WO2019237520A1 - 一种图像匹配方法、装置、计算机设备及存储介质 - Google Patents
一种图像匹配方法、装置、计算机设备及存储介质 Download PDFInfo
- Publication number
- WO2019237520A1 WO2019237520A1 PCT/CN2018/104050 CN2018104050W WO2019237520A1 WO 2019237520 A1 WO2019237520 A1 WO 2019237520A1 CN 2018104050 W CN2018104050 W CN 2018104050W WO 2019237520 A1 WO2019237520 A1 WO 2019237520A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- pixel
- feature
- points
- point
- Prior art date
Links
Images
Classifications
-
- 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
- G06V40/12—Fingerprints or palmprints
- G06V40/1365—Matching; Classification
- G06V40/1376—Matching features related to ridge properties or fingerprint texture
Definitions
- the present application relates to the field of computer technology, and in particular, to an image matching method, device, computer device, and storage medium.
- the traditional finger vein matching algorithm has unsatisfactory matching results, misrecognition rate, and large degree of implementation complexity, resulting in low accuracy of the matching.
- the accuracy of finger vein vein matching in the verification process Low, frequent verification failures, reducing the success rate of verification.
- An image matching method includes:
- a cosine value between a feature vector of the feature point and a feature vector of each legal feature point in a legitimate image is calculated, where the legitimate image refers to a finger vein that has been legally authenticated in advance image;
- the cosine value is less than a preset cosine threshold, a feature point obtained by the cosine value is confirmed as a matching point;
- An image matching device includes:
- An acquisition module for acquiring an image of a finger vein to be detected
- a grayscale module configured to perform grayscale processing on the finger vein image to be detected to obtain a grayscale image
- a feature point extraction module for extracting feature points of a vein pattern of a finger in the grayscale image using a fast algorithm
- a vector calculation module configured to process the feature points by using a sift algorithm to obtain a feature vector of each of the feature points;
- a similarity calculation module is configured to calculate, for each of the feature points, a cosine value between a feature vector of the feature points and a feature vector of each legal feature point in a legitimate image, where the legitimate image refers to Pre-certified images of finger veins;
- a recognition module configured to: if the cosine value is less than a preset cosine threshold, confirm a feature point obtained by the cosine value as a matching point;
- a first matching module is configured to confirm that the finger vein image to be detected matches the legal image if the number of matching points is greater than a preset threshold number of points.
- a computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor.
- the processor executes the computer-readable instructions, the image matching method is implemented. step.
- One or more non-volatile readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to execute the image matching method described above A step of.
- FIG. 1 is a schematic diagram of an application environment of an image matching method provided by an embodiment of the present application
- FIG. 2 is a flowchart of an image matching method according to an embodiment of the present application.
- step S2 in the image matching method according to an embodiment of the present application
- step S3 is a flowchart of step S3 in the image matching method according to an embodiment of the present application.
- step S4 is a flowchart of step S4 in the image matching method according to an embodiment of the present application.
- FIG. 6 is a flowchart of performing matching again by using a displacement calculation method in the image matching method according to an embodiment of the present application
- FIG. 7 is a schematic diagram of an image matching device according to an embodiment of the present application.
- FIG. 8 is a schematic diagram of a computer device according to an embodiment of the present application.
- FIG. 1 illustrates an application environment provided by an embodiment of the present application.
- the application environment includes a server and a client, where the server and the client are connected through a network, and the client is used to collect a finger vein image, and Send the captured finger vein image to the server.
- the client can specifically but not limited to a camera, camera, scanner or a finger vein image acquisition device with other photographing functions; the server is used to perform finger veins on the finger vein image. Grain matching, the server can be implemented by an independent server or a server cluster composed of multiple servers.
- the image matching method provided in the embodiment of the present application is applied to a server.
- an image matching method is provided.
- the method is applied to the server in FIG. 1 as an example, and includes the following steps:
- S1 Acquire an image of a finger vein to be detected.
- the finger vein image to be detected refers to a finger vein image directly acquired from a finger vein acquisition device without any processing.
- S2 Perform grayscale processing on the finger vein image to be detected to obtain a grayscale image.
- the pixel values of all pixels in the finger vein image are adjusted to be between 0 and 255 to obtain a gray image of the finger vein image.
- the feature point refers to a pixel point capable of identifying a vein pattern of a finger vein in a grayscale image.
- the grayscale image obtained in step S2 is performed by using a corner detection (Features from accelerated segmentation test, fast) algorithm. Processing, extracting the pixel points of the finger vein pattern in the grayscale image as feature points.
- the fast algorithm is an algorithm for corner detection.
- the principle of the algorithm is to take the detection point in the image and use the detection point as the center of the circle. By comparing the pixel value of the pixels around the center of the circle with the pixel value of the detection point To determine whether the detection point is a corner point.
- corner points determined by the fast algorithm are used as feature points.
- S4 The sift algorithm is used to process the feature points to obtain the feature vector of each feature point.
- the number of feature points may be multiple.
- the feature vector refers to a vector of feature points, that is, the amount of feature points having a size and direction. According to the feature points obtained in step S3, the scale is unchanged.
- a feature transformation (Scale Invariant Feature Transformation, sift) algorithm processes each feature point to obtain a feature vector for each feature point.
- the sift algorithm is a scale-invariant feature transformation algorithm, which is mainly used to extract the local features of an image.
- the feature vectors of the feature points are obtained by calculation on different scale spaces.
- S5 For each feature point, calculate a cosine value between the feature vector of the feature point and the feature vector of each legal feature point in the legal image, where the legal image refers to a finger vein image that has been legally authenticated in advance.
- the feature vector of the feature point and each of the legitimate images are calculated. Cosine value between feature vectors of legal feature points.
- Legal images are stored in a legal database in advance.
- Legal images refer to finger vein images that have been legally authenticated in advance, and are mainly used to match the finger vein images to be detected.
- the feature points of the legitimate image and the feature vectors of the feature points of the legitimate image can be obtained by the same method as in this case, and are stored in the legitimate database corresponding to the legitimate image.
- the preset cosine threshold is mainly used to determine whether the two feature points are matching points with each other.
- the specific value can be set according to actual application requirements, and is not limited here.
- the cosine value calculated in step S5 it is compared with a preset cosine threshold. When the cosine value is less than the preset cosine threshold, it is confirmed that the characteristic point of the cosine value is the cosine value. Matching points of legal feature points.
- the preset point threshold is used to determine whether the finger vein patterns of the two images match.
- the specific value can be set according to actual application requirements, and is not limited here.
- the number of matching points between the finger vein image to be detected and the legitimate image is obtained according to step S6, and it is compared with a preset number of points threshold. When the number of matching points is greater than the preset number of points threshold, it is confirmed
- the finger vein image to be detected matches a legitimate image, that is, the finger vein image to be detected and the legitimate image are homologous finger vein images.
- a grayscale image is obtained by performing grayscale processing on a finger vein image to be detected, and a fast algorithm is used to extract the feature points of the finger vein pattern in the grayscale image, and a feature vector of each feature point is obtained by the sift algorithm. And according to the feature vector, calculate the cosine value between the feature vector of the feature point in the grayscale image and the feature vector of each legal feature point in the legitimate image, and then determine the number of matching points according to the cosine value, and finally match the points The number of is compared with a preset threshold. If the number of matching points is greater than the preset threshold, it is confirmed that the finger vein image to be detected matches a legitimate image.
- the fast algorithm is used to extract the feature points of the finger vein veins, and the high efficiency of the fast algorithm is used to quickly extract the feature points of the finger vein veins, thereby improving the recognition efficiency of the finger vein vein feature points;
- the fast algorithm and sift algorithm are combined to obtain the feature vector of the feature points, and the matching point is determined by calculating the cosine value of the feature vector to achieve accurate matching of finger vein veins and effectively improve the accuracy of finger vein vein matching.
- step S2 the grayscale processing is performed on the finger vein image to obtain a grayscale image, which specifically includes the following steps:
- the pixels in the finger vein image to be detected are traversed according to a preset traversal method to obtain the RGB component value of each pixel, where R, G, and B respectively represent the three channels of red, green, and blue. colour.
- the preset traversal method may specifically start from the top left pixel point of the finger vein image to be detected, and traverse line by line from top to bottom from left to right, or it may be from the finger vein image to be detected.
- the position of the center line is traversed to both sides at the same time, and other traversal methods can be used, which is not limited here.
- x and y are the abscissa and ordinate of each pixel in the finger vein image to be detected
- g (x, y) is the gray value after the pixel (x, y) is grayed
- R ( x, y) is the color component of the R channel of the pixel (x, y)
- G (x, y) is the color component of the G channel of the pixel (x, y)
- B (x, y) is the pixel ( x, y) color component of the B channel.
- the grayscale processing of the finger vein image to be detected is required, and the R channel, the G channel and the B channel are respectively expressed by formula (1).
- the ratio of the channels is set, so that the finger vein image to be detected is converted into a grayscale image.
- the RGB model is a commonly used expression of color information. It uses the brightness of the three primary colors of red, green, and blue to quantify the color.
- This model is also called additive color mixing model, which is a method of mixing colors by superimposing RGB three-color light on each other. Therefore, this model is suitable for the display of light emitters such as displays.
- the component is used to calculate the gray value by using formula (1).
- the image may be grayed by using the weighting method, the maximum value method, or the average value method. There are no restrictions here.
- formula (1) is applied to the finger vein to be detected.
- the image is grayed out, so that the pixel value range of the pixels in the image is set between 0-255, which further reduces the amount of original data in the image and improves the calculation efficiency in subsequent processing calculations.
- step S3 the feature points of the vein veins of the fingers in the grayscale image are extracted using the fast algorithm, which specifically include the following steps:
- the pixel points in the grayscale image obtained in step S2 are traversed according to a preset traversal manner, the traversed pixels are used as basic pixel points, and the pixel value of each basic pixel point is obtained.
- the preset traversal method may be to traverse the pixels in the grayscale image line by line from top to bottom and from left to right, or it may be to step through the pixels in the grayscale image from the middle to the two sides simultaneously. Line traversal, no restrictions here.
- S32 The pixels on the circumference with the basic pixel as the center and a preset length threshold as the radius are determined as the target pixels corresponding to the basic pixel, and a target pixel of a preset position is selected from the target pixels as the target pixel Compare pixels.
- the comparison pixel point is a target pixel point for performing pixel value comparison with the basic pixel point.
- the basic pixel point in step S31 is a center and drawing a circle with a preset length threshold as a radius, All pixel points on the circumference of the circle are used as target pixel points corresponding to the basic pixel point, and a target pixel point at a preset position is selected as a comparison pixel point.
- the preset length threshold value can be set by the number of pixels, and the specific value can be set according to the needs of actual applications, which is not limited here.
- the preset length threshold is different, that is, the radius of the circumference is different, and the number of target pixels on the circumference is also different. For example, if the preset length threshold is 2 pixels, there will be 8 target pixels on the circumference Points; if the preset length threshold is 3 pixels, there will be 16 target pixels on its circumference.
- the preset position may be two pixels in the vertical direction on the circumference centered on the basic pixel point, or two pixels in the horizontal direction on the circumference centered on the basic pixel point, or may be
- the basic pixel point is the four pixel points in the horizontal direction and the vertical direction on the circumference of the center.
- the specific position can be set according to the needs of the actual application, which is not limited here.
- S33 For each basic pixel, calculate a first pixel difference between the pixel value of the basic pixel and the pixel value of each comparison pixel corresponding to the basic pixel. If there is a first preset number of If a pixel difference is greater than a preset first threshold, the basic pixel is determined as a candidate pixel.
- the absolute difference between the pixel value of the basic pixel and the pixel value of each comparison pixel corresponding to the basic pixel is taken as the first Pixel difference, and then comparing each first pixel difference with a preset first threshold to count the number of first pixel differences greater than the first threshold, and if the number is greater than or equal to the first preset number, confirm The basic pixel point is a candidate pixel point.
- the other basic pixel points are discarded, that is, the feature points can be obtained by further filtering only among the candidate pixel points.
