WO2020228181A1 - 手掌图像裁剪方法、装置、计算机设备及存储介质 - Google Patents

手掌图像裁剪方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2020228181A1
WO2020228181A1 PCT/CN2019/103179 CN2019103179W WO2020228181A1 WO 2020228181 A1 WO2020228181 A1 WO 2020228181A1 CN 2019103179 W CN2019103179 W CN 2019103179W WO 2020228181 A1 WO2020228181 A1 WO 2020228181A1
Authority
WO
WIPO (PCT)
Prior art keywords
palm
image
base
regression model
center point
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.)
Ceased
Application number
PCT/CN2019/103179
Other languages
English (en)
French (fr)
Inventor
侯丽
王福晴
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020228181A1 publication Critical patent/WO2020228181A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06V40/1347—Preprocessing; Feature extraction

Definitions

  • This application belongs to the field of image recognition, and in particular relates to a palm image cropping method, device, computer equipment and storage medium.
  • biometric recognition has the most extensive application range, and the recognition accuracy of iris is very high.
  • the collection area of these two biometrics is small, and high-resolution images are required to achieve satisfactory recognition accuracy.
  • Normal movement It is difficult for the imaging system of the equipment to collect high-resolution fingerprint or iris images that can meet the identification requirements.
  • the hand shape and palmprint area are larger, and the related recognition technology does not require high image resolution.
  • Palm print refers to the image of the palm from the end of the finger to the wrist. Many of these features can be used for identification: such as main lines, wrinkles, small textures, bifurcation points, etc. Palmprint recognition is also a non-invasive recognition method, which is relatively easy for users to accept and does not require high collection equipment.
  • This application provides a palm image cropping method, device, computer equipment and storage medium to solve the problem of palm image acquisition during palmprint recognition.
  • this application proposes a palm image cropping method, which includes the following steps:
  • the image is input to a pre-trained cascade posture regression model, and the palm key points output by the cascade posture regression model in response to the image are obtained, wherein the palm key points output by the cascade posture regression model include at least Index finger root A, little finger root B, middle finger tip D;
  • the image is cropped with a circle or rectangle that includes at least the base A of the index finger and the base B of the little finger to obtain a palm print image of the palm in the image to be processed.
  • a palm image cropping device including:
  • the acquisition module is used to acquire the image to be processed
  • the processing module is used to input the image to a pre-trained cascaded posture regression model, and obtain the palm key points output by the cascaded posture regression model in response to the image, wherein the output of the cascade posture regression model Key points of the palm include at least index finger root A, little finger root B, and middle finger tip D;
  • a calculation module configured to calculate the palm center point C of the palm in the image according to the palm key points
  • the execution module is configured to crop the image with a circle or rectangle containing at least the base A of the index finger and the base B of the little finger with the palm center point C as the center, and obtain the palm in the image to be processed Palm print image.
  • an embodiment of the present application further provides a computer device including a memory and a processor.
  • the memory stores computer-readable instructions.
  • the The processor executes the steps of the palm image cropping method as follows:
  • the image is input to a pre-trained cascade posture regression model, and the palm key points output by the cascade posture regression model in response to the image are obtained, wherein the palm key points output by the cascade posture regression model include at least Index finger root A, little finger root B, middle finger tip D;
  • the image is cropped with a circle or rectangle that includes at least the base A of the index finger and the base B of the little finger to obtain a palm print image of the palm in the image to be processed.
  • the embodiments of the present application also provide one or more non-volatile readable storage media storing computer readable instructions, and the non-volatile readable storage medium stores computer readable instructions,
  • the processor is caused to execute the following steps of the palm image cropping method:
  • the image is input to a pre-trained cascade posture regression model, and the palm key points output by the cascade posture regression model in response to the image are obtained, wherein the palm key points output by the cascade posture regression model include at least Index finger root A, little finger root B, middle finger tip D;
  • the image is cropped with a circle or rectangle that includes at least the base A of the index finger and the base B of the little finger to obtain a palm print image of the palm in the image to be processed.
  • the beneficial effects of the embodiments of the present invention are: by acquiring an image to be processed; inputting the image to a pre-trained cascaded posture regression model, and obtaining the palm key points output by the cascading posture regression model in response to the image; According to the key points of the palm, calculate the palm center point of the palm in the image; crop the palm center point as the center to obtain the palm print image of the palm in the image to be processed.
  • the cropped image contains key points for palmprint recognition, and the hand posture can be determined through the key points, while avoiding the interference of complex backgrounds and reducing the error rate of palmprint recognition.
  • FIG. 1 is a schematic diagram of the basic flow of a palm image cropping method according to an embodiment of this application;
  • FIG. 2 is a schematic diagram of cropping an image with a rectangle with palm center point C as the center and AB as one side according to an embodiment of the application;
  • Fig. 3 is a schematic diagram of a palm key point prediction process according to an embodiment of the application.
  • FIG. 4 is a schematic diagram of an image rotation process according to an embodiment of the application.
  • FIG. 5 is a schematic diagram of a training process of a cascaded posture regression model according to an embodiment of the application
  • FIG. 6 is a block diagram of the basic structure of a palm image cropping device according to an embodiment of this application.
  • FIG. 7 is a block diagram of the basic structure of the computer equipment implemented in this application.
  • terminal and “terminal equipment” used herein include both wireless signal receiver equipment, equipment that only has wireless signal receivers without transmitting capability, and equipment receiving and transmitting hardware.
  • a device which has a device capable of performing two-way communication receiving and transmitting hardware on a two-way communication link.
  • Such equipment may include: cellular or other communication equipment, which has a single-line display or multi-line display or cellular or other communication equipment without a multi-line display; PCS (Personal Communications Service, personal communication system), which can combine voice and data Processing, fax and/or data communication capabilities; PDA (Personal Digital Assistant), which can include radio frequency receivers, pagers, Internet/Intranet access, web browsers, notebooks, calendars and/or GPS (Global Positioning System (Global Positioning System) receiver; a conventional laptop and/or palmtop computer or other device, which has and/or includes a radio frequency receiver, a conventional laptop and/or palmtop computer or other device.
