CN110866932A (zh) 多通道舌体边缘检测装置、方法及存储介质
JP2007272435A (ja) 顔特徴抽出装置及び顔特徴抽出方法
CN106529531A (zh) 一种基于图像处理的中国象棋识别系统及方法
WO2018032630A1 (zh) 一种教育玩具套件及利用颜色和轮廓识别编程模块的方法
CA2189671C (en) Apparatus for extracting fingerprint features
JP5278984B2 (ja) 画像解析装置及び画像解析プログラム
US11216905B2 (en) Automatic detection, counting, and measurement of lumber boards using a handheld device
WO2021195873A1 (zh) 识别sfr测试卡图像中感兴趣区域的方法及装置、介质
CN114863492A (zh) 一种低质量指纹图像的修复方法及修复装置
CN105229665B (zh) 对基于图像的蛇形带子磨损评估的增强分析
CN109472257B (zh) 一种字符版面确定方法及装置
JP4293653B2 (ja) 外観検査方法
Deng et al. A new measuring method of wool fiber diameter based on image processing
KR100194583B1 (ko) 얼굴 구성요소 추출시스템 및 그 추출방법
CN110543802A (zh) 眼底图像中左右眼识别方法与装置
CN114155296A (zh) 一种确定手掌图像中心区域的方法

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: 17901740

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

WWE Wipo information: entry into national phase

Ref document number: 2017901740

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2017901740

Country of ref document: EP

Effective date: 20191024