CN114140620A - Object straight line contour detection method - Google Patents
Object straight line contour detection method Download PDFInfo
- Publication number
- CN114140620A CN114140620A CN202111314707.6A CN202111314707A CN114140620A CN 114140620 A CN114140620 A CN 114140620A CN 202111314707 A CN202111314707 A CN 202111314707A CN 114140620 A CN114140620 A CN 114140620A
- Authority
- CN
- China
- Prior art keywords
- scale
- image
- pixel
- point
- points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for detecting a linear contour of an object. The invention comprises the following steps: obtaining a card image and preprocessing the card image to obtain a card image to be detected; obtaining a corrected multi-scale sampling image through a multi-scale smoothing module and a preset smoothing module; processing the obtained continuous characteristic pixel points by using a nearest neighbor operator; and combining the reduction module and the binarization to carry out scale combination to obtain a linear contour map of the card image to be detected, and thus obtaining the linear contour of the object. The frame of the card is detected by the image edge detection method, so that the running time and the calculated amount are reduced, the condition of misjudgment under the condition of the existence of irrelevant complex textures, patterns or noise is avoided, the robustness and the precision of the detection of the linear contour of the object are improved, and convenience is provided for the subsequent processing of the extraction of object information and the like.
Description
Technology neighborhood
The invention relates to an image processing application neighborhood, in particular to a method for detecting a linear contour of an object.
Background
Under the current era background of rapid development of the internet and digitalization, objects with complex surface characteristics are often required to be identified in monitoring, automatic driving and industrial production lines, information input of identity cards, bank cards, student cards and the like is required to be carried out by users in a manual input or card identification mode in the industries of information borrowing of social software, third-party payment platforms, leasing platforms and the like, and card identification is more popular with society in a convenient and efficient manner. Most of the objects such as lanes, identification cards and products have a large amount of straight line features, and object recognition is usually performed on the basis of extracting a straight line profile of the object.
The accuracy of the object recognition area is one of the key factors affecting the accuracy of object recognition. The existing line or card contour detection method is based on a Hough transform algorithm, a line equation intersection, a Freeman chain code or a DeepLab v3 network and the like, but the time complexity and the space complexity of methods involving space transformation or artificial intelligence neighborhood, such as the Hough transform algorithm and network learning, are higher, the time influence experience of object identification can be increased, the interference resistance of methods involving the line equation intersection, the Freeman chain code and the like is poorer, the method is difficult to process when rich colors and textures exist at the edge of an object, and the influence experience of object identification precision can be reduced.
Disclosure of Invention
The invention aims to provide an object straight line contour detection method, which solves the problem that robust and accurate object contour detection is difficult to obtain on the basis of low time complexity and low space complexity in the prior art, and the object straight line contour obtained by the object straight line contour detection of the method can be fit with the real situation.
The first aspect of the embodiments of the present invention provides a method for detecting a linear profile of an object, taking card linear profile detection as an example, specifically comprising the following steps:
the technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step S10, obtaining an image of the card to be detected: acquiring a card image by using an image storage module or an embedded image sensor, and preprocessing the acquired card image to obtain a card image to be detected;
step S20, multi-scale smoothing: performing K-order unit segmentation on the card image to be detected to obtain K2Multiple multi-scale sampling subgraphs, and sequentially adding links to each scale sampling subgraph according to the sequence of the segmentation sub-stepsContinuing indexing; then the multi-scale sampling subgraph is transmitted to a preset smoothing module, irrelevant information is removed from the preset smoothing module, and a modified multi-scale sampling subgraph is obtained;
step S30, nearest neighbor operator processing: extracting pixel points and nearest neighbor pixel points in the corrected multi-scale sampling subgraph to establish separable lattice sequences, extracting head and tail two-point characteristics and nearest neighbor sequence characteristics of the separable lattice sequences, and obtaining continuous characteristic pixel points according to the point characteristics and the sequence characteristics;
step S40, binarization processing: marking the position of a continuous characteristic pixel point in the modified multi-scale sampling subgraph as 1, marking the position of the rest data points in the modified multi-scale sampling subgraph as 0, and obtaining a black-white graph of the continuous characteristic pixel point according to the binarization mark;
step S50, scale merging: establishing a linear contour map of each scale according to the black-and-white map of the continuous characteristic pixel points, and transmitting the linear contour map of each scale to a deduction module to obtain a linear contour map of the card image to be detected;
step S60, positioning the card straight line contour: the linear profile graph of the card image to be detected is used as a positioning range, and the card image is subjected to profile calibration to obtain a card identification area, so that the card identification precision is improved.
Further, step S20 is specifically implemented as follows:
in step S21, after the card image to be tested is obtained, K-order unit segmentation of the card image to be tested is first performed, where K is { a {1,a2,...,an-1,anAnd | n is less than 10}, sequentially extracting sampling subgraphs of each scale from left to right and from top to bottom from the card image to be detected, and extracting the pixel point set of the sampling subgraphs of each scale to be phikThe mathematical expression form of the pixel point set of each scale sampling subgraph is phik＝{pij,p(i+k)j,pi(j+k),p(i+k)(j+k),...,p(i+nk)(j+mk)|m,n≤PsizeK }; wherein p isijRepresenting pixel points, i and j represent relative coordinates of the pixel points in the multi-scale sampling subgraph, k represents offset of the pixel points, and PsizeRepresenting the size of the sub-graph of each scale sampling, K tableScale displaying; dividing the card image to be tested into K2After a plurality of multi-scale sampling subgraphs, sequentially adding continuous indexes F to each scale sampling subgraph according to the sequence of the segmentation sub-steps, wherein the value range of the indexes is that F is more than or equal to 1 and less than or equal to K2，
Step S22, importing the multi-scale sampling sub-images into a preset smoothing module, and sequentially setting the step length S ═ a for pixel points of each multi-scale sampling sub-image1,a2,...,an-1,anL n is less than K, and eight neighborhoods of the pixel points are extracted according to the step length;
step S23, calculating the difference value between the central point and the eight neighborhoods, and replacing the central point pixel with the pixel value of the point with the difference value from the central point ranked as n;
calculating a central point C and eight neighborhood pixels AiWherein i is 1,2,. 8; scanning eight neighborhoods of the pixel points extracted in the step S22 according to the Zigzag path, and connecting the center point C with the neighborhood pixel point AiIn order from small to large such that | C-A1|≤|C-A2|≤|C-A3|≤...≤|C-A8And use of A |)2Replaces the value of the center point C;
step S24, carrying out median filtering processing on the sampling subgraphs of all scales after the center point value replacement; the median filtering adopts self-adaptive median filtering, the initial length of the length L of the rectangular area block is set as Min _ L, the maximum length is set as Max _ L, the initial width of the width W of the rectangular area block is set as Min _ W, and the maximum width is set as Max _ W; the method comprises the following steps that a rectangular area block scans and corrects a multi-scale sampling sub-image according to a Zigzag path, the mean value M, the maximum value Max _ P and the minimum value Min _ P of pixel points covered in the rectangular area block are calculated, when the rectangular area block increasing condition is met, the matrix area block calculates the mean value again until the rectangular area block increasing condition is not met, the mean value of the pixel points replaces pixels at the center point of the rectangular area block, and the rectangular area block increasing condition is as follows: m is less than or equal to Min _ p, M is more than or equal to Max _ p, and L is less than or equal to Max _ L, Lambda W is less than or equal to Max _ W.