  • PCS Personal Communications Service, personal communication system
  • PDA Personal Digital Assistant
  • GPS Global Positioning System (Global Positioning System) receiver
  • a conventional laptop and/or palmtop computer or other device which has and/or includes a radio frequency receiver, a conventional laptop and/or palmtop computer or other device.
  • terminal and terminal equipment used here may be portable, transportable, installed in vehicles (aviation, sea and/or land), or suitable and/or configured to operate locally, and/or In a distributed form, it runs on the earth and/or any other location in space.
  • the "terminal” and “terminal device” used here can also be communication terminals, Internet terminals, music/video playback terminals, such as PDA, MID (Mobile Internet Device, mobile Internet device) and/or music/video playback Functional mobile phones can also be devices such as smart TVs and set-top boxes.
  • the terminal in this embodiment is the aforementioned terminal.
  • FIG. 1 is a schematic diagram of the basic flow of a palm image cropping method according to this embodiment.
  • a palm image cropping method includes the following steps:
  • the embodiment of the present application uses a cascaded posture regression model to predict key points of the palm of the input image.
  • the cascaded posture regression model is composed of a series of cascaded convolutional neural networks.
  • the cascaded posture regression model has been trained to predict the key points of the palm.
  • the key points of the palm of the hand detected in the embodiment of the present application include the base A of the index finger, the base B of the little finger, and the tip D of the middle finger.
  • the advantage of predicting the key position of the hand through the cascaded convolutional neural network is that a simpler network can be used in the initial stage, which reduces the amount of calculation and improves the efficiency of key point recognition.
  • the cascaded convolutional neural network model can also be called a regressor.
  • a specified initial prediction value is gradually refined through a series of regressors.
  • Each regressor relies on the output of the previous regressor to perform simple image operations.
  • the entire system learns the key point features of the hand from the training sample to predict the palm key point.
  • S103 Calculate the palm center point C of the palm in the image according to the palm key points
  • step S102 the base A of the index finger and the base B of the little finger are obtained.
  • C is the palm center.
  • Point, and palm center point C and middle finger fingertip D are located on both sides of straight line AB. That is, the palm point is on the midline of AB, and the middle fingertip D is on both sides of the straight line AB.
  • the image is cropped with a circle or rectangle containing at least the index finger root A and the little finger root B to obtain a palm print image containing the palm key points index finger root A and little finger root B.
  • the palm center point C is taken at 1/2 of the length of AB
  • the image is cropped with the palm center point C as the center and a rectangle with AB as one side
  • all the palms are retained.
  • Key points and palmprint images and remove the background image that interferes with palmprint recognition.
  • the cropped image is used for palmprint recognition, which reduces the amount of calculation for palmprint recognition and improves calculation efficiency. There is no interference from background images. Higher accuracy.
  • the palm point C is located on the midline of AB, with C as the center, the third vertex of the rectangle with AB as a side is extended from BC to F, and the length of CF is equal to BC.
  • the fourth vertex is extended from AC to G, the length of CG is equal to AC. Then the rectangle with C as the center and AB as one side is ABGF.
  • the step of inputting the image to a pre-trained cascaded posture regression model to obtain palm key points output by the cascaded posture regression model in response to the image includes the following steps :
  • S111 Input the image into a preset target detection model to remove the image background, and obtain a palm image to be recognized output by the target detection model in response to the image, wherein the preset target detection model adopts SSD target detection algorithm predicts the position of the hand in the image;
  • the preset target detection model adopts SSD (Single Shot MultiBox Detector) target detection algorithm.
  • SSD Single Shot MultiBox Detector
  • a forward propagation CNN network which generates a series of fixed-size bounding boxes, and the possibility of object instances contained in each box, that is, score. After that, perform a non-maximum suppression (Non-maximum suppression) to obtain the final prediction results.
  • the first part of the SSD model is a standard architecture for image classification.
  • an additional auxiliary network structure is added: that is, the convolutional layer used for feature extraction.
  • the size of these convolutional layers decreases layer by layer, and predictions can be performed at multiple scales.
  • Each newly added layer can be predicted using a series of convolution kernels.
  • a 3*3 convolution kernel is used for prediction, and a value is predicted at a certain position. This value is the score of a certain category.
  • the image is calculated by the model, and a value will be generated at each position.
  • a box of a certain size at a certain position with the highest probability is the predicted hand position.
  • the image to be processed is cropped according to this position and size to obtain the palm image.
  • step S103 after step S103; before step S104, it also includes the following steps:
  • S121 Calculate the inclination of the palm in the image according to the tip D of the middle finger, the base of the index finger A, and the base of the little finger B;
  • the inclination of the hand is judged according to the connection of the middle finger tip D, index finger root A and little finger root B.
  • the coordinates of index finger root A are (Ax, Ay)
  • the coordinates of little finger root B are (Bx, By)
  • B can be expressed as (Bx-Ax, By-Ay)
  • the inclination angle a between AB and the horizontal direction can be expressed as arctan (By-Ay)/(Bx-Ax).
  • D can be expressed as (D.x-A.x, D.y-A.y), and the inclination angle b between DA and the horizontal direction can be expressed as arctan(D.y-A.y)/(D.x-A.x).
  • a is smaller than b, the palm image is rotated clockwise by a degree, so that AB is in a horizontal position, and the middle finger tip D is located above AB, so that the palm posture is aligned with the preset posture of the palmprint recognition device. Understandably, rotating the palm image counterclockwise 360-a degrees can achieve the same effect.
  • the cascaded pose regression model is trained in the following steps:
  • training samples are a set of palm pictures marked with key points of the palm, wherein the key points of the palm marked by each sample include at least the base of the index finger, the base of the little finger, and the tip of the middle finger;
  • the cascaded posture regression model is composed of a cascaded convolutional neural network.
  • the training sample is a set of palm images marked with key points of the palm.
  • the key points of the palm marked by each sample include at least the root of the index finger, the root of the little finger and the tip of the middle finger.
  • the trained posture regression model can predict the position between the root of the index finger, the root of the little finger and the middle finger.
  • the cascaded convolutional neural network model of the cascaded pose regression model can also be called a regressor.
  • a specified initial prediction value is gradually refined through a series of regressors. Each regressor depends on the output of the previous regressor. To perform simple image operations, the entire system learns the key point features of the hand from the training samples to predict the key points of the palm.