- the first pixel difference between the basic pixel Q and the comparison pixels Q1, Q2, Q3, and Q4 is 8, 8, 5, 8, if the preset first threshold is 6, and the first preset number is 2, the first pixel difference between the three comparison pixels Q1, Q2, and Q4 and the basic pixel Q is greater than the preset
- the first threshold that is, the number of first pixel differences greater than the first threshold is 3 and greater than the first preset number, so it can be confirmed that the basic pixel Q is a candidate pixel.
- two comparative pixels in the vertical direction on the circumference centered on the basic pixel point can be preferentially selected for calculation, and the first one is selected from the basic pixel points according to the calculation result.
- Candidate pixel points then select four horizontal and vertical comparison pixels on the circumference centered on the basic pixel point, and use the same calculation method to calculate with the first candidate pixel point.
- Out of the pixel points candidate pixel points are filtered out. Excluding the basic pixels through this multiple stepwise screening method can reduce the calculation amount of filtering the candidate pixels and improve the calculation efficiency.
- both the first preset number and the first threshold can be set according to actual application requirements, and there is no limitation here.
- S34 For each candidate pixel point, calculate a second pixel difference between the pixel value of the candidate pixel point and the pixel value of the target pixel point corresponding to the candidate pixel point, if there is a second preset number of second pixels If the difference is greater than the first threshold, the candidate pixel point is determined as a feature point.
- the absolute difference between the pixel value of the candidate pixel point and the pixel value of each target pixel point corresponding to the candidate pixel point is taken as the second Pixel difference, and then comparing each second pixel difference with a preset first threshold to count the number of second pixel differences greater than the first threshold, and if the number is greater than or equal to the second preset number, confirm The candidate pixel point is a feature point.
- the neighborhood refers to an area composed of M * M pixels with the feature point as the center.
- M is a positive integer greater than 1. The specific value of M can be set according to actual application requirements, and is not limited here.
- some non-feature points can be effectively excluded, and then the pixel values of the candidate pixels are compared with the pixel values of their corresponding target pixel points, which can reduce the amount of calculation and thus improve the calculation efficiency of feature points. Realize fast extraction of feature points.
- step S4 the feature points are processed using the sift algorithm, and obtaining the feature vectors of each feature point specifically includes the following steps:
- a region of B * B size is selected as the sampling area of each feature point, where the size identifies the size of the image area, and B is The number of pixels and its specific value can be set according to the needs of the actual application, which is not limited here.
- S42 Rotate the sampling area according to a preset direction to obtain a target image area, and divide the target image area into a preset number of sub-areas.
- sampling area is rotated according to the preset direction according to formula (2):
- x and y are the original x-axis and y-axis of the sampling area
- ⁇ is the direction of rotation
- x 'and y' are the x'-axis and y'-axis after the sampling area is rotated according to a preset direction.
- the image area obtained by rotating the sampling area according to formula (2) is determined as the target image area, and the target image area is divided in an equal interval manner, that is, the B * B size area is equally divided into a preset number Subregions, that is, divided into BP * BP subregions.
- the preset direction may be specifically set according to actual application requirements, and is not limited herein.
- the preset number is a positive integer greater than 1.
- x and y are the horizontal and vertical coordinates of each pixel in each sub-region
- V (x, y) is the gradient of the pixel (x, y) in the horizontal direction
- H (x, y) is The vertical gradient of the pixel (x, y)
- ⁇ (x, y) is the gradient direction of the pixel (x, y)
- m (x, y) is the gradient amplitude of the pixel (x, y) value.
- the gradient direction and gradient magnitude of each pixel in the sub-region are obtained according to formula (3), and the gradient direction of each pixel is mapped to a corresponding preset according to a preset direction range.
- Direction For example, if the gradient direction of pixel E is 30 ° and the gradient direction of pixel F is 80 °, it is assumed that the preset direction includes 45 ° and 90 °, and the range of directions mapped to the 45 ° preset direction is (22.5 ° , 67.5 °], the direction range mapped to the 90 ° preset direction is (67.5 °, 112.5 °], then the pixel E is mapped to the 45 ° preset direction, and the pixel F is mapped to the 90 ° preset Set the direction.
- the gradient amplitude value of each pixel point in the sub-area in the same preset direction is accumulated as the gradient intensity information of the preset direction, that is, Gradient amplitude, and using the starting points of all preset directions as the same starting point to obtain the gradient direction histogram feature of each sub-region, the gradient direction histogram feature includes n preset directions and each preset direction Gradient intensity information.
- n preset directions include 4 preset directions of 0 °, 90 °, 180 °, and 270 °. If five pixel points Q1, Q2, Q3, Q4, and Q5 exist in a sub-region, the corresponding preset directions If the directions are 0 °, 0 °, 90 °, 180 °, and 270 °, and the corresponding gradient amplitudes are 1, 2, 1, 1, and 1, respectively, the gradient amplitudes in the same preset direction are accumulated.
- the gradient amplitudes of pixel Q1 and pixel Q2 are accumulated, and then the starting points of all preset directions are combined as the same starting point, and finally four points with four angles of 0 °, 90 °, 180 °, and 270 ° will be obtained.
- the preset directions, and the magnitudes of the gradients in each of the preset directions are plane vectors of 3, 1, 1, and 1, respectively, and the plane vectors are the characteristics of the gradient direction histogram.
- each sub-region contains gradient intensity information in n directions, and there are Bp * Bp sub-regions, the BP * BP * n-dimensional Feature vector, which is the basic feature vector of the feature point.
- a 32-dimensional feature vector will be obtained after combination, that is, the 32-dimensional feature vector is the feature point.
- Base feature vector For example, if there are 4 sub-areas in the sampling area of the feature point, and gradient intensity information in 8 directions exists in each sub-area, a 32-dimensional feature vector will be obtained after combination, that is, the 32-dimensional feature vector is the feature point. Base feature vector.
- the normalization processing manner may be that each basic feature vector is divided by a maximum value among all the basic feature vectors, or each basic feature vector is divided by an average value of all the basic feature vectors.
- a standard feature vector is obtained, which is convenient for data processing.
- the length of the feature vector can also be truncated. If the length of the feature vector is greater than a preset length, only the preset length is taken as the length of the feature vector, thereby improving the identification of the data processing. Sex.
- the preset length can be set according to the needs of practical applications, and there is no limitation here.
- the length of the feature vector is 0.2.
- the target image area is obtained by rotating the sampling area of each feature point, and the target image area is further divided into a preset number of sub-areas, and the gradient direction histogram features of each sub-area are calculated.
- the gradient histogram features of each sub-region are combined to obtain the basic feature vector of the feature points, and finally the basic feature vector is normalized to obtain the feature vector.
- the features of the sub-areas can be accurately extracted, and then the histogram features of the gradient direction of each sub-area are combined.
- the obtained basic feature vector can accurately reflect the characteristics of the feature point, so that the feature point can be analyzed according to the basic feature vector to improve the accuracy of the feature point feature vector.
- by normalizing the base feature vector It can ensure the discriminativeness of data processing, thereby achieving accurate extraction of feature vectors of feature points.
- step S5 that is, for each feature point, calculating the cosine value between the feature vector of the feature point and the feature vector of each legal feature point in the legitimate image specifically includes the following steps:
- cos ⁇ ij is the cosine value between the feature vector of the i-th feature point in the gray image and the feature vector of the j-th legitimate feature point in the legal image
- a i is the feature of the i-th feature point in the gray image Vector
- b j is the feature vector of the j-th legitimate feature point in the legitimate image
- is the length of a i
- is the length of b j
- i and j are positive integers.
- the feature vector of each feature point in the grayscale image and the feature vector of each legitimate feature point in the legitimate image are calculated according to formula (4).
- the cosine value between the feature vectors of each legal feature point is calculated according to formula (4).
- formula (4) can be used to quickly and accurately calculate the cosine value between the feature vector of each feature point and the feature vector of each legal feature point in the legitimate image, and use the cosine value to accurately determine the relationship between the feature point and the feature vector.
- the degree of feature similarity between legal feature points accurately recognizes matching points and improves the accuracy of vein matching of finger veins, thereby effectively improving the accuracy of authentication when legally taking finger vein images seriously.
- the image matching method further includes the following steps:
- the degree of matching between the finger vein image and the legitimate image may be further judged by relative displacement.
- the average value of the relative displacement between the matching point and the corresponding legal feature point in the legal image is calculated according to formula (5):
- x and y are the horizontal and vertical coordinates of each legal feature point in the legal image and each feature point in the gray image
- P e (x, y) is the e-th matching point in the gray image and the legal Relative displacement between corresponding legal feature points in the image
- P 2e (x, y) is the position of the e-th feature point in the grayscale image
- P 1e (x, y) is the position of the e-th feature point in the legal image
- bias is the average of the relative displacements between all matching points and corresponding legal feature points in the legitimate image
- o is the total number of matching points in the grayscale image, where e ranges from [1, o].
- the average value of the relative displacement between the matching point in the grayscale image and the corresponding legal feature point in the legal image is obtained, and the overlapping area of the grayscale image relative to the legal image is obtained in the abscissa direction by the relative displacement average.
- S81 Determine the overlapping area between the grayscale image and the legal image according to the x-component average and the y-component average.
- the x-axis mean value can be used to obtain the offset distance of the overlapping area of the legal image relative to the grayscale image
- the y-component mean value can be used to obtain the overlapping area of the legal image relative to the grayscale image. The distance offset in the ordinate direction to obtain the overlapping area between the grayscale image and the legal image.
- the length of the grayscale image is 20
- the width is 10
- the bottom left corner of the grayscale image is the origin
- the length direction is the abscissa direction
- the length of the grayscale image is the abscissa 0 to 20
- the width direction is the vertical direction.
- the width in the grayscale direction is 0 to 10 on the ordinate. If the overlapping area of the legal image relative to the grayscale image is shifted to the right by 10 in the abscissa direction and upward by 5 in the ordinate direction, then The overlapping area is between 10 to 20 in the abscissa direction and 5 to 10 in the ordinate direction.
- the overlapping area between the grayscale image and the legal image is obtained according to step S81, and the pixel difference between the pixel values of the grayscale image and the legal image pixel points at the same position is used to calculate the mismatch point according to formula (6).
- N m is the number of mismatched points in the overlapping area
- x and y are the horizontal and vertical coordinates of the pixel points in the legal image and the grayscale image
- the legal image and the grayscale image are in the same coordinate system.
- the bottom left pixel of the image and the bottom left pixel of the grayscale image are on the coordinate origin of the coordinate system
- s is the mean value of the x component of the offset of the legal image relative to the overlapping area of the grayscale image on the abscissa.
- the mean value of the y component of the offset of the overlapping area of the image relative to the grayscale image h is the height of the grayscale image and the legal image, w is the width of the grayscale image and the legal image, and P 1 and P 2 are both
- I (x, y) is the pixel value of the pixel (x, y) in the grayscale image
- R (xs, yt) is the pixel value of the pixel (xs, yt) in the legal image.
- S83 Calculate the degree of mismatch between the grayscale image and the legal image according to the number of mismatched points and the number of pixel points with preset pixel values in the overlapping area.
- the pixel point of the preset pixel value may specifically be a pixel point with a pixel value of 255 in the overlapping area of the grayscale image and a pixel point with a pixel value of 255 in the overlapping area of the legal image.
- step S82 the number of mismatch points between the grayscale image and the legal image in the overlapping area and the number of pixel points with preset pixel values in the overlapping area are obtained, and the grayscale image and the Mismatch between legal images:
- R m is the degree of mismatch between the grayscale image and the legitimate image.
- the mismatch between the grayscale image and the legal image is obtained according to step S83, and it is compared with a preset matching degree threshold. If the mismatch is smaller than the preset matching degree threshold, the finger to be detected is confirmed.
- the vein image matches the legitimate image, that is, the finger vein image to be detected and the legitimate image are homologous finger vein images.
- the specific value of the preset matching degree threshold can be set according to the needs of practical applications, and is not limited here.
- the average value of the x component of the overlapping area of the legal image relative to the overlapping area of the gray image is obtained by calculating the average value of the relative displacement between the matching point in the gray image and the corresponding legal feature point in the legal image.
- the mean value of the y component shifted in the ordinate direction so as to determine the overlapping area between the gray image and the legal image, and calculate the number of mismatch points between the gray image and the legal image in the overlapping area, and then pass the The number of matching points is used to calculate the degree of mismatch between the grayscale image and the legitimate image.