Further, the step length refers to the interval range of the central point selected at intervals in two directions from left to right and from top to bottom, and is divided into three areas A, B and C according to the position relationship of each point; the region A is the vertex of each scale sampling subgraph and is characterized in that eight neighborhoods of pixel points in the region have only 3 effective values; the B region is a four-side region of each scale sampling subgraph and is characterized in that eight neighborhoods of pixel points in the region have no value in a certain row or column; the C region is a non-image edge region of each scale sampling sub-image and is characterized in that eight neighborhoods of pixel points in the region are in a full-value state; and (4) carrying out zero value supplement on positions lacking effective values in the areas A and B, and extracting eight neighborhoods of the pixel points from the sampling subgraphs of all scales after the value supplement according to the step length.
Further, step S30 is specifically implemented as follows:
the nearest neighbor pixel point represents that the pixel point and the central pixel point have the nearest pixel point of the pixel point in the N x N neighborhood in the connecting direction; extracting a pixel point P and a nearest neighbor pixel point P1 in the modified multi-scale sampling subgraph, when two pixel points P and P1 are in separable lattices, adding separable lattice sequences into the two pixel points P and P1, continuously extracting a nearest neighbor pixel point P2 of the nearest neighbor pixel point P1 for iteration, namely judging whether the pixel point P1 and the nearest neighbor pixel point P2 are in the separable lattices, if so, adding the two pixel points P1 and P2 into the separable lattice sequences, and continuously repeating the extraction and judgment operations; and when the nearest neighbor pixel point exists as the image boundary or the pixel point exists in the lattice subsequence, finishing a separable lattice sequence.
Further, the separable grid is described as follows:
firstly, sorting the difference values of every two pixels of 2 x 2 pixel points from small to large, and then removing the difference value of diagonal pixels, so that the difference between the difference value of the 2 nd large pixel and the difference value of the 3 rd large pixel is greater than a set threshold value h1The 2 x 2 pixel points are separable lattices; the head and tail two-point characteristic of the separable lattice sequence is the angle between the connection line of the head and tail two points of the sequence and the actual connection line of the two points, and the nearest neighbor sequence characteristic is the steering condition of the sequence in the process from head to tail; the steering condition is the turning condition of the sequence in eight neighborhood directions of 'up, down, left, right, upper left, lower left, upper right and lower right'; according to the characteristics of the head and the tail points and the characteristics of the nearest neighbor sequence, the angle is smaller than a set threshold value h2And the sequence without turning is used as continuous characteristic pixel pointsThe points have scale attributes and comprise pixel points with various scales meeting requirements.
Further, step S50 is to establish a linear profile of each scale according to the black-and-white image of the continuous characteristic pixel points, obtain a linear profile of the card image to be detected using the branch reduction module, establish a linear profile of each scale, and determine the relative coordinates of the pixel points in the black-and-white image of the continuous characteristic pixel points and the step length S of each scale as { a ═ a {1,a2,...,an-1,anI n < K }, sequentially arranging from top to bottom and from left to right to obtain linear contour maps of all scales, and transmitting the linear contour maps of all scales to a deduction module to obtain a linear contour map of the card image to be detected, wherein the specific implementation is as follows:
step S51, marking the main direction of any pixel point in the linear contour graph of each scale under the adjacent m points; selecting any pixel point along the directions from left to right and from top to bottom, defining a point containing a straight line outline in each scale straight line outline graph as a straight line outline point, selecting a pixel point of the first straight line outline point at the upper left as a starting point, taking a straight line outline point existing in a rectangular block of m × m with the pixel point as the center as a point in the adjacent direction, directly marking more directions of the straight line outline points as main directions if the straight line outline points are in one adjacent direction or two adjacent directions, and marking the straight line outline points as any directions if the straight line outline points are more than two adjacent directions;
step S52, obtaining a continuous line segment set by using depth-first search in the recording direction for the straight line profile graph of each scale; substituting the principal directions of any pixel points under adjacent m points in the linear contour map of each scale obtained in the step S51 into depth-first search, setting the principal direction with the most linear contour points in the depth-first search process as a marked principal direction, covering the principal direction in which the linear contour points with any direction are found in the depth-first search process by using the marked principal direction, and extracting the principal direction ratio with the most linear contour points in the priority search process to reach a threshold value h of each scale3Obtaining a continuous line segment set by the straight line contour points;
step S53, extracting the threshold h larger than each scale according to the length of the continuous line segment set of the straight line contour graph of each scale4Each of the continuous line segments ofSorting the lengths of the continuous line segment sets of the scale straight line profile graph, and keeping the length larger than the threshold value h of each scale4The pixel value of the pixel point of the continuous line segment set is not changed, and the length is less than or equal to the threshold value h of each scale4Removing pixel values of pixel points of the continuous line segment set;
and S54, interpolating and merging the continuous line segments of all scales according to the sizes of the scales to obtain a linear contour map of the card image to be detected, determining the difference priority of the corresponding scale image according to the scale K, and extracting linear contour points from the continuous line segments of all scales according to the Zigzag path and the priorities of all scales to obtain the linear contour map of the card image to be detected.