  • the purpose of palm key point detection is to estimate the vector.
  • K represents the number of key points. Since each key point has two coordinates, the length of S is 2K.
  • feature extraction is first performed. Shape-indexed features are used here. Manually designed features such as HOG, SIFT, etc., or other learning based features can also be used.
  • the key point prediction value S t-1 output by the t-1th regressor and the training sample are input to the tth regressor to obtain the key point prediction output by the tth regressor Value S t
  • the key point predicted value S T output by the T-th regressor is the palm key point predicted by the cascade posture regression model.
  • the regressor includes a feature extraction layer and a key point increment estimation layer.
  • the key point predicted value S T output by the T-th regressor is the palm key point predicted by the cascaded posture regression model.
  • K different postures can be initialized for each image, and the cascaded neural network model can be run K times to obtain K different results. Cluster these results and select the result with the highest density as the final result. That is, the predicted values of K index finger roots, little finger roots and middle fingertips are obtained, and each predicted value is the center of the circle, and a circle is drawn with eps as the radius (called the neighborhood eps-neigbourhood), and then count how many points there are In this circle, this number is the density value of the point. Select a density threshold MinPts.
  • the center point of the circle less than MinPts is a low-density point
  • the center point of the circle greater than or equal to MinPts is a high-density point (called the core point).
  • the center point of the circle with the highest density is taken as the final predicted palm key point.
  • the specific algorithm is the DBSCAN algorithm.
  • the distance between the predicted palm key points and the marked palm key points refers to the distance between the corresponding key points. For example, if the marked palm key points are the root of the index finger, the tip of the middle finger, and the root of the little finger, then calculate the predicted The distance between the base of the index finger and the base of the labeled index finger, the predicted distance between the base of the little finger and the base of the labeled little finger, and the predicted distance between the tip of the middle finger and the tip of the labeled middle finger.
  • the distance between two points is the square of the difference between the two coordinates.
  • Adjust the parameters of each node of the cascaded posture regression model that is, adjust the weight of each node of the cascaded convolutional neural network model to minimize the distance between the predicted palm key point and the labeled palm key point, and the training ends. It is considered that the cascaded posture regression model defined by each node parameter has the function of predicting the key points of the palm.
  • FIG. 6 is a block diagram of the basic structure of the palm image cropping device of this embodiment.
  • a palm image cropping device includes: an acquisition module 210, a processing module 220, a calculation module 230, and an execution module 240.
  • the acquisition module 210 is used to acquire an image to be processed; and the processing module 220 is used to The image is input to a pre-trained cascade posture regression model, and the palm key points output by the cascade posture regression model in response to the image are obtained, wherein the palm key points output by the cascade posture regression model include at least Index finger base A, little finger base B, middle finger fingertip D; calculation module 230, used to calculate the palm center point C in the image according to the palm key points; execution module 240, used to use the palm center point C is the center, and the image is cropped with a circle or rectangle containing at least the base A of the index finger and the base B of the little finger to obtain the palm print image of the palm in the image to be processed.
  • the image to be processed is obtained; the image is input to a pre-trained cascade posture regression model, and the key points of the palm output by the cascade posture regression model in response to the image are obtained, wherein the level
  • the palm key points output by the joint posture regression model include at least index finger root A, little finger root B, and middle finger tip D; according to the palm key points, calculate the palm center point C of the palm in the image; use the palm center point C As a center, the image is cropped with a circle or rectangle that includes at least the base A of the index finger and the base B of the little finger to obtain a palm print image of the palm in the image to be processed.
  • the cropped image contains key points for palmprint recognition, and the hand posture can be determined through the key points, while avoiding the interference of complex backgrounds and reducing the error rate of palmprint recognition.
  • the calculation module 230 further includes: a midpoint determination sub-module for determining the midpoint E of the line segment AB formed by the base of the index finger A and the base of the little finger B as the endpoints; With the midpoint E as the end point, the palm center point C is determined in a direction perpendicular to the line segment AB, so that the length of CE is at least 1/2 of the length of the line segment AB, and the palm center point C and the middle finger The tips D are located on both sides of the line segment AB.
  • the processing module 220 further includes: a first detection sub-module for inputting the image into a preset target detection model to remove the image background and obtain the target detection model response The palm image to be recognized output from the image, wherein the preset target detection model uses the SSD target detection algorithm to predict the position of the hand in the image; the first processing sub-module is used to input the palm image to a pre-trained stage A joint posture regression model is used to obtain palm key points output by the cascade posture regression model in response to the palm image.
  • the palm image cropping device further includes: a first calculation sub-module configured to calculate the inclination of the palm in the image according to the middle finger tip D, index finger base A, and little finger base B;
  • the second processing sub-module is configured to rotate the image according to the inclination.
  • the execution module 240 further includes: a third processing sub-module, configured to take the line AB between the index finger root A and the little finger root B as one of the sides of the rectangle; Take the base A of the index finger or the base B of the little finger as an end point, the palm center point C as the center, and the diagonal line of the line AC or BC is twice the length of the rectangle. Cropped.
  • the palm image cropping device further includes: a first acquisition sub-module for acquiring training samples, the training samples being a set of palm images marked with key points of the palm, wherein each sample is marked with a palm
  • the key points include at least the root of the index finger, the root of the little finger, and the tip of the middle finger; the fourth processing sub-module is used to input the training sample into the cascade posture regression model to obtain the cascade posture regression model in response to the training sample.
  • Predicted palm key points Predicted palm key points; a second calculation sub-module, used to calculate the distance between the labeled palm key points and the predicted palm key points; a first adjustment sub-module, used to adjust the cascaded posture regression
  • the training ends when the distance between the parameters of each node of the model and the marked palm key point and the predicted palm key point is the smallest.
  • the key point prediction value S t , the key point prediction value S T output by the T-th regressor is the palm key point predicted by the cascade posture regression model.
  • FIG. 7 is a block diagram of the basic structure of the computer device in this embodiment.
  • the computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected through a system bus.
  • the non-volatile storage medium of the computer device stores an operating system, a database, and computer-readable instructions.
  • the database may store control information sequences.
  • the processor can realize a A method of cropping palm images.
  • the processor of the computer equipment is used to provide computing and control capabilities and support the operation of the entire computer equipment.