- the degree of mismatch is compared with a preset match degree threshold.
- the Finger vein image matches legitimate image.
- the number of matching points is less than or equal to the threshold number of points, matching is further performed by means of displacement calculation to avoid matching failure due to image quality problems, which can effectively improve the accuracy of matching vein veins on the fingers, and then determine the Whether the finger vein image matches the legal image can realize the legal authentication of the finger vein.
- an image matching device is provided, and the image matching device corresponds to the image matching method in the above embodiment one by one.
- the image matching device includes: an acquisition module 71, a grayscale module 72, a feature point extraction module 73, a vector calculation module 74, a similarity calculation module 75, a recognition module 76, and a first matching module 77.
- the detailed description of each function module is as follows:
- An acquisition module 71 configured to acquire a finger vein image to be detected
- a grayscale module 72 configured to perform grayscale processing on a finger vein image to be detected to obtain a grayscale image
- a feature point extraction module 73 for extracting feature points of finger vein patterns in a grayscale image using a fast algorithm
- a vector calculation module 74 configured to process a feature point by using a sift algorithm to obtain a feature vector of each feature point;
- a similarity calculation module 75 is configured to calculate, for each feature point, a cosine value between a feature vector of the feature point and a feature vector of each legal feature point in a legal image, where the legal image refers to legal authentication in advance Finger vein image
- the identification module 76 is configured to identify a feature point obtained as a matching point if the cosine value is less than a preset cosine threshold;
- the first matching module 77 is configured to confirm that the finger vein image to be detected matches a legitimate image if the number of matching points is greater than a preset threshold of the number of points.
- the grayscale module 72 includes:
- the processing sub-module 722 is configured to perform graying processing on the finger vein image to be detected according to the following formula according to the RGB component values of the pixels:
- x and y are the abscissa and ordinate of each pixel in the finger vein image to be detected
- g (x, y) is the pixel
- (x, y) is the gray value after the graying process
- R (x, y) is the color component of the R channel of the pixel (x, y)
- G (x, y) is the color component of the G channel of the pixel (x, y)
- B (x, y) is the pixel The color component of the B channel of point (x, y).
- the feature point extraction module 73 includes:
- a setting sub-module 731 configured to use pixels in a grayscale image as basic pixels, and obtain a pixel value of each basic pixel;
- a selection sub-module 732 is configured to determine pixels on a circle centered on a basic pixel and having a preset length threshold as a target pixel corresponding to the basic pixel, and select a preset position from the target pixel The target pixel as the comparison pixel;
- the candidate sub-module 733 is configured to calculate, for each basic pixel point, a first pixel difference between a pixel value of the basic pixel point and a pixel value of each comparison pixel point corresponding to the basic pixel point. If a preset number of first pixel differences are greater than a preset first threshold, determining the basic pixel as a candidate pixel;
- a feature submodule 734 configured to calculate, for each candidate pixel point, a second pixel difference between a pixel value of the candidate pixel point and a pixel value of a target pixel point corresponding to the candidate pixel point; if a second preset exists If the number of second pixel differences is greater than the first threshold, the candidate pixel point is determined as a feature point.
- the vector calculation module 74 includes:
- a sampling submodule 741 configured to obtain a sampling area of each feature point
- a rotation sub-module 742 configured to rotate the sampling area in a preset direction to obtain a target image area, and divide the target image area into a preset number of sub-areas;
- the histogram calculation submodule 743 is configured to calculate a gradient direction histogram feature of each sub-region, where the gradient direction histogram feature includes gradient magnitudes in n preset directions, where n is a positive integer;
- a combination sub-module 744 configured to combine the gradient direction histogram features of each sub-region to obtain a basic feature vector of feature points
- a normalization sub-module 745 is configured to perform normalization processing on the basic feature vector to obtain a feature vector.
- the similarity calculation module 75 includes:
- the cosine calculation sub-module 751 is configured to calculate a cosine value between a feature vector of each feature point in a grayscale image and a feature vector of each legitimate feature point in a legitimate image according to the following formula:
- cos ⁇ ij is the cosine value between the feature vector of the i-th feature point in the gray image and the feature vector of the j-th legitimate feature point in the legal image
- a i is the feature of the i-th feature point in the gray image Vector
- b j is the feature vector of the j-th legitimate feature point in the legitimate image
- is the length of a i
- is the length of b j
- i and j are positive integers.
- the image matching device further includes:
- a displacement calculation module 78 is configured to calculate a relative displacement average between the matching point and a corresponding legal feature point in the legal image if the number of matching points is less than or equal to the threshold threshold, wherein the average relative displacement includes a grayscale image and a legal image The mean value of the x component shifted in the abscissa direction, and the mean value of the y component shifted in the ordinate direction;
- An area determining module 79 configured to determine an overlapping area between a grayscale image and a legitimate image according to the x-component average and the y-component average;
- the pixel difference calculation module 710 is configured to calculate the number of unmatched points in the overlapping area according to the pixel difference between the grayscale image and the pixel value of the pixel point of the legal image at the same position;
- a mismatch degree calculation module 711 configured to calculate a degree of mismatch between a grayscale image and a legal image according to the number of mismatched points and the number of pixel points of a preset pixel value in an overlapping area;