Furthermore, the device corresponding to the method comprises an object image multi-scale sampling unit (201), a continuous characteristic pixel point extracting unit (202), a multi-scale merging unit (203) and a linear contour detecting unit (204);
the object image multi-scale sampling unit (201) is used for acquiring an object image, graying and standardizing the object image to obtain an object image to be detected, performing multi-scale smoothing on the object image to be detected, and removing irrelevant information, imaging equipment system errors and image compression noise by using a preset smoothing module to obtain a modified multi-scale sampling sub-image;
the continuous characteristic pixel point extraction unit (202) is used for processing nearest neighbor operators of the obtained modified multi-scale sampling subgraph, extracting pixel points in the modified multi-scale sampling subgraph and nearest neighbor pixel points to establish separable lattice sequences, obtaining sequence head and tail two-point characteristics and nearest neighbor sequence characteristics according to the separable lattice sequences, and obtaining continuous characteristic pixel points according to the point characteristics and the sequence characteristics;
the multi-scale merging unit (203) is used for acquiring a linear profile of an image of an object to be detected and obtaining black and white images of continuous characteristic pixel points according to the continuous characteristic pixel points; establishing a linear profile graph of each scale based on the black-and-white graph of the continuous characteristic pixel points, and then obtaining the linear profile graph of the object image to be detected by using a deduction module;
the linear contour detection unit (204) is used for positioning the linear contour of the object, taking the linear contour of the image of the object to be detected as a positioning range, carrying out contour calibration on the image of the object to obtain an object identification area, and improving the accuracy of object identification.
Further, to achieve the above object, a third aspect of the present invention provides an object straight line contour detection apparatus, which includes an image sensor, a processor, an external interface, and a preset smoothing module and a deduction module stored in the memory and executable on the processor. The processor executes a preset smoothing module, a branch reducing module and any one of the steps of the object straight line contour detection method.
Compared with the prior art, the invention has the following beneficial effects:
according to the technical scheme, the object straight line contour detection method comprises the steps of carrying out multi-scale sampling on an object image to be detected, obtaining a corrected multi-scale sampling sub-image by using a preset smoothing module, obtaining continuous characteristic pixel points by using nearest neighbor operator processing, obtaining a black-and-white image of the continuous characteristic pixel points, and carrying out scale combination by using a deduction module to obtain a straight line contour image of the object image to be detected.
Compared with the prior art, the method does not need to carry out training and learning on the product image and the conversion of the image characteristics, and the multi-scale segmentation is still the characteristics of the original image, so that the running time and the calculation amount can be reduced; in addition, the method does not need to establish a linear equation and a linear fitting equation, only needs to combine screening data of all scales based on the modified multi-scale sampling subgraph as a data source of linear contour detection, avoids interference of image irrelevant information and noise on the linear equation and the sum of fitting errors, and therefore can improve the robustness and accuracy of object contour detection.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed in the embodiments or the prior art description will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without inventive labor.
Fig. 1 is a schematic hardware structure diagram of an object straight line contour detection apparatus provided by an embodiment of the present invention;
fig. 2 is a schematic flow chart of a method for detecting a straight line contour of an object according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an object straight line contour detection apparatus according to an embodiment of the present invention;
FIG. 4 is a schematic flow chart of a default smoothing module according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of an embodiment of a predetermined smoothing module result provided by the present invention;
FIG. 6 is a schematic flow diagram of a branch reduction module provided by an embodiment of the present invention;
FIG. 7 is a schematic diagram of one embodiment of the flow results of the offload module of the present invention;
fig. 8 is a schematic diagram of an embodiment of a detection result of a straight line contour of an object according to the present invention.
Detailed Description
For a better understanding of the objects, aspects and advantages of the present invention, those skilled in the art will appreciate from the following detailed description that is to be read in connection with the accompanying drawings and the detailed description. It is to be understood that the described embodiments are illustrative of the relevant invention only and are not limiting of the invention. It is also noted that for the convenience of description, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
The object straight line contour detection method related by the embodiment of the invention is mainly applied to object straight line contour detection equipment or equipment, and the object straight line contour detection equipment or equipment can be equipment with a data processing function, such as a single chip microcomputer, a mobile phone, a computer, a server and the like. The frame of the card is detected by the image edge detection method, so that the running time and the calculated amount are reduced, the condition of misjudgment under the condition of the existence of irrelevant complex textures, patterns or noise is avoided, the robustness and the precision of the detection of the linear contour of the object are improved, and convenience is provided for the subsequent processing of the extraction of object information and the like.
Referring to fig. 1, a schematic diagram of a hardware structure of an object straight line contour detection apparatus according to an embodiment of the present invention is provided. In the embodiment of the present invention, the object straight line contour detection apparatus includes an image sensor 101, a memory 102, a processor 103, and an external interface 104.
The image sensor 101 is used to acquire a card image, and a Charge Coupled Device (CCD) or a Complementary Metal-Oxide Semiconductor (CMOS) may be used, so that image sensors of a consumer mobile phone, a computer, an industrial scanner, and the like are included in the coverage of the sensor 101.
The Memory 102 is used for storing data of the object straight line contour detection process, the preset smoothing module, the branch reduction module and other functional programs, and a Random Access Memory (RAM) or a Direct Access Memory (DAM) can be used, so that the Memory elements of the digital camera, the mobile phone, the computer and the like are all contained in the Memory 102. The memory 102 may also be independent of the processor 103, suitable for embedded development. The preset smoothing module in the memory 102 is shown in fig. 4, and may remove irrelevant information to obtain a modified multi-scale sampling subgraph. The branch reduction module in the memory 102 can obtain the linear contour map of the card image to be measured according to the linear contour map of each scale, as shown in fig. 6.
The processor 103 is used for all operations and control of the detection of the linear contour of the object, and can control the functional implementation and data input and output of the rest of the sensor 101, the memory 102 and the external interface 104.
The external interface 104 is a bidirectional interface, which can input data and control commands through input devices such as a handwriting pad, a keyboard and a mouse, and can output intermediate data and card linear profiles through a display, a server, an external device, and the like. Those skilled in the art will appreciate that the hardware configuration depicted in FIG. 1 is not a limitation of the present invention and may include more or fewer components than those depicted, or may be rearranged or replaced with different components.
Based on the hardware architecture, an embodiment of the method for detecting the linear profile of the object is provided, and card linear profile detection is taken as an example.