  • a computer readable instruction may be stored in the memory of the computer device. When the computer readable instruction is executed by the processor, the processor may execute a palm image cropping method.
  • the network interface of the computer device is used to connect and communicate with the terminal.
  • FIG. 7 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or less parts than shown in the figure, or combining some parts, or having a different part arrangement.
  • the processor is used to execute the specific content of the acquisition module 210, the processing module 220, the calculation module 230, and the execution module 240 in FIG. 6, and the memory stores computer readable instructions and various data required to execute the above modules.
  • the network interface is used for data transmission between user terminals or servers.
  • the memory in this embodiment stores computer-readable instructions and data required to execute all sub-modules in the palm image cropping method, and the server can call the computer-readable instructions and data of the server to perform the functions of all the sub-modules.
  • the computer device obtains the image to be processed; inputs the image to a pre-trained cascade posture regression model, and obtains the palm key points output by the cascade posture regression model in response to the image, wherein the cascade posture
  • the palm key points output by the regression model include at least index finger root A, little finger root B, and middle finger tip D; according to the palm key points, the palm center point C of the palm in the image is calculated; centering on the palm center point C , Crop the image with a circle or rectangle that includes at least the base A of the index finger and the base B of the little finger to obtain the palm print image of the palm in the image to be processed.
  • the cropped image contains key points for palmprint recognition, and the hand posture can be determined through the key points, while avoiding the interference of complex backgrounds and reducing the error rate of palmprint recognition.
  • the present application also provides a storage medium storing computer-readable instructions.
  • the computer-readable instructions are executed by one or more processors, the one or more processors execute the palm image cropping described in any of the above embodiments. Method steps.
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Human Computer Interaction (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

一种手掌图像裁剪的方法、装置、计算机设备及存储介质，所述方法包括下述步骤：获取待处理的图像（S101）；将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D（S102）；根据所述手掌关键点，计算所述图像中手掌的掌心点C（S103）；以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像（S104）。经过裁剪的图像包含掌纹识别的关键点，避免了复杂背景的干扰，降低了掌纹识别的误差率。

Description

手掌图像裁剪方法、装置、计算机设备及存储介质
【交叉引用】
本申请以2019年5月15日提交的申请号为201910401330.4，名称为“手掌图像裁剪方法、装置、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。
【技术领域】
本申请属于图像识别领域，尤其涉及一种手掌图像裁剪方法、装置、计算机设备及存储介质。
【背景技术】
在常用的生物特征识别技术中，指纹识别应用范围最为广泛，虹膜的识别精度很高，然而这两种生物特征的采集区域较小，需要高分辨率的图像才能取得满意的识别精度，普通移动设备的成像系统很难采集到能够满足识别要求的高分辨率指纹或虹膜图像。而手形和掌纹的区域较大，相关的识别技术并不需要很高的图像分辨率。掌纹是指手指末端到手腕部分的手掌图像。其中很多特征可以用来进行身份识别：如主线、皱纹、细小的纹理、分叉点等。掌纹识别也是一种非侵犯性的识别方法，用户比较容易接受，对采集设备要求不高。
但现有技术在获取手掌图像时，通常使用固定设备，通过模板匹配法，检测手的轮廓，要求手的姿态与图像采集设定的姿态一致，当手的姿态与图像采集设定的姿态不一致时，采集到的图像用于掌纹识别，存在识别误差率较高的问题。
【发明内容】
本申请提供一种手掌图像裁剪方法、装置、计算机设备及存储介质，以解决掌纹识别时手掌图像的获取问题。
为解决上述技术问题，本申请提出一种手掌图像裁剪方法，包括如下步骤：
获取待处理的图像；