- the second matching module 712 is configured to confirm that the finger vein image to be detected matches a legitimate image if the degree of mismatch is less than a preset matching degree threshold.
- Each module in the image matching device can be implemented in whole or in part by software, hardware, and a combination thereof.
- the above-mentioned modules may be embedded in the hardware in or independent of the processor in the computer device, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
- a computer device is provided.
- the computer device may be a server, and its internal structure diagram may be as shown in FIG. 8.
- the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
- the processor of the computer device is used to provide computing and control capabilities.
- the memory of the computer device includes a non-volatile storage medium and an internal memory.
- the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
- the internal memory provides an environment for the operation of the operating system and computer-readable instructions in a non-volatile storage medium.
- the database of the computer equipment is used to store data of finger vein images.
- the network interface of the computer device is used to communicate with an external terminal through a network connection.
- the computer-readable instructions are executed by a processor to implement an image matching method.
- a computer device including a memory, a processor, and computer-readable instructions stored on the memory and executable on the processor.
- the processor executes the computer-readable instructions
- the image matching of the foregoing embodiment is implemented
- the steps of the method are, for example, steps S1 to S7 shown in FIG. 2.
- the processor executes the computer-readable instructions
- the functions of each module / unit of the image matching device in the foregoing embodiment are implemented, for example, the functions of modules 71 to 77 shown in FIG. 7. To avoid repetition, we will not repeat them here.
- one or more non-volatile readable storage media are provided, and computer-readable instructions are stored thereon.
- the computer-readable instructions are executed by one or more processors, the images in the foregoing method embodiments are implemented
- Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
- Volatile memory can include random access memory (RAM) or external cache memory.
- RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Collating Specific Patterns (AREA)
Abstract
一种图像匹配方法、装置、计算机设备及存储介质,图像匹配方法包括:获取待检测的手指静脉图像(S1);对待检测的手指静脉图像进行灰度化处理,得到灰度图像(S2);使用fast算法提取灰度图像中手指静脉纹路的特征点(S3);采用sift算法对特征点进行处理,得到每个特征点的特征向量(S4);针对每个特征点,计算该特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,合法图像是指预先经过合法认证的手指静脉图像(S5);若余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点(S6);若匹配点的数量大于预设的点数阈值,则确认待检测的手指静脉图像与合法图像相匹配(S7)。上述方法实现了对手指静脉纹路的准确识别和匹配,从而有效提高手指静脉纹路匹配的准确性。
Description
本申请以2018年6月11日提交的申请号为201810593681.5,名称为“一种图像匹配方法、装置、计算机设备及存储介质”的中国发明专利申请为基础,并要求其优先权。
本申请涉及计算机技术领域,尤其涉及一种图像匹配方法、装置、计算机设备及存储介质
目前,传统的手指静脉匹配算法匹配效果不理想,存在误识别率且实现复杂程度较大,导致匹配的准确性不高,使得在手指静脉纹路验证的环节中,由于手指静脉纹路匹配的准确性不高,经常存在验证失败情况,降低验证成功率。
发明内容
基于此,有必要针对上述技术问题,提供一种可以提高手指静脉纹路匹配的准确性的图像匹配方法、装置、计算机设备及存储介质。
一种图像匹配方法,包括:
获取待检测的手指静脉图像;
对所述待检测的手指静脉图像进行灰度化处理,得到灰度图像;
使用fast算法提取所述灰度图像中手指静脉纹路的特征点;
采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量;
针对每个所述特征点,计算所述特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,所述合法图像是指预先经过合法认证的手指静脉图像;
若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点;
若所述匹配点的数量大于预设的点数阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
一种图像匹配装置,包括:
获取模块,用于获取待检测的手指静脉图像;
灰度模块,用于对所述待检测的手指静脉图像进行灰度化处理,得到灰度图像;
特征点提取模块,用于使用fast算法提取所述灰度图像中手指静脉纹路的特征点;
向量计算模块,用于采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量;
相似度计算模块,用于针对每个所述特征点,计算所述特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,所述合法图像是指预先经过合法认证的手指静脉图像;
识别模块,用于若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点;
第一匹配模块,用于若所述匹配点的数量大于预设的点数阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
一种计算机设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令,所述处理器执行所述计算机可读指令时实现上述图像匹配方法的步骤。
一个或多个存储有计算机可读指令的非易失性可读存储介质,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行上述图像匹配方法的步骤。
本申请的一个或多个实施例的细节在下面的附图和描述中提出,本申请的其他特征和优点将从说明书、附图以及权利要求变得明显。
为了更清楚地说明本申请实施例的技术方案,下面将对本申请实施例的描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的图像匹配理方法的应用环境示意图;
图2是本申请实施例提供的图像匹配方法的流程图;
图3是本申请实施例提供的图像匹配方法中步骤S2的流程图;
图4是本申请实施例提供的图像匹配方法中步骤S3的流程图;
图5是本申请实施例提供的图像匹配方法中步骤S4的流程图;
图6是本申请实施例提供的图像匹配方法中通过位移计算的方式再次进行匹配的流程图;
图7是本申请实施例提供的图像匹配装置的示意图;
图8是本申请实施例提供的计算机设备的示意图。
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
图1示出了本申请实施例提供的应用环境,该应用环境包括服务端和客户端,其中,服务端和客户端之间通过网络进行连接,客户端用于对手指静脉图像进行采集,并且将采集到的手指静脉图像发送到服务端,客户端具体可以但不限于是摄像机、相机、扫描仪或者带有其他拍照功能的手指静脉图像采集设备;服务端用于对手指静脉图像进行手指静脉纹路匹配,服务端具体可以用独立的服务器或者多个服务器组成的服务器集群实现。本申请实施例提供的图像匹配方法应用于服务端。
在一实施例中,如图2所示,提供一种图像匹配方法,以该方法应用在图1中的服务端为例进行说明,包括如下步骤:
S1:获取待检测的手指静脉图像。
在本申请实施例中,待检测的手指静脉图像是指未经过任何处理,直接从指静脉采集设备中采集到的手指静脉图像。
S2:对待检测的手指静脉图像进行灰度化处理,得到灰度图像。
在本申请实施例中,灰度化是指在RGB模型中,如果R=G=B时,则色彩表示只有一种灰度颜色,其中R=G=B的值叫灰度值,因此,灰度图像每个像素只需一个字节存放灰度值,灰度范围为0-255。
通过对待检测的手指静脉图像进行灰度化处理,将手指静脉图像中所有像素点的像素值都调整到0-255之间,得到手指静脉图像的灰度图像。
S3:使用fast算法提取灰度图像中手指静脉纹路的特征点。
在本申请实施例中,特征点是指能够标识灰度图像中的手指静脉纹路的像素点,通过运用角点检测(Features from accelerated segment test,fast)算法,对步骤S2得到的灰度图像进行处理,提取该灰度图像中手指静脉纹路的像素点作为特征点。
fast算法是一种用于角点检测的算法,该算法的原理是取图像中的检测点,以该检测点为圆心,通过将圆心周围的像素点的像素值与检测点的像素值进行比较,从而判断检测点是否为角点。在本申请实施例中,将fast算法确定的角点作为特征点。
S4:采用sift算法对特征点进行处理,得到每个特征点的特征向量。
在本申请实施例中,特征点的数量可为多个,其中,特征向量是指特征点的向量,即是特征点具有大小和方向的量,根据步骤S3获取的特征点,运用尺度不变特征变换(Scale Invariant Feature Transform,sift)算法,对每个特征点进行处理,获取每个特征点的特征向量。
sift算法是一种尺度不变特征变换算法,主要用于提取图像的局部性特征,通过在不同的尺度空间上的计算得到特征点的特征向量。
S5:针对每个特征点,计算该特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,合法图像是指预先经过合法认证的手指静脉图像。
在本申请实施例中,针对每个特征点,根据灰度图像中特征点的特征向量与合法图像中每个合法特征点的特征向量,计算该特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值。
需要说明的是,合法图像预先存储在合法数据库中,合法图像是指预先经过合法认证的手指静脉图像,主要用来与待检测的手指静脉图像进行匹配。
其中,合法图像特征点、合法图像特征点的特征向量均可以采用与本案相同的方法得到,并且与合法图像对应保存在合法数据库中。
S6:若余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点。
在本申请实施例中,预设的余弦阈值主要用来判断两个特征点之间是否互为匹配点,其具体的取值可以根据实际应用的需要进行设置,此处不做限制。
具体地,根据步骤S5计算得出的余弦值,将其与预设的余弦阈值进行比较,当该余弦值小于预设的余弦阈值时,则确认得到该余弦值的特征点为得到该余弦值的合法特征点的匹配点。
S7:若匹配点的数量大于预设的点数阈值,则确认待检测的手指静脉图像与合法图像相匹配。
在本申请实施例中,预设的点数阈值用于判断两幅图像的手指静脉纹路是否相匹配,其具体的取值可以根据实际应用的需要进行设置,此处不做限制。
具体地,根据步骤S6获取待检测的手指静脉图像与合法图像之间匹配点的数量,将其与预设的点数阈值进行比较,当该匹配点的数量大于预设的点数阈值时,则确认待检测的手指静脉图像与合法图像相匹配,即待检测的手指静脉图像与合法图像是同源手指静脉图像。
需要说明的是,通过判断待检测的手指静脉图像与合法图像是否为同源手指静脉图像,能够实现对手指静脉的合法认证,即当待检测的手指静脉图像与合法图像是同源手指静脉图像时,确认该待检测的手指静脉图像中的手指静脉为合法有效的手指静脉,进而可以为该手指静脉的用户开放相应的权限。
本实施例中,通过对待检测的手指静脉图像进行灰度化处理,得到灰度图像,利用fast算法提取灰度图像中手指静脉纹路的特征点,通过sift算法得到每个特征点的特征向量,并根据特征向量,计算灰度图像中的特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,再根据余弦值判断匹配点的个数,最后将匹配点的个数与预设阈值进行比较,若匹配点的个数大于预设阈值,则确认待检测的手指静脉图像与合法图像相匹配。一方面,采用fast算法提取手指静脉纹路的特征点,利用fast算法的高效性,实现对手指静脉纹路的特征点的快速提取,从而提高对手指静脉纹路特征点的识别效率;另一方面,采用fast算法和sift算法相结合的方式得到特征点的特征向量,并通过计算特征向量的余弦值确定匹配点,实现了对手指静脉纹路的准确匹配,有效提高手指静脉纹路匹配的准确性。
在一实施例中,如图3所示,步骤S2中,即对手指静脉图像进行灰度化处理,得到灰度图像具体包括如下步骤:
S21:对待检测的手指静脉图像中的像素点进行遍历,获取每个像素点的RGB分量值。