Referring to fig. 2, a flow chart of an embodiment of a method for detecting a straight line contour of an object is shown, which is detailed as follows:
in step S10, a card image is obtained, and the image is preprocessed to obtain an image of the card to be tested.
In the embodiment of the invention, the acquisition of the card image can be carried out by a device with a CCD or CMOS image sensor, such as a mobile phone, a digital camera, a notebook computer and the like. Because the card image can express the image content by gray scale, RGB, HSV, CMYK and the like, in order to unify the expression form of the image content of the card to be detected and reduce the space complexity, the card image is subjected to the preprocessing of gray scale and standardization, the data formats of RGB, HSV, CMYK and the like are converted into the gray scale value, the image size is scaled to the specified size range in an equal proportion, and the card image to be detected is obtained.
In step S20, the card image to be detected is smoothed in multiple scales, and is corrected by using a preset smoothing module, so as to obtain a corrected multi-scale sampling sub-image.
In this example, after obtaining the card image to be tested, K-order unit segmentation of the card image to be tested is first performed, where K ═ a1,a2,...,an-1,anAnd | n is less than 10}, sequentially extracting sampling subgraphs of each scale from left to right and from top to bottom from the card image to be detected, and extracting the pixel point set of the sampling subgraphs of each scale to be phikThe mathematical expression form of the pixel point set of each scale sampling subgraph is phik＝{pij,p(i+k)j,pi(j+k),p(i+k)(j+k),...,p(i+nk)(j+mk)|m,n≤Psizeand/K }. Wherein p isijRepresenting pixel points, i and j represent relative coordinates of the pixel points in the multi-scale sampling subgraph, k represents offset of the pixel points, and PsizeAnd the size of the sampling subgraph of each scale is shown, and K represents the scale. Dividing the card image to be tested into K2After a plurality of multi-scale sampling subgraphs, sequentially adding continuous indexes F to each scale sampling subgraph according to the sequence of the segmentation sub-steps, wherein the value range of the indexes is that F is more than or equal to 1 and less than or equal to K2Said substepsFor example, as shown below, if K is 2, there are 4 sampling sub-graphs, and each adjacent 2 × 2 pixel point of the card image to be detected belongs to different sampling sub-graphs in turn, the process of sequentially extracting each 2 × 2 pixel point is a segmentation sub-step. And then, the multi-scale sampling subgraph is transmitted to a preset smoothing module, irrelevant information is removed, and a modified multi-scale sampling subgraph is obtained. Referring to FIG. 4, a flow diagram of an embodiment of a preset smoothing module is shown.
In step S21, the multi-scale sampling sub-images are imported into the preset smoothing module, and for the pixel points, the step length of each scale sampling sub-image is set to be S ═ a in sequence1,a2,...,an-1,anAnd | n is less than K }, and extracting eight neighborhoods of the pixel points according to the step length.
In this example, the step length refers to the interval range of the central point selected from the left to the right and from the top to the bottom, and can be divided into three areas a, B, and C according to the position relationship of each point. The A region is the vertex of each scale sampling subgraph and is characterized in that eight neighborhoods of pixel points in the region only have 3 effective values. The B area is a four-side area of each scale sampling subgraph and is characterized in that eight neighborhoods of pixel points in the area have no value in a certain row or column. The C area is a non-image edge area of each scale sampling sub-image and is characterized in that eight neighborhoods of pixel points in the area are in a full-value state. And (4) carrying out zero value supplement on positions lacking effective values in the areas A and B, and extracting eight neighborhoods of the pixel points from the sampling subgraphs of all scales after the value supplement according to the step length.
In step S22, the difference between the center point and the eight neighborhoods is calculated, and the pixel value of the point with the difference value from the center point ranked as n is substituted for the center point pixel.
In this example, a center point C and eight neighborhood pixels A are calculatedi(i ═ 1, 2.. 8) scanning the eight neighborhoods of the pixels extracted in step S21 according to the Zigzag path, and comparing the center point C with the neighborhood pixels aiIn order from small to large such that | C-A1|≤|C-A2|≤|C-A3|≤...≤|C-A8And use of A |)2Replaces the value of the center point C.
In step S23, the sampled subgraphs of each scale after being replaced by the center point values are subjected to median filtering processing.
In this example, the median filtering is adaptive median filtering, and the initial length of the rectangular region block length L is Min _ L, the maximum length is Max _ L, the initial width of the rectangular region block width W is Min _ W, and the maximum width is Max _ W. The rectangular area block scans the multi-scale sampling sub-image corrected in the step S22 according to the Zigzag path, calculates the mean value M, the maximum value Max _ P and the minimum value Min _ P of the pixel points covered in the rectangular area block, when the rectangular area block increasing condition is met, the matrix area block calculates the mean value again until the rectangular area block increasing condition is not met, the mean value of the pixel points replaces the pixel of the central point of the rectangular area block, and the rectangular area block increasing condition is as follows: m is less than or equal to Min _ p, M is more than or equal to Max _ p, and L is less than or equal to Max _ L, Lambda W is less than or equal to Max _ W.
By the mode, the card image to be detected can be subjected to multi-scale smoothing, and the corrected multi-scale sampling subgraph is obtained by correcting the card image to be detected by using the preset smoothing module. Referring to FIG. 5, a schematic diagram of an embodiment of the pre-set smoothing module result is shown. It can be seen that the preset smoothing module can remove irrelevant information, imaging device system errors and image compression noise, and completely retain the required information.
In step S30, the modified multi-scale sampling sub-image is processed using a nearest neighbor operator, a separable lattice sequence is established by extracting pixel points and nearest neighbor pixel points in the modified multi-scale sampling sub-image, head and tail two-point characteristics and nearest neighbor sequence characteristics are obtained from the separable lattice sequence, and continuous characteristic pixel points are obtained according to the point characteristics and the sequence characteristics.