将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归 模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
根据所述手掌关键点，计算所述图像中手掌的掌心点C；
以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
为解决上述问题，本申请还提供一种手掌图像裁剪装置，包括：
获取模块，用于获取待处理的图像；
处理模块，用于将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
计算模块，用于根据所述手掌关键点，计算所述图像中手掌的掌心点C；
执行模块，用于以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
为解决上述技术问题，本申请实施例还提供一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行如下手掌图像裁剪方法的步骤:
获取待处理的图像；
将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
根据所述手掌关键点，计算所述图像中手掌的掌心点C；
以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
为解决上述技术问题，本申请实施例还提供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，所述非易失性可读存储介质上存储有计算机可读指令，所述计算机可读指令被处理器执行时，使得所述处理器执行如下手掌图像裁剪方法的步骤:
获取待处理的图像；
将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的 手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
根据所述手掌关键点，计算所述图像中手掌的掌心点C；
以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
本发明实施例的有益效果为：通过获取待处理的图像；将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点；根据所述手掌关键点，计算所述图像中手掌的掌心点；以掌心点为中心进行裁剪，获取所述待处理图像中手掌的掌纹图像。经过裁剪后的图像包含了用于掌纹识别的关键点，且通过关键点可以确定手的姿态，同时避免了复杂背景的干扰，降低了掌纹识别的误差率。
【附图说明】
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例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，对于本领域技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图
图1为本申请实施例一种手掌图像裁剪方法基本流程示意图；
图2为本申请实施例以掌心点C为中心，以AB为一条边的矩形对图像进行裁剪的示意图；
图3为本申请实施例手掌关键点预测流程示意图；
图4为本申请实施例图像旋转流程示意图；
图5为本申请实施例级联姿势回归模型训练流程示意图；
图6为本申请实施例一种手掌图像裁剪装置基本结构框图；
图7为本申请实施计算机设备基本结构框图。
【具体实施方式】
为了使本技术领域的人员更好地理解本申请方案，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。
在本申请的说明书和权利要求书及上述附图中的描述的一些流程中，包含了按照特定顺序出现的多个操作，但是应该清楚了解，这些操作可以不按照其在本文中出现的顺序来执行或并行执行，操作的序号如101、102等，仅仅是用 于区分开各个不同的操作，序号本身不代表任何的执行顺序。另外，这些流程可以包括更多或更少的操作，并且这些操作可以按顺序执行或并行执行。需要说明的是，本文中的“第一”、“第二”等描述，是用于区分不同的消息、设备、模块等，不代表先后顺序，也不限定“第一”和“第二”是不同的类型。
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，所描述的实施例仅仅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。
实施例
本技术领域技术人员可以理解，这里所使用的“终端”、“终端设备”既包括无线信号接收器的设备，其仅具备无发射能力的无线信号接收器的设备，又包括接收和发射硬件的设备，其具有能够在双向通信链路上，执行双向通信的接收和发射硬件的设备。这种设备可以包括：蜂窝或其他通信设备，其具有单线路显示器或多线路显示器或没有多线路显示器的蜂窝或其他通信设备；PCS(Personal Communications Service，个人通信系统)，其可以组合语音、数据处理、传真和/或数据通信能力；PDA(Personal Digital Assistant，个人数字助理)，其可以包括射频接收器、寻呼机、互联网/内联网访问、网络浏览器、记事本、日历和/或GPS(Global Positioning System，全球定位系统)接收器；常规膝上型和/或掌上型计算机或其他设备，其具有和/或包括射频接收器的常规膝上型和/或掌上型计算机或其他设备。这里所使用的“终端”、“终端设备”可以是便携式、可运输、安装在交通工具(航空、海运和/或陆地)中的，或者适合于和/或配置为在本地运行，和/或以分布形式，运行在地球和/或空间的任何其他位置运行。这里所使用的“终端”、“终端设备”还可以是通信终端、上网终端、音乐/视频播放终端，例如可以是PDA、MID(Mobile Internet Device，移动互联网设备)和/或具有音乐/视频播放功能的移动电话，也可以是智能电视、机顶盒等设备。
本实施方式中的终端即为上述的终端。
具体地，请参阅图1，图1为本实施例一种手掌图像裁剪方法的基本流程示意图。
如图1所示，一种手掌图像裁剪方法，包括下述步骤：
S101、获取待处理的图像；
通过具有图像采集功能的设备获取待处理的图片，包括带摄像头的手持设 备、移动设备等。
S102、将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
本申请实施例通过级联姿势回归模型对输入的图像进行手掌关键点的预测。级联姿势回归模型由一系列的卷积神经网络级联构成，级联姿势回归模型经过训练具备预测手掌关键点的功能，具体的训练过程请参见图5。本申请实施例检测的手的掌关键点包含食指指根A、小指指根B、中指指尖D。通过级联的卷积神经网络预测手的关键位置的优点是可以在初始阶段采用较简单的网络,减少计算量,提高关键点识别的效率。
级联的卷积神经网络模型也可以称之为回归器。通过一系列回归器将一个指定的初始预测值逐步细化，每一个回归器都依靠前一个回归器的输出来执行简单的图像操作，整个系统从训练样本中学习手的关键点特征来预测手掌关键点。
S103、根据所述手掌关键点，计算所述图像中手掌的掌心点C；
根据步骤S102获取了食指指根A、小指指根B，以AB的中点E为一个端点，以垂直于AB的方向，并以至少等于AB长度的1/2取线段CE，C即为掌心点，且掌心点C与中指指尖D分别位于直线AB的两边。即掌心点在AB的中线上，与中指指尖D分别位于直线AB的两边。
S104、以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
以掌心点C为中心，以至少包含食指指根A和小指指根B的圆形或矩形对图像进行裁剪，得到包含手掌关键点食指指根A和小指指根B的掌纹图像。特别地，如图2所示，当以AB长度的1/2取手掌掌心点C，并以掌心点C为中心，以AB为一条边的矩形对图像进行裁剪时，既全部保留了手掌的关键点和掌纹图像，又去除了干扰掌纹识别的背景图像，裁剪后的图像用于掌纹识别减少了掌纹识别的计算量，提高计算效率，没有背景图像的干扰，掌纹识别的精度更高。由于掌心点C位于AB的中线上，以C为中心，以AB为一条边的矩形的第3个顶点为BC延长到F，CF的长度等于BC，同理，第4个顶点为AC延长到G，CG的长度等于AC。则以C为中心，以AB为一条边的矩形为ABGF。