具体地,按照预设的遍历方式对待检测的手指静脉图像中的像素点进行遍历,获取每个像素点的RGB分量值,其中,R、G、B分别代表红、绿、蓝三个通道的颜色。
其中,预设的遍历方式具体可以是以待检测的手指静脉图像的左上角像素点为起点,从上往下从左往右的顺序进行逐行遍历,也可以是从待检测的手指静脉图像的中线位置同时向两边遍历,还可以是其他遍历方式,此处不做限制。
S22:根据像素点的RGB分量值,按照公式(1)对待检测的手指静脉图像作灰度化处理:
其中,x和y为待检测的手指静脉图像中每个像素点的横坐标和纵坐标,g(x,y)为像素点(x,y)灰度化处理后的灰度值,R(x,y)为像素点(x,y)的R通道的颜色分量,G(x,y)为像素点(x,y)的G通道 的颜色分量,B(x,y)为像素点(x,y)的B通道的颜色分量。
在本申请实施例中,为了实现对待检测的手指静脉图像中信息内容的准确提取,首先需要对待检测的手指静脉图像进行灰度化处理,通过公式(1)分别对R通道,G通道和B通道的占比进行设置,从而将待检测的手指静脉图像转换为灰度图像。
RGB模型是目前常用的一种彩色信息表达方式,它使用红、绿、蓝三原色的亮度来定量表示颜色。该模型也称为加色混色模型,是以RGB三色光互相叠加来实现混色的方法,因而适合于显示器等发光体的显示。
需要说明的是,在本申请实施例中,通过公式(1)进行分量计算灰度值,在其他实施例中还可以采用加权法、最大值法或者平均值法对图像进行灰度化处理,此处不做限制。
本实施例中,通过遍历待检测的手指静脉图像中的像素点并获取对应像素点的RGB分量值,根据获取到的每个像素点的RGB分量值,利用公式(1)对待检测的手指静脉图像进行灰度化处理,从而实现将图像中像素点的像素值范围设定在0-255之间,进一步减少图像原始数据量,提高在后续处理计算中的计算效率。
在一实施例中,如图4所示,步骤S3中,即使用fast算法提取灰度图像中手指静脉纹路的特征点具体包括如下步骤:
S31:将灰度图像中的像素点作为基本像素点,并获取每个基本像素点的像素值。
具体地,按照预设的遍历方式对步骤S2获取的灰度图像中的像素点进行遍历,将遍历到的像素点作为基本像素点,并且,获取每个基本像素点的像素值。
其中,预设的遍历方式可以是从上往下从左往右的顺序对灰度图像中的像素点进行逐行遍历,也可以是从中间同时向两边对灰度图像中的像素点进行逐行遍历,此处不做限制。
S32:将以基本像素点为中心,以预设长度阈值为半径的圆周上的像素点确定为该基本像素点对应的目标像素点,并从目标像素点中选取预设位置的目标像素点作为比较像素点。
在本申请实施例中,比较像素点是用于与基本像素点进行像素值比较的目标像素点,通过以步骤S31中的基本像素点为中心,以预设长度阈值为半径进行画圆,将该圆的圆周上的所有像素点作为该基本像素点对应的目标像素点,并选取预设位置的目标像素点作为比较像素点。
需要说明的是,预设长度阈值可以以像素点的数量进行取值,其具体取值可以根据实际应用的需要进行设置,此处不做限制。
可以理解的是,预设长度阈值不同,即圆周的半径不同,圆周上的目标像素点的数量也不同,例如,若预设长度阈值为2个像素点,则圆周上将存在8个目标像素点;若预设长度阈值为3个像素点,则其圆周上将存在16个目标像素点。
其中,预设位置可以是以基本像素点为中心的圆周上的垂直方向的两个像素点,也可以是以基本像素点为中心的圆周上的水平方向的两个像素点,还可以是以基本像素点为中心的圆周上的水平方向和垂直方向的四个像素点,其具体的位置可以根据实际应用的需要进行设置,此处不做限制。
S33:针对每个基本像素点,计算该基本像素点的像素值与该基本像素点对应的每个比较像素点的像素值之间的第一像素差值,若存在第一预设数量个第一像素差值大于预设的第一阈值,则将该基本像素点确定为候选像素点。
具体地,针对每个基本像素点,根据步骤S32获得的比较像素点,将基本像素点的像素值与该基本像素点对应的每个比较像素点的像素值之间的绝对差值作为第一像素差值,再将每个第一像素差值与预设的第一阈值进行比较,统计第一像素差值大于第一阈值的数量,若该数量大于或者等于第一预设数量,则确认该基本像素点为候选像素点。
需要说明的是,在确定所有候选像素点之后,将其他的基本像素点都丢弃,即特征点只需在候选像素点中进行进一步筛选得到。
例如,若基本像素点Q对应四个比较像素点Q1、Q2、Q3和Q4,基本像素点Q和比较像素点Q1、Q2、Q3、Q4之间的第一像素差值分别为8、8、5、8,若预设的第一阈值为6,并且第一预设数量为2, 则Q1、Q2和Q4这3个比较像素点与基本像素点Q的第一像素差值大于预设的第一阈值,即第一像素差值大于第一阈值的数量为3,大于第一预设数量,因此可以确认基本像素点Q为候选像素点。
需要说明的是,在计算候选点的过程中,可以优先选取以基本像素点为中心的圆周上的垂直方向的两个比较像素点进行计算,并根据计算结果从基本像素点中筛选出第一候选像素点,再选取以基本像素点为中心的圆周上的水平方向和垂直方向的四个比较像素点,采用相同的计算方式与第一候选像素点进行计算,并根据计算结果从第一候选像素点中再次筛选出候选像素点。通过这种多次逐步筛选的方式排除基本像素点,能够减少对候选像素点进行筛选的计算量,提高计算效率。
需要说明的是,第一预设数量和第一阈值均可根据实际应用的需要进行设置,此处不做限制。
S34:针对每个候选像素点,计算该候选像素点的像素值与该候选像素点对应的目标像素点的像素值之间的第二像素差值,若存在第二预设数量个第二像素差值大于第一阈值,则将该候选像素点确定为特征点。
具体地,针对每个候选像素点,根据步骤S32获得的目标像素点,将候选像素点的像素值与该候选像素点对应的每个目标像素点的像素值之间的绝对差值作为第二像素差值,再将每个第二像素差值与预设的第一阈值进行比较,统计第二像素差值大于第一阈值的数量,若该数量大于或者等于第二预设数量,则确认该候选像素点为特征点。
进一步地,一个邻域内可以只保留一个特征点,当以特征点为中心的邻域内存在多个特征点时,可以分别计算各个特征点的像素值与其对应的所有目标像素点的像素值之间的绝对差值的总和,选取绝对差值的总和最大的特征点进行保留,其余的丢弃。其中,邻域是指以特征点为中心的M*M个像素点构成的区域,M为大于1的正整数,M的具体取值可根据实际应用的需要进行设置,此处不做限制。
本实施例中,通过获取基本像素点,获取基本像素点对应的目标像素点,以及获取比较像素点,针对每个基本像素点,通过计算基本像素点的像素值与该基本像素点对应的每个比较像素点的像素值之间的像素差值,确定候选像素点,再针对每个候选像素点,计算候选像素点的像素值与该候选像素点对应的目标像素点的像素值之间的第二像素差值,从而确定特征点。通过先确定候选像素点,可以有效排除部分非特征点,再利用候选像素点的像素值与其对应的目标像素点的像素值进行比较,能够减少计算量,从而能够提高对特征点的计算效率,实现对特征点的快速提取。
在一实施例中,如图5所示,步骤S4中,即采用sift算法对特征点进行处理,得到每个特征点的特征向量具体包括如下步骤:
S41:获取每个特征点的采样区域。
具体地,根据步骤S3获取到的每个特征点,以每个特征点为中心,选取B*B规格的区域作为每个特征点的采样区域,其中,规格标识了图像区域的大小,B为像素点的个数,且其具体的取值可以根据实际应用的需要进行设置,此处不做限制。
S42:将采样区域按照预设方向进行旋转,得到目标图像区域,并将目标图像区域划分成预设个数的子区域。
具体地,按照公式(2)将采样区域按照预设方向进行旋转:
其中,x和y分别为采样区域的原始x轴和y轴,α为旋转的方向,x'和y'分别为采样区域按照预设方向旋转后的x'轴和y'轴。
将采样区域按照公式(2)进行旋转后得到的图像区域确定为目标图像区域,并对该目标图像区域按照等间隔的方式进行划分,即将B*B规格的区域等间隔划分成预设个数的子区域,即划分成BP*BP个子区域。
需要说明的是,预设方向具体可以根据实际应用的需要进行设置,此处不做限制。预设个数为大于 1的正整数。
S43:计算每个子区域的梯度方向直方图特征,其中,梯度方向直方图特征包括n个预设方向上的梯度幅值,n为正整数。
具体地,按照公式(3)计算每个子区域中像素点的梯度方向及梯度幅值:
其中,x和y为每个子区域中每个像素点的横坐标和纵坐标,V(x,y)为像素点(x,y)在水平方向上的梯度大小,H(x,y)为像素点(x,y)在垂直方向上的梯度大小,β(x,y)为像素点(x,y)的梯度方向,m(x,y)为像素点(x,y)的梯度幅值。
在每个子区域中,根据公式(3)得到该子区域中每个像素点的梯度方向和梯度幅值,将每个像素点的梯度方向按照预设的方向范围映射到对应的某一个预设方向上。例如,若像素点E的梯度方向为30°,像素点F的梯度方向为80°,假设预设方向包括45°和90°,并且映射到45°预设方向上的方向范围为(22.5°,67.5°],映射到90°预设方向上的方向范围为(67.5°,112.5°],则将像素点E映射到45°的预设方向上,将像素点F映射到90°的预设方向上。
待每个像素点的梯度方向都映射到对应的预设方向上后,将该子区域中每个像素点同一预设方向的梯度幅值进行累加,作为该预设方向的梯度强度信息,即梯度幅值,并以所有预设方向的起始点作为同一起始点进行组合,得到每个子区域的梯度方向直方图特征,该梯度方向直方图特征包括n个预设方向和每个预设方向上的梯度强度信息。
例如,假设n个预设方向包括0°、90°、180°和270°共4个预设方向,若一子区域中存在五个像素点Q1、Q2、Q3、Q4和Q5,对应的预设方向分别为0°、0°、90°、180°和270°,对应的梯度幅值分别为1、2、1、1和1,则将同一预设方向的梯度幅值进行累加,即将像素点Q1和像素点Q2的梯度幅值进行累加,再将所有预设方向的起始点作为同一起始点进行组合后,最后将得到一个具有0°、90°、180°和270°这四个预设方向,且每个预设方向的梯度幅值分别为3、1、1、1的平面向量,该平面向量即为梯度方向直方图特征。
S44:将每个子区域的梯度方向直方图特征进行组合,得到特征点的基础特征向量。
具体地,将每个子区域内的梯度方向直方图特征进行组合,由于每个子区域包含n个方向的梯度强度信息,且存在Bp*Bp个子区域,通过组合之后将形成BP*BP*n维的特征向量,该特征向量即为特征点的基础特征向量。
例如,若特征点的采样区域存在4个子区域,且每个子区域中都存在8个方向的梯度强度信息,通过组合之后将得到32维的特征向量,即该32维的特征向量为该特征点的基础特征向量。
S45:对基础特征向量进行归一化处理,得到特征向量。
在本申请实施例中,归一化处理方式可以是将每个基础特征向量除以所有基础特征向量中的最大值,也可以将每个基础特征向量除以所有基础特征向量的均值。通过对基础特征向量进行归一化处理,得到标准的特征向量,方便对数据进行处理。
进一步地,在归一化处理后,还可以对特征向量的长度进行截断处理,若特征向量的长度大于预设长度,则只取预设长度作为特征向量的长度,以此提高数据处理的鉴别性。其中,预设长度可以根据实际应用的需要进行设置,此处不做限制。
例如,若预设长度为0.2,特征向量的长度为0.5,通过对该特征向量进行截断处理后,该特征向量的长度为0.2。
本实施例中,通过对每个特征点的采样区域进行旋转得到目标图像区域,进一步将目标图像区域划分成预设个数的子区域,并计算每个子区域的梯度方向直方图特征,再将每个子区域的梯度方向直方图 特征进行组合后获取特征点的基础特征向量,最后对基础特征向量进行归一化处理,获取特征向量。一方面,通过将目标图像区域划分成多个子区域,并对每个子区域的计算不同梯度方向的梯度幅值,能够准确提取子区域的特征,进而对每个子区域的梯度方向直方图特征进行组合后得到的基础特征向量能够准确的反应特征点的特征,从而能够根据该基础特征向量对特征点进行分析,提高特征点特征向量的精度;另一方面,通过对基础特征向量进行归一化,能够保证数据处理的鉴别性,从而实现对特征点的特征向量的准确提取。
在一实施例中,步骤S5中,即针对每个特征点,计算该特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值具体包括如下步骤:
按照公式(4)计算灰度图像中每个特征点的特征向量与合法图像中每个合法特征点的特征向量之间的余弦值:
其中,cosθ
ij为灰度图像中第i个特征点的特征向量与合法图像中第j个合法特征点的特征向量之间的余弦值,a
i为灰度图像中第i个特征点的特征向量,b
j为合法图像中第j个合法特征点的特征向量,|a
i|为a
i的长度,|b
j|为b
j的长度,i和j均为正整数。
具体地,根据灰度图像中每个特征点的特征向量与合法图像中每个合法特征点的特征向量,按照公式(4)计算手指静脉图像中每个特征点的特征向量与合法图像中每个合法特征点的特征向量之间的余弦值。
本实施例中,利用公式(4)能够快速准确地计算出每个特征点的特征向量与合法图像中每个合法特征点的特征向量之间的余弦值,利用该余弦值准确判断特征点与合法特征点之间的特征相似程度,进而准确识别匹配点,提高手指静脉纹路匹配的准确性,从而在对手指静脉图像进行合法认真时,有效提高认证的准确性。
在一实施例中,如图6所示,步骤S6之后,该图像匹配方法还包括如下步骤:
S80:若匹配点的数量小于或者等于点数阈值,则计算匹配点与合法图像中对应的合法特征点之间的相对位移均值,其中,相对位移均值包括灰度图像相对合法图像在横坐标方向上偏移的x分量均值,以及在纵坐标方向上偏移的y分量均值。
在本申请实施例中,当匹配点的数量小于或者等于点数阈值时,为了避免由于图像质量问题等导致匹配失败,可以进一步通过相对位移再一次判断手指静脉图像与合法图像的匹配度。
具体地,按照公式(5)计算匹配点与合法图像中对应的合法特征点之间的相对位移均值:
其中,x和y为合法图像中每个合法特征点和灰度图像中每个特征点的横坐标和纵坐标,P
e(x,y)为灰度图像中的第e个匹配点与合法图像中对应的合法特征点点之间的相对位移,P
2e(x,y)为灰度图像中第e个特征点的位置,P
1e(x,y)为合法图像中第e个特征点的位置,bias为所有匹配点与合法图像中对应的合法特征点之间的相对位移均值,o为灰度图像中匹配点的总数,其中,e的取值范围为[1,o]。
根据公式(5)得出灰度图像中的匹配点与合法图像中对应的合法特征点之间的相对位移均值,通过该相对位移均值获取灰度图像相对合法图像的重叠区域在横坐标方向上偏移的x分量均值,以及在纵坐标方向上偏移的y分量均值。
S81:根据x分量均值和y分量均值,确定灰度图像与合法图像之间的重叠区域。
在本申请实施例中,通过x分量均值可以获取合法图像相对于灰度图像的重叠区域在横坐标方向上偏移的距离,通过y分量均值可以获取合法图像相对于灰度图像的重叠区域在纵坐标方向上偏移的距离,进而获取灰度图像与合法图像之间的重叠区域。
例如,若灰度图像的长为20,宽为10,以灰度图像左下角为原点,以长度方向为横坐标方向,则灰度图像的长度为横坐标0到20,以宽度方向为纵坐标方向,则灰度方向的宽度为纵坐标0到10,若合法图像相对灰度图像的重叠区域在横坐标方向上向右偏移了10,在纵坐标方向上向上偏移了5,则重叠区域在横坐标方向上为横坐标10到20之间,在纵坐标方向上为纵坐标5到10之间。
S82:在重叠区域中,根据灰度图像与合法图像在同一位置像素点的像素值之间的像素差值,计算不匹配点的个数。
具体的,根据步骤S81得到灰度图像与合法图像之间的重叠区域,利用灰度图像与合法图像在同一位置像素点的像素值之间的像素差值,按照公式(6)计算不匹配点的个数:
其中,N
m为重叠区域内不匹配点的个数,x和y为合法图像和灰度图像中像素点的横坐标和纵坐标,且合法图像和灰度图像都在同一坐标系中,合法图像的左下角像素和灰度图像的左下角像素点均在坐标系的坐标原点上,s为合法图像相对灰度图像的重叠区域在横坐标上的偏移的x分量的均值,t为合法图像相对灰度图像的重叠区域在纵坐标上的偏移的y分量的均值,h为灰度图像与合法图像的高度,w为灰度图像与合法图像的宽度,P
1和P
2都为像素点的像素值,I(x,y)为灰度图像中像素点(x,y)的像素值,R(x-s,y-t)为合法图像中像素点(x-s,y-t)的像素值。
当灰度图像与合法图像在同一位置像素点的像素值的绝对值相差255时,即一个为白色像素点,一个为黑色像素点,则这两个像素点为不匹配的像素点,利用公式(6),计算得出灰度图像与合法图像在重叠区域内不匹配点的个数。