In this example, the nearest neighbor pixel point indicates that the pixel point and the central pixel point have the nearest pixel point of the pixel point in the connection direction in the N × N neighborhood. Extracting and correcting a pixel point P and a nearest neighbor pixel point P1 in the multi-scale sampling subgraph, adding separable lattice sequences into the two pixel points P and P1 when the two pixel points P and P1 are both in separable lattices, continuously extracting a nearest neighbor pixel point P2 of the nearest neighbor pixel point P1 for iteration, namely judging whether the pixel point P1 and the nearest neighbor pixel point P2 are in separable lattices, if so, adding the two pixel points P1 and P2 into the separable lattice sequences, and continuously extracting the nearest neighbor pixel points P2 of the nearest neighbor pixel point P1 for iterationThe extraction and judgment operations are repeated. And when the nearest neighbor pixel point exists as the image boundary or the pixel point exists in the lattice subsequence, finishing a separable lattice sequence. The divisible grids are expressed as follows, firstly, the difference value of every two pixels of 2 x 2 pixel points is sorted from small to large, then the difference value of diagonal pixels is removed, and the difference value of the 2 nd large pixel difference value and the 3 rd large pixel difference value is greater than a set threshold value h1The 2 x 2 pixel points are divided into separable lattices. The head and tail two-point characteristic of the separable lattice sequence is the angle between the connection line of the head and tail two-point of the sequence and the actual connection line of the two points, and the nearest neighbor sequence characteristic is the steering condition of the sequence in the process from head to tail. The turning condition is the turning condition of the sequence in eight neighborhood directions of 'up, down, left, right, left up, left down, right up and right down'. According to the characteristics of the head and the tail points and the characteristics of the nearest neighbor sequence, the angle is smaller than a set threshold value h2And the sequence without the turning condition is used as continuous characteristic pixel points which have scale attributes and comprise pixel points with various scales meeting the requirements.
In step S40, binarization processing is performed on all the pixel points of the modified multi-scale sampling subgraph according to the continuous characteristic pixel points.
In this example, according to the continuous characteristic pixel points obtained in step S30, marking 1 in the corresponding modified multi-scale sampling sub-image, marking 0 for the remaining points of the modified multi-scale sampling sub-image, and according to the binarization marks in the modified multi-scale sampling sub-image, setting the gray value of the pixel point marked as 1 as 255 and the gray value of the pixel point marked as 0, to obtain a black-and-white image of the continuous characteristic pixel points.
In step S50, a linear contour map of each scale is created according to the black-and-white images of the continuous characteristic pixel points, and a linear contour map of the card image to be detected is obtained by using the branch reduction module.
In this example, the linear profile of each scale is established by the relative coordinates of the pixels in the black-and-white image of the continuous characteristic pixels and the step length s of the scale ═ a1,a2,...,an-1,anI n < K }, sequentially arranging from top to bottom and from left to right to obtain linear contour graphs of all scales, and then arranging the linear contour graphs of all scalesAnd transmitting the image to a deduction module to obtain a linear profile of the image of the card to be detected. Referring to FIG. 6, a flow diagram of an embodiment of a mitigation module is shown.
In step S51, the principal direction of any pixel point in each scale straight line contour map under m adjacent points is marked.
In this example, the selection of any pixel point follows the directions from left to right and from top to bottom, a point containing a straight line profile in each scale straight line profile graph is defined as a straight line profile point, the pixel point of the first straight line profile point at the upper left is selected as a starting point, the straight line profile point existing in the m × m rectangular block with the pixel point as the center is taken as a point in the adjacent direction, if the straight line profile point is in one adjacent direction or two adjacent directions, more directions of the straight line profile points are directly marked as main directions, and if the straight line profile point is larger than the two adjacent directions, the straight line profile point is marked as any direction.
In step S52, a set of continuous line segments is obtained for each scale straight line contour map using Depth First Search (DFS) in the recording direction.
In this example, the principal directions of any pixel points in the linear contour map of each scale obtained in step S51 under the condition of m adjacent points are substituted into the depth-first search, the principal direction with the most linear contour points in the depth-first search process is set as the marked principal direction, the marked principal direction is used to cover the principal direction in which the linear contour points with any direction are found in the depth-first search process, and the principal direction ratio of the principal direction with the most linear contour points in the priority search process is extracted to reach the threshold value h of each scale3Obtaining a continuous line segment set.
For example: the method comprises the following steps that two straight line profiles exist in a straight line profile, 5 straight line profile points exist in one straight line profile, the main directions of 4 points are the same and are in the direction a, and the main direction of 1 point is not in the direction a; the other line has 10 points, the main directions of the 5 points are the same as the b direction, and the main directions of the 5 points are not the b direction. The ratio of the main directions of the 5 straight line contour points is 80%, the ratio of the main directions of the 10 straight line contour points is 50%, and if the threshold value is 70%, the 5 straight line contour points are added into the continuous line segment set.
In step S53, the line segment is determined from the continuous line segment of the straight-line profile for each scaleSet length extraction greater than each scale threshold h4A continuous line segment of (a).
In the example, the continuous line segment sets of the straight line contour diagrams of all scales are subjected to length sorting, and the length is kept to be larger than the threshold value h of all scales4The pixel value of the pixel point of the continuous line segment set is not changed, and the length is less than or equal to the threshold value h of each scale4Removing the pixel values of the pixel points of the continuous line segment set.
In step S54, the continuous line segments of each scale are interpolated and combined according to the scale size to obtain a straight line contour map of the card image to be measured.
In the embodiment, the difference priority of the image of the corresponding scale is determined according to the scale K, and the linear contour points are extracted from the continuous line segments of the scales according to the Zigzag path and the priorities of the scales, so that the linear contour graph of the image of the card to be detected is obtained.
For example: if the scale is K ═ 3,1,2, a blank linear contour map is established first, the size of the blank linear contour map is consistent with that of the card image to be detected, whether a pixel point is inserted at the position of the linear contour map (0,0) or not is judged, whether the point has a value in a continuous line segment with the scale of 3 is judged first, and if the point has no value in the continuous line segment with the scale of 3, the subsequent judgment is cancelled to insert a black pixel point; and then sequentially judging whether the point in the continuous line segments with the scales of 1 and 2 has a value or not, and if the point in the continuous line segment with a certain scale has no value, cancelling the subsequent judgment and inserting the black pixel point.
By the method, the linear contour map of each scale can be established according to the attributes of the continuous characteristic pixel points in the linear space of each scale, and then the linear contour map of the card image to be detected is obtained by using the deduction module. Referring to FIG. 7, a schematic diagram of one embodiment of the result of the branch reduction module is shown. The branch reducing module can reduce other straight line branches and curves outside the straight line profile of the card, and completely reserve the straight line profile of the card.