如图3所示，在所述将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点的步骤中，包括下述步骤：
S111、将所述图像输入到预设的目标检测模型中，以去除所述图像背景，获取所述目标检测模型响应所述图像而输出的待识别的手掌图像，其中预设的目标检测模型采用SSD目标检测算法预测图像中手的位置；
当图像采集设备采集到的手掌图像背景较复杂时，通过预设的目标检测模型预测手的位置，从而输出只包含待识别的手掌图像。预设的目标检测模型采用SSD(Single Shot MultiBox Detector)目标检测算法。SSD是基于一个前向传播CNN网络，产生一系列固定大小(fixed-size)的bounding boxes，以及每一个box中包含物体实例的可能性，即score。之后，进行一个非极大值抑制(Non-maximum suppression)得到最终的预测结果predictions。
SSD模型的最开始部分，称作base network，是用于图像分类的标准架构。在base network之后，添加了额外辅助的网络结构：即用来进行特征提取的卷积层，这些卷积层的大小是逐层递减的，可以在多尺度下进行predictions。
每一个新添加的层，可以使用一系列的卷积核进行预测。对于一个大小为m*n、p通道的特征层，使用3*3的卷积核进行预测，在某个位置上预测出一个值，该值是某一类别的得分，本申请实施例中也可以理解为包含在该位置上，以某一尺寸的box包含手的概率。图像经该模型计算，在每个位置都将产生一个值。概率最高的某一位置某一尺寸的box即为预测的手的位置。按照此位置和尺寸对待处理的图片进行裁剪，得到手掌图像。
S112、将所述手掌图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述手掌图像而输出的手掌关键点。
将去除了背景的手掌图像输入到预先训练的级联姿势回归模型中，获得级联姿势回归模型输出的手掌关键点。
如图4所示，在步骤S103之后；步骤之S104前，还包括下述步骤：
S121、根据所述中指指尖D、食指指根A和小指指根B计算所述图像中手掌的倾斜度；
根据中指指尖D、食指指根A和小指指根B的连线来判断手的倾斜度，食指指根A的坐标为(A.x,A.y),小指指根B的坐标为(B.x,B.y)，以A为原点，则B可以表示(B.x-A.x，B.y-A.y)，AB与水平方向的倾斜角度a可表示为arctan (B.y-A.y)/(B.x-A.x)。
S122、根据所述倾斜度，对所述图像进行旋转。
以A为原点，D可以表示为(D.x-A.x，D.y-A.y)，DA与水平方向的倾斜角度b可表示为arctan(D.y-A.y)/(D.x-A.x)。当a小于b时，将手掌图像顺时针旋转a度，使AB处于水平位置，中指指尖D位于AB的上方，实现手掌姿势与掌纹识别装置预设的姿势对齐。可以理解地，将手掌图像逆时针旋转360-a度，可以达到相同的效果。当a大于b时，将手掌图像顺时针旋转180+a度，使AB处于水平位置，中指指尖D位于AB的上方，实现手掌姿势与掌纹识别装置预设的姿势对齐。可以理解地，将手掌图像逆时针旋转180-a度，可以达到相同的效果。
如图5所示，级联姿势回归模型按照下述步骤训练：
S131、获取训练样本，所述训练样本为标注了手掌关键点的手掌图片集，其中，每个样本标注的手掌关键点至少包含食指指根、小指指根和中指指尖；
级联姿势回归模型通过级联的卷积神经网络构成，训练样本为标注了手掌关键点的手掌图片集，每个样本标注的手掌关键点至少包含食指指根、小指指根和中指指尖，使经过训练的姿势回归模型可以预测出食指指根、小指指根和中指之间的位置。
S132、将所述训练样本输入到级联姿势回归模型获取所述级联姿势回归模型响应所述训练样本而预测的手掌关键点；
级联姿势回归模型的级联的卷积神经网络模型也可以称之为回归器，通过一系列回归器将一个指定的初始预测值逐步细化，每一个回归器都依靠前一个回归器的输出来执行简单的图像操作，整个系统从训练样本中学习手的关键点特征来预测手掌关键点。
手掌关键点检测的目的是估计向量，K表示关键点的个数，由于每个关键点有横纵两个坐标，所以S的长度为2K。在每个阶段中首先进行特征提取，这里使用的是shape-indexed features，也可以使用诸如HOG、SIFT等人工设计的特征，或者其他可学习特征(learning based features)。
训练过程是先初始化一个S 0，然后提取特征，用特征进行回归，再更新S。即当t＝1时，将预设的关键点初始化值S 0和所述训练样本输入到第1个回归器，获取第1个回归器输出的关键点预测值S 1；
当t＝2～T时，将第t-1个回归器输出的关键点预测值S t-1和所述训练样本输 入到第t个回归器，获取第t个回归器输出的关键点预测值S t，所述第T个回归器输出的关键点预测值S T即为所述级联姿势回归模型预测的手掌关键点。
其中，回归器包括特征提取层、关键点增量估计层，在将第t-1个回归器输出的关键点预测值S t-1和所述训练样本输入到第t个回归器，获取第t个回归器输出的关键点预测值S t的具体过程为：
先将第t-1个回归器输出的关键点预测值S t-1和所述训练样本输入到特征提取层，获取特征提取层输出的特征向量；然后将特征向量输入到所述关键点增量估计层，获取第t个回归器输出的关键点增量值ΔS t；然后将所述第t-1个回归器输出的关键点预测值S t-1与所述第t个回归器输出的关键点增量值ΔS t相加，得到所述第t个回归器输出的关键点预测值S t。
第T个回归器输出的关键点预测值S T即为级联姿势回归模型预测的手掌关键点。
有时候初始化S 0选的不合适就会导致错误预测。所以可以对每一个图像，初始化K个不同姿态，级联神经网络模型运行K次得到K个不同结果。对这些结果进行聚类，选取密度最高的结果作为最终结果。即得到K个食指指根、小指指根和中指指尖的预测值，以每个预测值为圆心，以eps为半径画个圈(称为邻域eps-neigbourhood)，然后数有多少个点在这个圈内，这个数就是该点密度值。选取一个密度阈值MinPts，如圈内点数小于MinPts的圆心点为低密度的点，而大于或等于MinPts的圆心点高密度的点(称为核心点Core point)。则取密度最高的圆心点作为最终预测的手掌关键点。具体算法为DBSCAN算法。
S133、计算所述标注的手掌关键点和所述预测的手掌关键点之间的距离；
计算预测的手掌关键点与标注的手掌关键点之间的距离，这里指对应关键点之间的距离，例如标注的手掌关键点为食指指根、中指指尖、小指指根，则计算预测的食指指根与标注的食指指根的距离，预测的小指指根与标注的小指指根的距离，预测的中指指尖与标注中指指尖的距离。两个点之间的距离即两个坐标的差的平方。
S134、调整所述级联姿势回归模型各节点的参数，至所述标注的手掌关键点和所述预测的手掌关键点之间的距离最小时，训练结束。
调整级联姿势回归模型各节点的参数，即调整级联的各卷积神经网络模型各节点的权重，使预测的手掌关键点与标注的手掌关键的之间的距离最小时，训练结束，此时，认为各节点参数定义的级联姿势回归模型具备预测手掌关键 点的功能。
为解决上述技术问题本申请实施例还提供一种手掌图像裁剪装置。具体请参阅图6，图6为本实施例手掌图像裁剪装置的基本结构框图。
如图6所示，一种手掌图像裁剪装置，包括：获取模块210、处理模块220、计算模块230和执行模块240，其中获取模块210，用于获取待处理的图像；处理模块220，用于将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；计算模块230，用于根据所述手掌关键点，计算所述图像中手掌的掌心点C；执行模块240，用于以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