S83:根据不匹配点的个数和重叠区域中预设像素值的像素点的个数,计算灰度图像与合法图像之间的不匹配度。
在本申请实施例中,预设像素值的像素点具体可以为灰度图像在重叠区域中像素值为255的像素点,以及合法图像在重叠区域中像素值为255的像素点。
具体地,根据步骤S82获取灰度图像与合法图像在重叠区域内不匹配点的个数,以及在重叠区域中预设像素值的像素点的个数,按照公式(7)计算灰度图像与合法图像之间的不匹配度:
其中,R
m为灰度图像与合法图像的不匹配度。
S84:若不匹配度小于预设的匹配度阈值,则确认待检测的手指静脉图像与合法图像相匹配。
具体地,根据步骤S83获取灰度图像与合法图像的不匹配度,将其与预设的匹配度阈值进行比较,若该不匹配度小于预设的匹配度阈值时,则确认待检测的手指静脉图像与合法图像相匹配,即待检测的 手指静脉图像与合法图像是同源手指静脉图像。
需要说明的是,预设的匹配度阈值具体的取值可以根据实际应用的需要进行设置,此处不做限制。
本实施例中,通过计算灰度图像中匹配点与合法图像中对应的合法特征点之间的相对位移均值,获取合法图像相对灰度图像的重叠区域在横坐标方向上偏移的x分量均值,以及在纵坐标方向上偏移的y分量均值,从而确定灰度图像与合法图像之间的重叠区域,并在重叠区域内计算灰度图像与合法图像不匹配点的个数,再通过不匹配点的个数计算灰度图像与合法图像之间的不匹配度,最后将不匹配度与预设的匹配度阈值进行比较,若不匹配度小于预设匹配度阈值,则确认待检测的手指静脉图像与合法图像相匹配。当匹配点的数量小于或者等于点数阈值时,进一步通过位移计算的方式再次进行匹配,避免由于图像质量问题导致匹配失败,从而能够有效提高对手指静脉纹路匹配的准确性,进而通过判断待检测的手指静脉图像与合法图像是否相匹配,能够实现对手指静脉的合法认证。
应理解,上述实施例中各步骤的序号的大小并不意味着执行顺序的先后,各过程的执行顺序应以其功能和内在逻辑确定,而不应对本申请实施例的实施过程构成任何限定。
在一实施例中,提供一种图像匹配装置,该图像匹配装置与上述实施例中图像匹配方法一一对应。如图7所示,该图像匹配装置包括:获取模块71、灰度模块72、特征点提取模块73、向量计算模块74、相似度计算模块75、识别模块76和第一匹配模块77。各功能模块详细说明如下:
获取模块71,用于获取待检测的手指静脉图像;
灰度模块72,用于对待检测的手指静脉图像进行灰度化处理,得到灰度图像;
特征点提取模块73,用于使用fast算法提取灰度图像中手指静脉纹路的特征点;
向量计算模块74,用于采用sift算法对特征点进行处理,得到每个特征点的特征向量;
相似度计算模块75,用于针对每个特征点,计算该特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,合法图像是指预先经过合法认证的手指静脉图像;
识别模块76,用于若余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点;
第一匹配模块77,用于若匹配点的数量大于预设的点数阈值,则确认待检测的手指静脉图像与合法图像相匹配。
进一步地,灰度模块72包括:
遍历子模块721,用于对待检测的手指静脉图像中的像素点进行遍历,获取每个像素点的RGB分量值;
处理子模块722,用于根据像素点的RGB分量值,按照如下公式对待检测的手指静脉图像作灰度化处理:
其中,x和y为待检测的手指静脉图像中每个像素点的横坐标和纵坐标,g(x,y)为像素点,(x,y)为灰度化处理后的灰度值,R(x,y)为像素点(x,y)的R通道的颜色分量,G(x,y)为像素点(x,y)的G通道的颜色分量,B(x,y)为像素点(x,y)的B通道的颜色分量。
进一步地,特征点提取模块73包括:
设定子模块731,用于将灰度图像中的像素点作为基本像素点,并获取每个基本像素点的像素值;
选取子模块732,用于将以基本像素点为中心,以预设长度阈值为半径的圆周上的像素点确定为该基本像素点对应的目标像素点,并从目标像素点中选取预设位置的目标像素点作为比较像素点;
候选子模块733,用于针对每个基本像素点,计算该基本像素点的像素值与该基本像素点对应的每个比较像素点的像素值之间的第一像素差值,若存在第一预设数量个第一像素差值大于预设的第一阈值,则将该基本像素点确定为候选像素点;
特征子模块734,用于针对每个候选像素点,计算该候选像素点的像素值与该候选像素点对应的目 标像素点的像素值之间的第二像素差值,若存在第二预设数量个第二像素差值大于第一阈值,则将该候选像素点确定为特征点。
进一步地,向量计算模块74包括:
采样子模块741,用于获取每个特征点的采样区域;
旋转子模块742,用于将采样区域按照预设方向进行旋转,得到目标图像区域,并将目标图像区域划分成预设个数的子区域;
直方图计算子模块743,用于计算每个子区域的梯度方向直方图特征,其中,梯度方向直方图特征包括n个预设方向上的梯度幅值,n为正整数;
组合子模块744,用于将每个子区域的梯度方向直方图特征进行组合,得到特征点的基础特征向量;
归一子模块745,用于对基础特征向量进行归一化处理,得到特征向量。
进一步地,相似度计算模块75包括:
余弦计算子模块751,用于按照如下公式计算灰度图像中每个特征点的特征向量与合法图像中每个合法特征点的特征向量之间的余弦值:
其中,cosθ
ij为灰度图像中第i个特征点的特征向量与合法图像中第j个合法特征点的特征向量之间的余弦值,a
i为灰度图像中第i个特征点的特征向量,b
j为合法图像中第j个合法特征点的特征向量,|a
i|为a
i的长度,|b
j|为b
j的长度,i和j均为正整数。
进一步地,该图像匹配装置还包括:
位移计算模块78,用于若匹配点的数量小于或者等于点数阈值,则计算匹配点与合法图像中对应的合法特征点之间的相对位移均值,其中,相对位移均值包括灰度图像相对合法图像在横坐标方向上偏移的x分量均值,以及在纵坐标方向上偏移的y分量均值;
区域确定模块79,用于根据x分量均值和y分量均值,确定灰度图像与合法图像之间的重叠区域;
像素差计算模块710,用于在重叠区域中,根据灰度图像与合法图像在同一位置像素点的像素值之间的像素差值,计算不匹配点的个数;
不匹配度计算模块711,用于根据不匹配点的个数和重叠区域中预设像素值的像素点的个数,计算灰度图像与合法图像之间的不匹配度;
第二匹配模块712,用于若不匹配度小于预设的匹配度阈值,则确认待检测的手指静脉图像与合法图像相匹配。
关于图像匹配装置的具体限定可以参见上文中对于图像匹配方法的限定,在此不再赘述。上述图像匹配装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中,也可以以软件形式存储于计算机设备中的存储器中,以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中,提供了一种计算机设备,该计算机设备可以是服务器,其内部结构图可以如图8所示。该计算机设备包括通过***总线连接的处理器、存储器、网络接口和数据库。其中,该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作***、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作***和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储手指静脉图像的数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种图像匹配方法。
在一个实施例中,提供了一种计算机设备,包括存储器、处理器及存储在存储器上并可在处理器上 运行的计算机可读指令,处理器执行计算机可读指令时实现上述实施例图像匹配方法的步骤,例如图2所示的步骤S1至步骤S7。或者,处理器执行计算机可读指令时实现上述实施例中图像匹配装置的各模块/单元的功能,例如图7所示模块71至模块77的功能。为避免重复,这里不再赘述。
在一个实施例中,提供了一个或多个非易失性可读存储介质,其上存储有计算机可读指令,计算机可读指令被一个或多个处理器执行时实现上述方法实施例中图像匹配方法,或者,该计算机可读指令被一个或多个处理器执行时实现上述装置实施例中图像匹配装置中各模块的功能。为避免重复,这里不再赘述。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机可读指令来指令相关的硬件来完成,所述的计算机可读指令可存储于一一个或多个非易失性可读取存储介质中,该计算机可读指令在执行时,可包括如上述各方法的实施例的流程。其中,本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限,RAM以多种形式可得,诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
所属领域的技术人员可以清楚地了解到,为了描述的方便和简洁,仅以上述各功能单元、模块的划分进行举例说明,实际应用中,可以根据需要而将上述功能分配由不同的功能单元、模块完成,即将所述装置的内部结构划分成不同的功能单元或模块,以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围,均应包含在本申请的保护范围之内。
Claims (20)
- 一种图像匹配方法,其特征在于,所述图像匹配方法包括:获取待检测的手指静脉图像;对所述待检测的手指静脉图像进行灰度化处理,得到灰度图像;使用fast算法提取所述灰度图像中手指静脉纹路的特征点;采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量;针对每个所述特征点,计算所述特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,所述合法图像是指预先经过合法认证的手指静脉图像;若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点;若所述匹配点的数量大于预设的点数阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
- 如权利要求1所述的图像匹配方法,其特征在于,所述使用fast算法提取所述灰度图像中手指静脉纹路的特征点包括:将所述灰度图像中的像素点作为基本像素点,并获取每个所述基本像素点的像素值;将以所述基本像素点为中心,以预设长度阈值为半径的圆周上的像素点确定为所述基本像素点对应的目标像素点,并从所述目标像素点中选取预设位置的目标像素点作为比较像素点;针对每个所述基本像素点,计算所述基本像素点的像素值与所述基本像素点对应的每个所述比较像素点的像素值之间的第一像素差值,若存在第一预设数量个所述第一像素差值大于预设的第一阈值,则将所述基本像素点确定为候选像素点;针对每个所述候选像素点,计算所述候选像素点的像素值与所述候选像素点对应的所述目标像素点的像素值之间的第二像素差值,若存在第二预设数量个所述第二像素差值大于所述第一阈值,则将所述候选像素点确定为所述特征点。
- 如权利要求1所述的图像匹配方法,其特征在于,所述采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量包括:获取每个所述特征点的采样区域;将所述采样区域按照预设方向进行旋转,得到目标图像区域,并将所述目标图像区域划分成预设个数的子区域;计算每个所述子区域的梯度方向直方图特征,其中,所述梯度方向直方图特征包括n个预设方向上的梯度幅值,n为正整数;将每个所述子区域的所述梯度方向直方图特征进行组合,得到所述特征点的基础特征向量;对所述基础特征向量进行归一化处理,得到所述特征向量。
- 如权利要求1至5中任意一项所述的图像匹配方法,其特征在于,所述若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点之后,所述图像匹配方法还包括:若所述匹配点的数量小于或者等于所述点数阈值,则计算所述匹配点与所述合法图像中对应的合法特征点之间的相对位移均值,其中,所述相对位移均值包括所述灰度图像相对所述合法图像在横坐标方向上偏移的x分量均值,以及在纵坐标方向上偏移的y分量均值;根据所述x分量均值和所述y分量均值,确定所述灰度图像与所述合法图像之间的重叠区域;在所述重叠区域中,根据所述灰度图像与所述合法图像在同一位置像素点的像素值之间的像素差值,计算不匹配点的个数;根据所述不匹配点的个数和所述重叠区域中预设像素值的像素点的个数,计算所述灰度图像与所述合法图像之间的不匹配度;若所述不匹配度小于预设的匹配度阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
- 一种图像匹配装置,其特征在于,所述图像匹配装置包括:获取模块,用于获取待检测的手指静脉图像;灰度模块,用于对所述待检测的手指静脉图像进行灰度化处理,得到灰度图像;特征点提取模块,用于使用fast算法提取所述灰度图像中手指静脉纹路的特征点;向量计算模块,用于采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量;相似度计算模块,用于针对每个所述特征点,计算所述特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,所述合法图像是指预先经过合法认证的手指静脉图像;识别模块,用于若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点;第一匹配模块,用于若所述匹配点的数量大于预设的点数阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
- 一种计算机设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令,其特征在于,所述处理器执行所述计算机可读指令时实现如下步骤:获取待检测的手指静脉图像;对所述待检测的手指静脉图像进行灰度化处理,得到灰度图像;使用fast算法提取所述灰度图像中手指静脉纹路的特征点;采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量;针对每个所述特征点,计算所述特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,所述合法图像是指预先经过合法认证的手指静脉图像;若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点;若所述匹配点的数量大于预设的点数阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
- 如权利要求9所述的计算机设备,其特征在于,所述使用fast算法提取所述灰度图像中手指静脉纹路的特征点包括:将所述灰度图像中的像素点作为基本像素点,并获取每个所述基本像素点的像素值;将以所述基本像素点为中心,以预设长度阈值为半径的圆周上的像素点确定为所述基本像素点对应的目标像素点,并从所述目标像素点中选取预设位置的目标像素点作为比较像素点;针对每个所述基本像素点,计算所述基本像素点的像素值与所述基本像素点对应的每个所述比较像素点的像素值之间的第一像素差值,若存在第一预设数量个所述第一像素差值大于预设的第一阈值,则将所述基本像素点确定为候选像素点;针对每个所述候选像素点,计算所述候选像素点的像素值与所述候选像素点对应的所述目标像素点的像素值之间的第二像素差值,若存在第二预设数量个所述第二像素差值大于所述第一阈值,则将所述候选像素点确定为所述特征点。
- 如权利要求9所述的计算机设备,其特征在于,所述采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量包括:获取每个所述特征点的采样区域;将所述采样区域按照预设方向进行旋转,得到目标图像区域,并将所述目标图像区域划分成预设个数的子区域;计算每个所述子区域的梯度方向直方图特征,其中,所述梯度方向直方图特征包括n个预设方向上的梯度幅值,n为正整数;将每个所述子区域的所述梯度方向直方图特征进行组合,得到所述特征点的基础特征向量;对所述基础特征向量进行归一化处理,得到所述特征向量。