In step S60, the card linear contour is positioned, and the linear contour of the card image to be detected is used as a positioning range, and contour calibration is performed on the card image to obtain a card identification area, so as to improve the card identification accuracy.
Referring to fig. 8, a schematic diagram of an embodiment of a detection result of a straight line contour of an object is shown. The card linear contour positioning can be seen to completely mark the area of the card to be identified, and the card linear contour positioning is completely attached to the edge of the card, so that the card identification precision is improved.
In addition, the embodiment of the invention also provides an object straight line contour detection device.
Referring to fig. 3, a schematic structural diagram of an object straight line contour detection apparatus is shown, and for convenience of explanation, only the parts related to the embodiment of the present invention are shown. In the embodiment of the present invention, the apparatus for detecting the linear contour of an object includes an object image multi-scale sampling unit 201, a continuous feature pixel point extracting unit 202, a multi-scale merging unit 203, and a linear contour detecting unit 204.
The object image multi-scale sampling unit 201 is configured to obtain an object image, perform graying and normalization on the object image to obtain an object image to be detected, perform multi-scale smoothing on the object image to be detected, and remove irrelevant information, imaging device system errors, and image compression noise by using a preset smoothing module to obtain a modified multi-scale sampling sub-image.
The continuous characteristic pixel point extracting unit 202 is configured to process a nearest neighbor operator of the obtained modified multi-scale sampling sub-image, extract pixel points in the modified multi-scale sampling sub-image and nearest neighbor pixel points to establish a separable lattice sequence, obtain a sequence head-tail two-point characteristic and a nearest neighbor sequence characteristic, and obtain continuous characteristic pixel points according to the point characteristic and the sequence characteristic.
The multi-scale merging unit 203 is configured to obtain a linear profile of the image of the object to be measured, and obtain a black-and-white image of the continuous characteristic pixel points according to the continuous characteristic pixel points. And establishing a linear profile map of each scale based on the black-and-white map of the continuous characteristic pixel points, and then obtaining the linear profile map of the object image to be detected by using the deduction module.
The linear contour detection unit 204 is used for positioning the linear contour of the object, and using the linear contour of the image of the object to be detected as a positioning range to perform contour calibration on the image of the object to obtain an object identification area, so that the accuracy of object identification is improved.
Alternatively, the acquired object image may be derived from the image sensor 101 or may be image data stored in the memory 102.
Optionally, an empty data container is prestored in the memory 102 for filling, and the empty data container may also be directly imported after being processed by the processor 103.
Alternatively, the object image located by the straight line contour detection unit 204 may be an RGB color image, a grayscale image, and a black-and-white image.
Further, the multi-scale threshold is greater than or equal to 30.
Further, each virtual function unit of an object straight line contour detection device is stored in the memory 102 of an object straight line contour detection device shown in fig. 1, and is used for realizing a preset smoothing module, a deduction module and all function programs; when executed by the processor 103, the units can implement a function of detecting the straight line contour of the object.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-mentioned division of the functional units and modules is illustrated, and in practical applications, the above-mentioned function distribution may be performed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-mentioned functions. Each functional unit and module in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units are integrated in one unit, and the integrated unit may be implemented in a form of hardware, or in a form of software functional unit. In addition, specific names of the functional units and modules are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present application.
The function implementation of each module in the above object straight line contour detection apparatus corresponds to each step in the above object straight line contour detection method embodiment, and the function and implementation process thereof are not described in detail here.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system.
In summary, compared with the prior art, the method, the device and the equipment for detecting the object straight line contour do not need space transformation and data set training, reduce the running time and the calculated amount, do not need intersection of a straight line equation and straight line fitting, realize robust object straight line contour detection and improve the accuracy of the straight line contour. Those skilled in the art will clearly understand that the method of the above embodiments can be implemented by embedded software in combination with necessary general hardware, or can be implemented by only hardware stored in integrated hardware, but the former is a better implementation mode in most cases. Based on such understanding, the technical solutions of the present invention may be used as a stand-alone software product, which is stored in a storage medium (such as a ROM/RAM, an optical disc, a usb disk, etc.) with a reading function, and includes several instructions to enable a device (such as a mobile phone, a computer, a server, or a network device, etc.) with the functions of reading, compiling, and executing instructions and data to execute the methods according to the embodiments of the present invention.
The above are merely preferred embodiments of the present application and are not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (7)
1. A method for detecting the straight line contour of an object is characterized by comprising the following steps:
step S10, obtaining an image of the card to be detected: acquiring a card image by using an image storage module or an embedded image sensor, and preprocessing the acquired card image to obtain a card image to be detected;
step S20, multi-scale smoothing: performing K-order unit segmentation on the card image to be detected to obtain K2Sequentially adding continuous indexes to each scale sampling subgraph according to the sequence of the segmentation sub-steps; then the multi-scale sampling subgraph is transmitted to a preset smoothing module, irrelevant information is removed from the preset smoothing module, and a modified multi-scale sampling subgraph is obtained;
step S30, nearest neighbor operator processing: extracting pixel points and nearest neighbor pixel points in the corrected multi-scale sampling subgraph to establish separable lattice sequences, extracting head and tail two-point characteristics and nearest neighbor sequence characteristics of the separable lattice sequences, and obtaining continuous characteristic pixel points according to the point characteristics and the sequence characteristics;
step S40, binarization processing: marking the position of a continuous characteristic pixel point in the modified multi-scale sampling subgraph as 1, marking the position of the rest data points in the modified multi-scale sampling subgraph as 0, and obtaining a black-white graph of the continuous characteristic pixel point according to the binarization mark;
step S50, scale merging: establishing a linear contour map of each scale according to the black-and-white map of the continuous characteristic pixel points, and transmitting the linear contour map of each scale to a deduction module to obtain a linear contour map of the card image to be detected;
step S60, positioning the card straight line contour: the linear profile graph of the card image to be detected is used as a positioning range, and the card image is subjected to profile calibration to obtain a card identification area, so that the card identification precision is improved.