本申请实施例通过获取待处理的图像；将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；根据所述手掌关键点，计算所述图像中手掌的掌心点C；以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。经过裁剪后的图像包含了用于掌纹识别的关键点，且通过关键点可以确定手的姿态，同时避免了复杂背景的干扰，降低了掌纹识别的误差率。
在一些实施方式中，计算模块230还包括：中点确定子模块，用于确定以食指指根A和所述小指指根B为端点构成的线段AB的中点E；掌心确定子模块，用于以所述中点E为端点，以垂直于所述线段AB的方向，确定掌心点C，使CE的长度至少为所述线段AB长度的1/2，且掌心点C与所述中指指尖D位于所述线段AB的两边。
在一些实施方式中，所述处理模块220还包括：第一检测子模块，用于将所述图像输入到预设的目标检测模型中，以去除所述图像背景，获取所述目标检测模型响应所述图像而输出的待识别的手掌图像，其中预设的目标检测模型采用SSD目标检测算法预测图像中手的位置；第一处理子模块，用于将所述手掌图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述手掌图像而输出的手掌关键点。
在一些实施方式中，所述手掌图像裁剪装置还包括：第一计算子模块，用 于根据所述中指指尖D、食指指根A和小指指根B计算所述图像中手掌的倾斜度；第二处理子模块，用于根据所述倾斜度，对所述图像进行旋转。
在一些实施方式中，所述执行模块240中还包括：第三处理子模块，用于以所述食指指根A和所述小指指根B之间的连线AB为矩形的其中一条边；以所述食指指根A或所述小指指根B为一个端点，以所述掌心点C为中心，以连线AC或者BC长度的2倍为矩形的对角线的矩形对所述图像进行裁剪。
在一些实施方式中，所述手掌图像裁剪装置还包括：第一获取子模块，用于获取训练样本，所述训练样本为标注了手掌关键点的手掌图片集，其中，每个样本标注的手掌关键点至少包含食指指根、小指指根和中指指尖；第四处理子模块，用于将所述训练样本输入到级联姿势回归模型获取所述级联姿势回归模型响应所述训练样本而预测的手掌关键点；第二计算子模块，用于计算所述标注的手掌关键点和所述预测的手掌关键点之间的距离；第一调整子模块，用于调整所述级联姿势回归模型各节点的参数，至所述标注的手掌关键点和所述预测的手掌关键点之间的距离最小时，训练结束。
在一些实施方式中，所述手掌图像裁剪装置中的级联姿势回归模型包含T个回归器，T为大于1的正整数，所述第四处理子模块中包括：第一预测子模块，用于当t＝1时，将预设的关键点初始化值S 0和所述训练样本输入到第1个回归器，获取第1个回归器输出的关键点预测值S 1；第二预测子模块，用于当t＝2～T时，将第t-1个回归器输出的关键点预测值S t-1和所述训练样本输入到第t个回归器，获取第t个回归器输出的关键点预测值S t，所述第T个回归器输出的关键点预测值S T即为所述级联姿势回归模型预测的手掌关键点。
为解决上述技术问题，本申请实施例还提供计算机设备。具体请参阅图7，图7为本实施例计算机设备基本结构框图。
如图7所示，计算机设备的内部结构示意图。如图7所示，该计算机设备包括通过系统总线连接的处理器、非易失性存储介质、存储器和网络接口。其中，该计算机设备的非易失性存储介质存储有操作系统、数据库和计算机可读指令，数据库中可存储有控件信息序列，该计算机可读指令被处理器执行时，可使得处理器实现一种手掌图像裁剪的方法。该计算机设备的处理器用于提供计算和控制能力，支撑整个计算机设备的运行。该计算机设备的存储器中可存储有计算机可读指令，该计算机可读指令被处理器执行时，可使得处理器执行一种手掌图像裁剪的方法。该计算机设备的网络接口用于与终端连接通信。本 领域技术人员可以理解，图7中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
本实施方式中处理器用于执行图6中获取模块210、处理模块220、计算模块230和执行模块240的具体内容，存储器存储有执行上述模块所需的计算机可读指令和各类数据。网络接口用于向用户终端或服务器之间的数据传输。本实施方式中的存储器存储有手掌图像裁剪方法中执行所有子模块所需的计算机可读指令及数据，服务器能够调用服务器的计算机可读指令及数据执行所有子模块的功能。
计算机设备通过获取待处理的图像；将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；根据所述手掌关键点，计算所述图像中手掌的掌心点C；以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。经过裁剪后的图像包含了用于掌纹识别的关键点，且通过关键点可以确定手的姿态，同时避免了复杂背景的干扰，降低了掌纹识别的误差率。
本申请还提供一种存储有计算机可读指令的存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行上述任一实施例所述手掌图像裁剪方法的步骤。
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，该计算机可读指令可存储于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被执行时，可包括如上述各方法的实施例的流程。其中，前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性存储介质，或随机存储记忆体(Random Access Memory，RAM)等。
应该理解的是，虽然附图的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，其可以以其他的顺序执行。而且，附图的流程图中的至少一部分步骤可以包括多个子步骤或者多个阶段，这 些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，其执行顺序也不必然是依次进行，而是可以与其他步骤或者其他步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
以上所述仅是本申请的部分实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本申请的保护范围。

Claims (20)

  1. 一种手掌图像裁剪方法，其特征在于，包括下述步骤：
    获取待处理的图像；
    将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
    根据所述手掌关键点，计算所述图像中手掌的掌心点C；以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
  2. 根据权利要求1所述的手掌图像裁剪方法，其特征在于，在所述根据所述手掌关键点，计算所述图像中手掌的掌心点C的步骤中，包括下述步骤：
    确定以所述食指指根A和所述小指指根B为端点构成的线段AB的中点E；
    以所述中点E为端点，以垂直于所述线段AB的方向，确定掌心点C，使CE的长度至少为所述线段AB长度的1/2，且掌心点C与所述中指指尖D位于所述线段AB的两边。
  3. 根据权利要求1所述的手掌图像裁剪方法，其特征在于，在所述将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点的步骤中，包括下述步骤：
    将所述图像输入到预设的目标检测模型中，以去除所述图像背景，获取所述目标检测模型响应所述图像而输出的待识别的手掌图像，其中预设的目标检测模型采用SSD目标检测算法预测图像中手的位置；
    将所述手掌图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述手掌图像而输出的手掌关键点。
  4. 根据权利要求1所述的手掌图像裁剪方法，其特征在于，在所述根据所述手掌关键点，计算所述图像中手掌的掌心点C的步骤之后；所述以掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪的步骤之前，还包括下述步骤：