- 如权利要求9至13中任意一项所述的计算机设备,其特征在于,所述若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点之后,所述处理器执行所述计算机可读指令时还实现如下步骤:若所述匹配点的数量小于或者等于所述点数阈值,则计算所述匹配点与所述合法图像中对应的合法特征点之间的相对位移均值,其中,所述相对位移均值包括所述灰度图像相对所述合法图像在横坐标方向上偏移的x分量均值,以及在纵坐标方向上偏移的y分量均值;根据所述x分量均值和所述y分量均值,确定所述灰度图像与所述合法图像之间的重叠区域;在所述重叠区域中,根据所述灰度图像与所述合法图像在同一位置像素点的像素值之间的像素差值,计算不匹配点的个数;根据所述不匹配点的个数和所述重叠区域中预设像素值的像素点的个数,计算所述灰度图像与所述合法图像之间的不匹配度;若所述不匹配度小于预设的匹配度阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
- 一个或多个存储有计算机可读指令的非易失性可读存储介质,其特征在于,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行如下步骤:获取待检测的手指静脉图像;对所述待检测的手指静脉图像进行灰度化处理,得到灰度图像;使用fast算法提取所述灰度图像中手指静脉纹路的特征点;采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量;针对每个所述特征点,计算所述特征点的特征向量与合法图像中的每个合法特征点的特征向量之间的余弦值,其中,所述合法图像是指预先经过合法认证的手指静脉图像;若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点;若所述匹配点的数量大于预设的点数阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
- 如权利要求15所述的非易失性可读存储介质,其特征在于,所述使用fast算法提取所述灰度图像中手指静脉纹路的特征点包括:将所述灰度图像中的像素点作为基本像素点,并获取每个所述基本像素点的像素值;将以所述基本像素点为中心,以预设长度阈值为半径的圆周上的像素点确定为所述基本像素点对应的目标像素点,并从所述目标像素点中选取预设位置的目标像素点作为比较像素点;针对每个所述基本像素点,计算所述基本像素点的像素值与所述基本像素点对应的每个所述比较像素点的像素值之间的第一像素差值,若存在第一预设数量个所述第一像素差值大于预设的第一阈值,则将所述基本像素点确定为候选像素点;针对每个所述候选像素点,计算所述候选像素点的像素值与所述候选像素点对应的所述目标像素点的像素值之间的第二像素差值,若存在第二预设数量个所述第二像素差值大于所述第一阈值,则将所述候选像素点确定为所述特征点。
- 如权利要求15所述的非易失性可读存储介质,其特征在于,所述采用sift算法对所述特征点进行处理,得到每个所述特征点的特征向量包括:获取每个所述特征点的采样区域;将所述采样区域按照预设方向进行旋转,得到目标图像区域,并将所述目标图像区域划分成预设个数的子区域;计算每个所述子区域的梯度方向直方图特征,其中,所述梯度方向直方图特征包括n个预设方向上的梯度幅值,n为正整数;将每个所述子区域的所述梯度方向直方图特征进行组合,得到所述特征点的基础特征向量;对所述基础特征向量进行归一化处理,得到所述特征向量。
- 如权利要求15至19中任意一项所述的非易失性可读存储介质,其特征在于,所述若所述余弦值小于预设的余弦阈值,则将得到该余弦值的特征点确认为匹配点之后,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器还执行如下步骤:若所述匹配点的数量小于或者等于所述点数阈值,则计算所述匹配点与所述合法图像中对应的合法特征点之间的相对位移均值,其中,所述相对位移均值包括所述灰度图像相对所述合法图像在横坐标方向上偏移的x分量均值,以及在纵坐标方向上偏移的y分量均值;根据所述x分量均值和所述y分量均值,确定所述灰度图像与所述合法图像之间的重叠区域;在所述重叠区域中,根据所述灰度图像与所述合法图像在同一位置像素点的像素值之间的像素差 值,计算不匹配点的个数;根据所述不匹配点的个数和所述重叠区域中预设像素值的像素点的个数,计算所述灰度图像与所述合法图像之间的不匹配度;若所述不匹配度小于预设的匹配度阈值,则确认所述待检测的手指静脉图像与所述合法图像相匹配。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810593681.5 | 2018-06-11 | ||
CN201810593681.5A CN109101867A (zh) | 2018-06-11 | 2018-06-11 | 一种图像匹配方法、装置、计算机设备及存储介质 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019237520A1 true WO2019237520A1 (zh) | 2019-12-19 |
Family
ID=64796799
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2018/104050 WO2019237520A1 (zh) | 2018-06-11 | 2018-09-05 | 一种图像匹配方法、装置、计算机设备及存储介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109101867A (zh) |
WO (1) | WO2019237520A1 (zh) |
Cited By (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111223134A (zh) * | 2020-01-19 | 2020-06-02 | 北京航空航天大学 | 直线特征的特征向量构建方法、特征匹配方法和计算设备 |
CN111445402A (zh) * | 2020-03-20 | 2020-07-24 | 上海中通吉网络技术有限公司 | 一种图像去噪方法及装置 |
CN111462074A (zh) * | 2020-03-31 | 2020-07-28 | 珠海格力智能装备有限公司 | 轴承外观检测方法、装置、***、计算机设备及存储介质 |
CN111695498A (zh) * | 2020-06-10 | 2020-09-22 | 西南林业大学 | 一种木材同一性检测方法 |
CN111815578A (zh) * | 2020-06-23 | 2020-10-23 | 浙江大华技术股份有限公司 | 图像条纹检测方法、视频监控***及相关装置 |
CN112084883A (zh) * | 2020-08-18 | 2020-12-15 | 珠海格力电器股份有限公司 | 指静脉采集方法、装置、指静脉识别设备及存储介质 |
CN112164032A (zh) * | 2020-09-14 | 2021-01-01 | 浙江华睿科技有限公司 | 一种点胶方法、装置、电子设备及存储介质 |
CN112308886A (zh) * | 2020-09-27 | 2021-02-02 | 深圳疆程技术有限公司 | 一种确定hud图像尺寸的方法、装置及*** |
CN112381105A (zh) * | 2020-11-16 | 2021-02-19 | 平安科技(深圳)有限公司 | 滑块校验的自动化处理方法、装置、设备及介质 |
CN112686302A (zh) * | 2020-12-29 | 2021-04-20 | 科大讯飞股份有限公司 | 图像特征点匹配方法、计算机设备及存储装置 |
CN112883827A (zh) * | 2021-01-28 | 2021-06-01 | 腾讯科技(深圳)有限公司 | 图像中指定目标的识别方法、装置、电子设备和存储介质 |
CN112950540A (zh) * | 2021-02-01 | 2021-06-11 | 联宝(合肥)电子科技有限公司 | 一种条码识别方法及设备 |
CN113033550A (zh) * | 2021-03-15 | 2021-06-25 | 合肥联宝信息技术有限公司 | 一种图像的检测方法、装置及计算机可读介质 |
CN113592921A (zh) * | 2021-06-30 | 2021-11-02 | 北京旷视科技有限公司 | 图像匹配方法,装置,电子设备及存储介质 |
CN113627446A (zh) * | 2021-08-18 | 2021-11-09 | 成都工业学院 | 基于梯度向量的特征点描述算子的图像匹配方法及*** |
CN113658261A (zh) * | 2020-05-12 | 2021-11-16 | 武汉Tcl集团工业研究院有限公司 | 棋盘格角点位置检测方法、装置、智能终端及存储介质 |
CN114217756A (zh) * | 2021-11-26 | 2022-03-22 | 海信电子科技(武汉)有限公司 | 显示设备、图像检测方法及装置 |
CN115184380A (zh) * | 2022-08-01 | 2022-10-14 | 国网上海市电力公司 | 基于机器视觉的印刷电路板焊点异常检测方法 |
CN115314634A (zh) * | 2022-06-28 | 2022-11-08 | 上海艾为电子技术股份有限公司 | 防抖检测方法、装置、终端设备和可读存储介质 |
CN115830719A (zh) * | 2023-02-16 | 2023-03-21 | 青岛旭华建设集团有限公司 | 基于图像处理的建筑工地危险行为识别方法 |
CN116310448A (zh) * | 2023-05-24 | 2023-06-23 | 山东曙岳车辆有限公司 | 基于计算机视觉的集装箱拼装匹配性检测方法 |
CN116452586A (zh) * | 2023-06-15 | 2023-07-18 | 山东飞宏工程机械有限公司 | 一种隧道小导管余料自动对焊质量检测*** |
CN116612306A (zh) * | 2023-07-17 | 2023-08-18 | 山东顺发重工有限公司 | 基于计算机视觉的法兰盘智能对位方法及*** |
CN116703909A (zh) * | 2023-08-07 | 2023-09-05 | 威海海泰电子有限公司 | 一种电源适配器生产质量智能检测方法 |
CN116993643A (zh) * | 2023-09-27 | 2023-11-03 | 山东建筑大学 | 基于人工智能的无人机摄影测量图像校正方法 |
CN117292142A (zh) * | 2023-09-26 | 2023-12-26 | 北京鹰之眼智能健康科技有限公司 | 一种图像的匹配方法 |
CN117495852A (zh) * | 2023-12-29 | 2024-02-02 | 天津中荣印刷科技有限公司 | 一种基于图像分析的数字印刷质量检测方法 |
Families Citing this family (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109977909B (zh) * | 2019-04-04 | 2021-04-20 | 山东财经大学 | 基于细节点区域匹配的手指静脉识别方法及*** |
CN110070564B (zh) * | 2019-05-08 | 2021-05-11 | 广州市百果园信息技术有限公司 | 一种特征点匹配方法、装置、设备及存储介质 |
CN110188815B (zh) * | 2019-05-24 | 2023-09-01 | 广州市百果园信息技术有限公司 | 一种特征点采样方法、装置、设备及存储介质 |
CN110348289B (zh) * | 2019-05-27 | 2023-04-07 | 广州中国科学院先进技术研究所 | 一种基于二值图的手指静脉识别方法 |
CN110162953A (zh) * | 2019-05-31 | 2019-08-23 | Oppo(重庆)智能科技有限公司 | 生物识别方法及相关产品 |
CN110210394B (zh) * | 2019-05-31 | 2021-05-04 | Oppo广东移动通信有限公司 | 静脉图像采集方法及相关产品 |
CN110162954B (zh) * | 2019-05-31 | 2021-09-28 | Oppo(重庆)智能科技有限公司 | 权限管理方法及相关产品 |
CN110310399B (zh) * | 2019-06-26 | 2021-11-02 | Oppo广东移动通信有限公司 | 静脉门锁的解锁方法及相关产品 |
CN110287161B (zh) * | 2019-07-02 | 2022-09-23 | 北京字节跳动网络技术有限公司 | 图像处理方法和装置 |
CN110633733B (zh) * | 2019-08-14 | 2024-05-03 | 中国平安财产保险股份有限公司 | 图像智能匹配方法、装置及计算机可读存储介质 |
CN110738236B (zh) * | 2019-09-16 | 2022-07-22 | 深圳市国信合成科技有限公司 | 图像匹配方法、装置、计算机设备及存储介质 |
CN110851349B (zh) * | 2019-10-10 | 2023-12-26 | 岳阳礼一科技股份有限公司 | 页面异常显示的检测方法、终端设备及存储介质 |
CN110880003B (zh) * | 2019-10-12 | 2023-01-17 | 中国第一汽车股份有限公司 | 一种图像匹配方法、装置、存储介质及汽车 |
CN110689485B (zh) * | 2019-10-14 | 2022-11-04 | 中国空气动力研究与发展中心超高速空气动力研究所 | 一种应用于大型压力容器红外无损检测的sift图像拼接方法 |
CN110766095B (zh) * | 2019-11-01 | 2022-04-19 | 易思维(杭州)科技有限公司 | 基于图像灰度特征的缺陷检测方法 |
CN110866503B (zh) * | 2019-11-19 | 2024-01-05 | 圣点世纪科技股份有限公司 | 指静脉设备的异常检测方法及*** |
CN112270372B (zh) * | 2020-11-06 | 2023-09-29 | 首都师范大学 | 一种确定目标对象的方法、装置、计算机设备和介质 |
CN113048912B (zh) * | 2021-02-26 | 2022-07-19 | 山东师范大学 | 投影仪的标定***及方法 |
CN113469971B (zh) * | 2021-06-30 | 2023-10-13 | 深圳中科飞测科技股份有限公司 | 一种图像匹配方法及检测装置、存储介质 |
CN115063613B (zh) * | 2022-08-09 | 2023-07-14 | 海纳云物联科技有限公司 | 一种验证商品标签的方法及装置 |
CN115331212B (zh) * | 2022-10-13 | 2023-10-27 | 南通东鼎彩印包装厂 | 一种易拉罐底喷码异常识别方法 |
CN115953407B (zh) * | 2023-03-15 | 2023-06-09 | 深圳市科姆特精密科技有限公司 | 基于计算机视觉的半导体设备维修*** |
CN117037159B (zh) * | 2023-10-09 | 2024-03-19 | 网思科技股份有限公司 | 基于卷积神经网络的油画真伪鉴别方法、装置及存储介质 |
CN117893502A (zh) * | 2024-01-15 | 2024-04-16 | 广州市科帕电子科技有限公司 | 图像检测方法、装置、设备及存储介质 |
CN117930375B (zh) * | 2024-03-22 | 2024-06-25 | 国擎(山东)信息科技有限公司 | 多维度检测技术融合的通道式太赫兹人体安检*** |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101470800A (zh) * | 2007-12-30 | 2009-07-01 | 沈阳工业大学 | 手形识别方法 |