2. The method for detecting the linear profile of the object according to claim 1, wherein the step S20 is implemented as follows:
in step S21, after the card image to be tested is obtained, K-order unit segmentation of the card image to be tested is first performed, where K is { a {1,a2,...,an-1,anAnd | n is less than 10}, sequentially extracting sampling subgraphs of each scale from left to right and from top to bottom from the card image to be detected, and extracting the pixel point set of the sampling subgraphs of each scale to be phikThe mathematical expression form of the pixel point set of each scale sampling subgraph is phik＝{pij,p(i+k)j,pi(j+k),p(i+k)(j+k),...,p(i+nk)(j+mk)|m,n≤PsizeK }; wherein p isijRepresenting pixel points, i and j represent relative coordinates of the pixel points in the multi-scale sampling subgraph, k represents offset of the pixel points, and PsizeThe size of each scale sampling subgraph is represented, and K represents a scale; dividing the card image to be tested into K2After a plurality of multi-scale sampling subgraphs, sequentially adding continuous indexes F to each scale sampling subgraph according to the sequence of the segmentation sub-steps, wherein the value range of the indexes is that F is more than or equal to 1 and less than or equal to K2，
Step S22, importing the multi-scale sampling sub-images into a preset smoothing module, and sequentially setting the step length S ═ a for pixel points of each multi-scale sampling sub-image1,a2,...,an-1,anL n is less than K, and eight neighborhoods of the pixel points are extracted according to the step length;
step S23, calculating the difference value between the central point and the eight neighborhoods, and replacing the central point pixel with the pixel value of the point with the difference value from the central point ranked as n;
calculating a central point C and eight neighborhood pixels AiWherein i is 1,2,. 8; scanning eight neighborhoods of the pixel points extracted in the step S22 according to the Zigzag path, and connecting the center point C with the neighborhood pixel point AiIn order from small to large such that | C-A1|≤|C-A2|≤|C-A3|≤...≤|C-A8And use of A |)2Replaces the value of the center point C;
step S24, carrying out median filtering processing on the sampling subgraphs of all scales after the center point value replacement; the median filtering adopts self-adaptive median filtering, the initial length of the length L of the rectangular area block is set as Min _ L, the maximum length is set as Max _ L, the initial width of the width W of the rectangular area block is set as Min _ W, and the maximum width is set as Max _ W; the method comprises the following steps that a rectangular area block scans and corrects a multi-scale sampling sub-image according to a Zigzag path, the mean value M, the maximum value Max _ P and the minimum value Min _ P of pixel points covered in the rectangular area block are calculated, when the rectangular area block increasing condition is met, the matrix area block calculates the mean value again until the rectangular area block increasing condition is not met, the mean value of the pixel points replaces pixels at the center point of the rectangular area block, and the rectangular area block increasing condition is as follows: m is less than or equal to Min _ p, M is more than or equal to Max _ p, and L is less than or equal to Max _ L, Lambda W is less than or equal to Max _ W.
3. The method for detecting the linear contour of an object according to claim 1, wherein the step length refers to the interval range of the central point selected at intervals in two directions from left to right and from top to bottom, and is divided into three areas A, B and C according to the position relationship of each point; the region A is the vertex of each scale sampling subgraph and is characterized in that eight neighborhoods of pixel points in the region have only 3 effective values; the B region is a four-side region of each scale sampling subgraph and is characterized in that eight neighborhoods of pixel points in the region have no value in a certain row or column; the C region is a non-image edge region of each scale sampling sub-image and is characterized in that eight neighborhoods of pixel points in the region are in a full-value state; and (4) carrying out zero value supplement on positions lacking effective values in the areas A and B, and extracting eight neighborhoods of the pixel points from the sampling subgraphs of all scales after the value supplement according to the step length.
4. The method for detecting the linear profile of the object according to claim 3, wherein the step S30 is implemented as follows:
the nearest neighbor pixel point represents that the pixel point and the central pixel point have the nearest pixel point of the pixel point in the N x N neighborhood in the connecting direction; extracting a pixel point P and a nearest neighbor pixel point P1 in the modified multi-scale sampling subgraph, when two pixel points P and P1 are in separable lattices, adding separable lattice sequences into the two pixel points P and P1, continuously extracting a nearest neighbor pixel point P2 of the nearest neighbor pixel point P1 for iteration, namely judging whether the pixel point P1 and the nearest neighbor pixel point P2 are in the separable lattices, if so, adding the two pixel points P1 and P2 into the separable lattice sequences, and continuously repeating the extraction and judgment operations; and when the nearest neighbor pixel point exists as the image boundary or the pixel point exists in the lattice subsequence, finishing a separable lattice sequence.
5. The method as claimed in claim 4, wherein the separable lattice is described as follows:
firstly, the difference value of every two pixels of 2 x 2 pixel points is increased from small to largeSorting, and removing the diagonal pixel difference value, the difference between the 2 nd large pixel difference value and the 3 rd large pixel difference value is larger than the set threshold value h1The 2 x 2 pixel points are separable lattices; the head and tail two-point characteristic of the separable lattice sequence is the angle between the connection line of the head and tail two points of the sequence and the actual connection line of the two points, and the nearest neighbor sequence characteristic is the steering condition of the sequence in the process from head to tail; the steering condition is the turning condition of the sequence in eight neighborhood directions of 'up, down, left, right, upper left, lower left, upper right and lower right'; according to the characteristics of the head and the tail points and the characteristics of the nearest neighbor sequence, the angle is smaller than a set threshold value h2And the sequence without the turning condition is used as continuous characteristic pixel points which have scale attributes and comprise pixel points with various scales meeting the requirements.
6. The method for detecting the linear contour of an object according to claim 4 or 5, wherein the step S50 is to establish linear contour maps of various scales according to the black-and-white images of the continuous characteristic pixel points, obtain the linear contour map of the card image to be detected by using the deduction module, and establish the linear contour maps of various scales by setting the relative coordinates and the step length S ═ a of the scale of the pixel points in the black-and-white images of the continuous characteristic pixel points1,a2,...,an-1,anI n < K }, sequentially arranging from top to bottom and from left to right to obtain linear contour maps of all scales, and transmitting the linear contour maps of all scales to a deduction module to obtain a linear contour map of the card image to be detected, wherein the specific implementation is as follows:
step S51, marking the main direction of any pixel point in the linear contour graph of each scale under the adjacent m points; selecting any pixel point along the directions from left to right and from top to bottom, defining a point containing a straight line outline in each scale straight line outline graph as a straight line outline point, selecting a pixel point of the first straight line outline point at the upper left as a starting point, taking a straight line outline point existing in a rectangular block of m × m with the pixel point as the center as a point in the adjacent direction, directly marking more directions of the straight line outline points as main directions if the straight line outline points are in one adjacent direction or two adjacent directions, and marking the straight line outline points as any directions if the straight line outline points are more than two adjacent directions;
step S52Obtaining a continuous line segment set by using depth-first search in the recording direction for the linear profile graph of each scale; substituting the principal directions of any pixel points under adjacent m points in the linear contour map of each scale obtained in the step S51 into depth-first search, setting the principal direction with the most linear contour points in the depth-first search process as a marked principal direction, covering the principal direction in which the linear contour points with any direction are found in the depth-first search process by using the marked principal direction, and extracting the principal direction ratio with the most linear contour points in the priority search process to reach a threshold value h of each scale3Obtaining a continuous line segment set by the straight line contour points;
step S53, extracting the threshold h larger than each scale according to the length of the continuous line segment set of the straight line contour graph of each scale4The continuous line segments of the linear contour graph of each scale are sorted according to the length, and the length is kept to be larger than the threshold value h of each scale4The pixel value of the pixel point of the continuous line segment set is not changed, and the length is less than or equal to the threshold value h of each scale4Removing pixel values of pixel points of the continuous line segment set;
and S54, interpolating and merging the continuous line segments of all scales according to the sizes of the scales to obtain a linear contour map of the card image to be detected, determining the difference priority of the corresponding scale image according to the scale K, and extracting linear contour points from the continuous line segments of all scales according to the Zigzag path and the priorities of all scales to obtain the linear contour map of the card image to be detected.
7. The method for detecting the linear contour of the object according to claim 6, wherein the corresponding device of the method comprises an object image multi-scale sampling unit (201), a continuous feature pixel point extracting unit (202), a multi-scale merging unit (203) and a linear contour detecting unit (204);
the object image multi-scale sampling unit (201) is used for acquiring an object image, graying and standardizing the object image to obtain an object image to be detected, performing multi-scale smoothing on the object image to be detected, and removing irrelevant information, imaging equipment system errors and image compression noise by using a preset smoothing module to obtain a modified multi-scale sampling sub-image;
the continuous characteristic pixel point extraction unit (202) is used for processing nearest neighbor operators of the obtained modified multi-scale sampling subgraph, extracting pixel points in the modified multi-scale sampling subgraph and nearest neighbor pixel points to establish separable lattice sequences, obtaining sequence head and tail two-point characteristics and nearest neighbor sequence characteristics according to the separable lattice sequences, and obtaining continuous characteristic pixel points according to the point characteristics and the sequence characteristics;
the multi-scale merging unit (203) is used for acquiring a linear profile of an image of an object to be detected and obtaining black and white images of continuous characteristic pixel points according to the continuous characteristic pixel points; establishing a linear profile graph of each scale based on the black-and-white graph of the continuous characteristic pixel points, and then obtaining the linear profile graph of the object image to be detected by using a deduction module;
the linear contour detection unit (204) is used for positioning the linear contour of the object, taking the linear contour of the image of the object to be detected as a positioning range, carrying out contour calibration on the image of the object to obtain an object identification area, and improving the accuracy of object identification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111314707.6A CN114140620A (en) | 2021-11-08 | 2021-11-08 | Object straight line contour detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111314707.6A CN114140620A (en) | 2021-11-08 | 2021-11-08 | Object straight line contour detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114140620A true CN114140620A (en) | 2022-03-04 |
Family
ID=80393226
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111314707.6A Pending CN114140620A (en) | 2021-11-08 | 2021-11-08 | Object straight line contour detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114140620A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117236247A (en) * | 2023-11-16 | 2023-12-15 | 零壹半导体技术(常州)有限公司 | Signal shielding wire generation method for chip test |
-
2021
- 2021-11-08 CN CN202111314707.6A patent/CN114140620A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117236247A (en) * | 2023-11-16 | 2023-12-15 | 零壹半导体技术(常州)有限公司 | Signal shielding wire generation method for chip test |
| CN117236247B (en) * | 2023-11-16 | 2024-01-23 | 零壹半导体技术(常州)有限公司 | Signal shielding wire generation method for chip test |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110046529B (en) | Two-dimensional code identification method, device and equipment | |
| CN111028213A (en) | Image defect detection method and device, electronic equipment and storage medium | |
| CN101650783B (en) | Image identification method and imaging apparatus | |
| KR101939936B1 (en) | Method and apparatus for recognizing fingerprints | |
| US20070253040A1 (en) | Color scanning to enhance bitonal image | |
| US7916957B2 (en) | Line noise eliminating apparatus, line noise eliminating method, and line noise eliminating program | |
| US20090016608A1 (en) | Character recognition method | |
| CN109509200A (en) | Checkerboard angle point detection process, device and computer readable storage medium based on contours extract | |
| US10769478B2 (en) | Convolutional neutral network identification efficiency increasing method and related convolutional neutral network identification efficiency increasing device | |
| US8995730B2 (en) | Image processing apparatus for analyzing and enhancing fingerprint images | |
| CN112183511A (en) | Method, system, storage medium and equipment for deriving table from image | |
| CN111220235B (en) | Water level monitoring method and device | |
| CN108960247B (en) | Image significance detection method and device and electronic equipment | |
| CN114140620A (en) | Object straight line contour detection method | |
| KR101842535B1 (en) | Method for the optical detection of symbols | |
| CN116091933A (en) | Geological analysis method and device for water area topography based on remote sensing technology | |
| CN111242051A (en) | Vehicle identification optimization method and device and storage medium | |
| JP5424694B2 (en) | Image recognition apparatus and program | |
| CN114445814A (en) | Character region extraction method and computer-readable storage medium | |
| CN114022856A (en) | Unstructured road travelable area identification method, electronic device and medium | |
| CN113159027A (en) | Seven-segment type digital display instrument identification method based on minimum external rectangle variant | |
| CN111476800A (en) | Character region detection method and device based on morphological operation | |
| CN112308812A (en) | Method, terminal and storage medium for detecting picture definition | |
| CN113255657B (en) | Method and device for detecting scratch on bill surface, electronic equipment and machine-readable medium | |
| CN109271986B (en) | Digital identification method based on Second-Confirm |
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 |