CN103268483A (zh) * | 2013-05-31 | 2013-08-28 | 沈阳工业大学 | 开放式环境非接触采集下的掌纹识别方法 |
CN106529407A (zh) * | 2016-10-09 | 2017-03-22 | 北京电子科技职业学院 | 一种车载指纹识别方法 |
CN107273844A (zh) * | 2017-06-12 | 2017-10-20 | 成都芯软科技股份公司 | 掌静脉识别匹配方法及装置 |
-
2018
- 2018-06-11 CN CN201810593681.5A patent/CN109101867A/zh not_active Withdrawn
- 2018-09-05 WO PCT/CN2018/104050 patent/WO2019237520A1/zh active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101470800A (zh) * | 2007-12-30 | 2009-07-01 | 沈阳工业大学 | 手形识别方法 |
CN103268483A (zh) * | 2013-05-31 | 2013-08-28 | 沈阳工业大学 | 开放式环境非接触采集下的掌纹识别方法 |
CN106529407A (zh) * | 2016-10-09 | 2017-03-22 | 北京电子科技职业学院 | 一种车载指纹识别方法 |
CN107273844A (zh) * | 2017-06-12 | 2017-10-20 | 成都芯软科技股份公司 | 掌静脉识别匹配方法及装置 |
Cited By (47)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111223134A (zh) * | 2020-01-19 | 2020-06-02 | 北京航空航天大学 | 直线特征的特征向量构建方法、特征匹配方法和计算设备 |
CN111223134B (zh) * | 2020-01-19 | 2023-04-14 | 北京航空航天大学 | 直线特征的特征向量构建方法、特征匹配方法和计算设备 |
CN111445402A (zh) * | 2020-03-20 | 2020-07-24 | 上海中通吉网络技术有限公司 | 一种图像去噪方法及装置 |
CN111445402B (zh) * | 2020-03-20 | 2023-09-19 | 上海中通吉网络技术有限公司 | 一种图像去噪方法及装置 |
CN111462074A (zh) * | 2020-03-31 | 2020-07-28 | 珠海格力智能装备有限公司 | 轴承外观检测方法、装置、***、计算机设备及存储介质 |
CN111462074B (zh) * | 2020-03-31 | 2024-03-19 | 珠海格力智能装备有限公司 | 轴承外观检测方法、装置、***、计算机设备及存储介质 |
CN113658261A (zh) * | 2020-05-12 | 2021-11-16 | 武汉Tcl集团工业研究院有限公司 | 棋盘格角点位置检测方法、装置、智能终端及存储介质 |
CN111695498A (zh) * | 2020-06-10 | 2020-09-22 | 西南林业大学 | 一种木材同一性检测方法 |
CN111815578A (zh) * | 2020-06-23 | 2020-10-23 | 浙江大华技术股份有限公司 | 图像条纹检测方法、视频监控***及相关装置 |
CN111815578B (zh) * | 2020-06-23 | 2024-05-28 | 浙江大华技术股份有限公司 | 图像条纹检测方法、视频监控***及相关装置 |
CN112084883A (zh) * | 2020-08-18 | 2020-12-15 | 珠海格力电器股份有限公司 | 指静脉采集方法、装置、指静脉识别设备及存储介质 |
CN112164032B (zh) * | 2020-09-14 | 2023-12-29 | 浙江华睿科技股份有限公司 | 一种点胶方法、装置、电子设备及存储介质 |
CN112164032A (zh) * | 2020-09-14 | 2021-01-01 | 浙江华睿科技有限公司 | 一种点胶方法、装置、电子设备及存储介质 |
CN112308886A (zh) * | 2020-09-27 | 2021-02-02 | 深圳疆程技术有限公司 | 一种确定hud图像尺寸的方法、装置及*** |
CN112381105B (zh) * | 2020-11-16 | 2024-05-07 | 平安科技(深圳)有限公司 | 滑块校验的自动化处理方法、装置、设备及介质 |
CN112381105A (zh) * | 2020-11-16 | 2021-02-19 | 平安科技(深圳)有限公司 | 滑块校验的自动化处理方法、装置、设备及介质 |
CN112686302A (zh) * | 2020-12-29 | 2021-04-20 | 科大讯飞股份有限公司 | 图像特征点匹配方法、计算机设备及存储装置 |
CN112686302B (zh) * | 2020-12-29 | 2024-02-13 | 科大讯飞股份有限公司 | 图像特征点匹配方法、计算机设备及存储装置 |
CN112883827A (zh) * | 2021-01-28 | 2021-06-01 | 腾讯科技(深圳)有限公司 | 图像中指定目标的识别方法、装置、电子设备和存储介质 |
CN112883827B (zh) * | 2021-01-28 | 2024-03-29 | 腾讯科技(深圳)有限公司 | 图像中指定目标的识别方法、装置、电子设备和存储介质 |
CN112950540A (zh) * | 2021-02-01 | 2021-06-11 | 联宝(合肥)电子科技有限公司 | 一种条码识别方法及设备 |
CN112950540B (zh) * | 2021-02-01 | 2022-04-15 | 联宝(合肥)电子科技有限公司 | 一种条码识别方法及设备 |
CN113033550A (zh) * | 2021-03-15 | 2021-06-25 | 合肥联宝信息技术有限公司 | 一种图像的检测方法、装置及计算机可读介质 |
CN113033550B (zh) * | 2021-03-15 | 2022-04-15 | 合肥联宝信息技术有限公司 | 一种图像的检测方法、装置及计算机可读介质 |
CN113592921A (zh) * | 2021-06-30 | 2021-11-02 | 北京旷视科技有限公司 | 图像匹配方法,装置,电子设备及存储介质 |
CN113627446B (zh) * | 2021-08-18 | 2023-10-31 | 成都工业学院 | 基于梯度向量的特征点描述算子的图像匹配方法及*** |
CN113627446A (zh) * | 2021-08-18 | 2021-11-09 | 成都工业学院 | 基于梯度向量的特征点描述算子的图像匹配方法及*** |
CN114217756A (zh) * | 2021-11-26 | 2022-03-22 | 海信电子科技(武汉)有限公司 | 显示设备、图像检测方法及装置 |
CN114217756B (zh) * | 2021-11-26 | 2023-09-19 | 海信电子科技(武汉)有限公司 | 显示设备、图像检测方法及装置 |
CN115314634A (zh) * | 2022-06-28 | 2022-11-08 | 上海艾为电子技术股份有限公司 | 防抖检测方法、装置、终端设备和可读存储介质 |
CN115314634B (zh) * | 2022-06-28 | 2024-05-31 | 上海艾为电子技术股份有限公司 | 防抖检测方法、装置、终端设备和可读存储介质 |
CN115184380A (zh) * | 2022-08-01 | 2022-10-14 | 国网上海市电力公司 | 基于机器视觉的印刷电路板焊点异常检测方法 |
CN115184380B (zh) * | 2022-08-01 | 2024-04-26 | 国网上海市电力公司 | 基于机器视觉的印刷电路板焊点异常检测方法 |
CN115830719A (zh) * | 2023-02-16 | 2023-03-21 | 青岛旭华建设集团有限公司 | 基于图像处理的建筑工地危险行为识别方法 |
CN116310448A (zh) * | 2023-05-24 | 2023-06-23 | 山东曙岳车辆有限公司 | 基于计算机视觉的集装箱拼装匹配性检测方法 |
CN116452586A (zh) * | 2023-06-15 | 2023-07-18 | 山东飞宏工程机械有限公司 | 一种隧道小导管余料自动对焊质量检测*** |
CN116452586B (zh) * | 2023-06-15 | 2023-09-26 | 山东飞宏工程机械有限公司 | 一种隧道小导管余料自动对焊质量检测*** |
CN116612306B (zh) * | 2023-07-17 | 2023-09-26 | 山东顺发重工有限公司 | 基于计算机视觉的法兰盘智能对位方法及*** |
CN116612306A (zh) * | 2023-07-17 | 2023-08-18 | 山东顺发重工有限公司 | 基于计算机视觉的法兰盘智能对位方法及*** |
CN116703909B (zh) * | 2023-08-07 | 2023-10-27 | 威海海泰电子有限公司 | 一种电源适配器生产质量智能检测方法 |
CN116703909A (zh) * | 2023-08-07 | 2023-09-05 | 威海海泰电子有限公司 | 一种电源适配器生产质量智能检测方法 |
CN117292142A (zh) * | 2023-09-26 | 2023-12-26 | 北京鹰之眼智能健康科技有限公司 | 一种图像的匹配方法 |
CN117292142B (zh) * | 2023-09-26 | 2024-05-10 | 北京鹰之眼智能健康科技有限公司 | 一种图像的匹配方法 |
CN116993643B (zh) * | 2023-09-27 | 2023-12-12 | 山东建筑大学 | 基于人工智能的无人机摄影测量图像校正方法 |
CN116993643A (zh) * | 2023-09-27 | 2023-11-03 | 山东建筑大学 | 基于人工智能的无人机摄影测量图像校正方法 |
CN117495852A (zh) * | 2023-12-29 | 2024-02-02 | 天津中荣印刷科技有限公司 | 一种基于图像分析的数字印刷质量检测方法 |
CN117495852B (zh) * | 2023-12-29 | 2024-05-28 | 天津中荣印刷科技有限公司 | 一种基于图像分析的数字印刷质量检测方法 |
Also Published As
Publication number | Publication date |
---|---|
CN109101867A (zh) | 2018-12-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2019237520A1 (zh) | 一种图像匹配方法、装置、计算机设备及存储介质 | |
WO2019232945A1 (zh) | 图像处理方法、装置、计算机设备及存储介质 | |
WO2018219054A1 (zh) | 一种车牌识别方法、装置及*** | |
WO2021027336A1 (zh) | 基于***和签名的身份验证方法、装置和计算机设备 | |
WO2019205290A1 (zh) | 一种图像检测方法、装置、计算机设备及存储介质 | |
KR100374708B1 (ko) | 회전영상의 보정에 의한 비접촉식 홍채인식방법 | |
JP4868530B2 (ja) | 画像認識装置 | |
CN110546651B (zh) | 用于识别对象的方法、***和计算机可读介质 | |
US9710691B1 (en) | Touchless fingerprint matching systems and methods | |
CN106981077B (zh) | 基于dce和lss的红外图像和可见光图像配准方法 | |
CN111209845A (zh) | 人脸识别方法、装置、计算机设备及存储介质 | |
US20050196044A1 (en) | Method of extracting candidate human region within image, system for extracting candidate human region, program for extracting candidate human region, method of discerning top and bottom of human image, system for discerning top and bottom, and program for discerning top and bottom | |
TWI639136B (zh) | 即時視訊畫面拼接方法 | |
US11281922B2 (en) | Face recognition system, method for establishing data of face recognition, and face recognizing method thereof | |
US12014498B2 (en) | Image enhancement processing method, device, equipment, and medium based on artificial intelligence | |
US20180247152A1 (en) | Method and apparatus for distance measurement | |
Lijing et al. | A warped document image mosaicing method based on registration and TRS transform | |
US20190279392A1 (en) | Medium recognition device and medium recognition method | |
KR20190027165A (ko) | 복수 이미지의 특징을 통일하는 이미지 조정 시스템 및 방법 | |
CN112308044B (zh) | 针对掌静脉图像的图像增强处理方法和掌静脉识别方法 | |
CN113344961A (zh) | 图像背景分割方法、装置、计算设备及存储介质 | |
Hu et al. | Digital video stabilization based on multilayer gray projection | |
Jiang et al. | The fast recognition of vehicle license plate based on the improved template matching | |
CN112132054A (zh) | 一种基于深度学习的文档定位和分割方法 | |
CN112381042A (zh) | 从掌静脉图像中提取掌静脉特征的方法、掌静脉识别方法 |
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: 18922123 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 18922123 Country of ref document: EP Kind code of ref document: A1 |