    根据所述中指指尖D、食指指根A和小指指根B计算所述图像中手掌的倾斜度；
    根据所述倾斜度，对所述图像进行旋转。
  5. 根据权利要求3所述的手掌图像裁剪方法，其特征在于，在所述以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪的步骤中，包括下述步骤：
    以所述食指指根A和所述小指指根B之间的连线AB为矩形的其中一条边；
    以所述食指指根A或所述小指指根B为一个端点，以所述掌心点C为中心，以连线AC或者BC长度的2倍为矩形的对角线对所述图像进行裁剪。
  6. 根据权利要求1～5所述的手掌图像裁剪方法，其特征在于，所述级联姿势回归模型按照下述步骤训练：
    获取训练样本，所述训练样本为标注了手掌关键点的手掌图片集，其中，每个样本标注的手掌关键点至少包含食指指根、小指指根和中指指尖；
    将所述训练样本输入到级联姿势回归模型获取所述级联姿势回归模型响应所述训练样本而预测的手掌关键点；
    计算所述标注的手掌关键点和所述预测的手掌关键点之间的距离；
    调整所述级联姿势回归模型各节点的参数，至所述标注的手掌关键点和所述预测的手掌关键点之间的距离最小时，训练结束。
  7. 根据权利要求6所述的手掌图像裁剪方法，其特征在于，所述级联姿势回归模型包含T个回归器，T为大于1的正整数，在所述将所述训练样本输入到级联姿势回归模型获取所述级联姿势回归模型响应所述训练样本而预测的手掌关键点的步骤中，包括下述步骤：
    当t＝1时，将预设的关键点初始化值S 0和所述训练样本输入到第1个回归器，获取第1个回归器输出的关键点预测值S 1；
    当t＝2～T时，将第t-1个回归器输出的关键点预测值S t-1和所述训练样本输入到第t个回归器，获取第t个回归器输出的关键点预测值S t，所述第T个回归器输出的关键点预测值S T即为所述级联姿势回归模型预测的手掌关键点。
  8. 一种手掌图像裁剪装置，其特征在于，包括：
    获取模块，用于获取待处理的图像；
    处理模块，用于将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
    计算模块，用于根据所述手掌关键点，计算所述图像中手掌的掌心点C；
    执行模块，用于以所述掌心点C为中心，以至少包含所述食指指根A和所 述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
  9. 根据权利要求8所述的手掌图像裁剪装置，其特征在于，所述计算模块包括：
    中点确定子模块，用于确定以所述食指指根A和所述小指指根B为端点构成的线段AB的中点E；
    掌心确定子模块，用于以所述中点E为端点，以垂直于所述线段AB的方向，确定掌心点C，使CE的长度至少为所述线段AB长度的1/2，且掌心点C与所述中指指尖D位于所述线段AB的两边。
  10. 根据权利要求8所述的手掌图像裁剪装置，其特征在于，所述处理模块：
    第一检测子模块，用于将所述图像输入到预设的目标检测模型中，以去除所述图像背景，获取所述目标检测模型响应所述图像而输出的待识别的手掌图像，其中预设的目标检测模型采用SSD目标检测算法预测图像中手的位置；
    第一处理子模块，用于将所述手掌图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述手掌图像而输出的手掌关键点。
  11. 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，所述计所述处理器执行所述计算机可读指令时，实现如下手掌图像裁剪方法的步骤：
    获取待处理的图像；
    将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
    根据所述手掌关键点，计算所述图像中手掌的掌心点C；以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
  12. 根据权利要求11所述的计算机设备，其特征在于，在所述根据所述手掌关键点，计算所述图像中手掌的掌心点C的步骤中，包括下述步骤：
    确定以所述食指指根A和所述小指指根B为端点构成的线段AB的中点E；
    以所述中点E为端点，以垂直于所述线段AB的方向，确定掌心点C，使CE的长度至少为所述线段AB长度的1/2，且掌心点C与所述中指指尖D位于 所述线段AB的两边。
  13. 根据权利要求11所述的计算机设备，其特征在于，在所述将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点的步骤中，包括下述步骤：
    将所述图像输入到预设的目标检测模型中，以去除所述图像背景，获取所述目标检测模型响应所述图像而输出的待识别的手掌图像，其中预设的目标检测模型采用SSD目标检测算法预测图像中手的位置；
    将所述手掌图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述手掌图像而输出的手掌关键点。
  14. 根据权利要求11所述的计算机设备，其特征在于，在所述根据所述手掌关键点，计算所述图像中手掌的掌心点C的步骤之后；所述以掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪的步骤之前，还包括下述步骤：
    根据所述中指指尖D、食指指根A和小指指根B计算所述图像中手掌的倾斜度；
    根据所述倾斜度，对所述图像进行旋转。
  15. 根据权利要求13所述的计算机设备，其特征在于，在所述以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪的步骤中，包括下述步骤：
    以所述食指指根A和所述小指指根B之间的连线AB为矩形的其中一条边；
    以所述食指指根A或所述小指指根B为一个端点，以所述掌心点C为中心，以连线AC或者BC长度的2倍为矩形的对角线对所述图像进行裁剪。
  16. 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处理器执行时实现如下手掌图像裁剪方法的步骤：
    获取待处理的图像；
    将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点，其中，所述级联姿势回归模型输出的手掌关键点至少包括食指指根A、小指指根B、中指指尖D；
    根据所述手掌关键点，计算所述图像中手掌的掌心点C；以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪，获取所述待处理图像中手掌的掌纹图像。
  17. 根据权利要求16所述的非易失性可读存储介质，其特征在于，在所述根据所述手掌关键点，计算所述图像中手掌的掌心点C的步骤中，包括下述步骤：
    确定以所述食指指根A和所述小指指根B为端点构成的线段AB的中点E；
    以所述中点E为端点，以垂直于所述线段AB的方向，确定掌心点C，使CE的长度至少为所述线段AB长度的1/2，且掌心点C与所述中指指尖D位于所述线段AB的两边。
  18. 根据权利要求16所述的非易失性可读存储介质，其特征在于，在所述将所述图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述图像而输出的手掌关键点的步骤中，包括下述步骤：
    将所述图像输入到预设的目标检测模型中，以去除所述图像背景，获取所述目标检测模型响应所述图像而输出的待识别的手掌图像，其中预设的目标检测模型采用SSD目标检测算法预测图像中手的位置；
    将所述手掌图像输入到预先训练的级联姿势回归模型，获取所述级联姿势回归模型响应所述手掌图像而输出的手掌关键点。
  19. 根据权利要求16所述的非易失性可读存储介质，其特征在于，在所述根据所述手掌关键点，计算所述图像中手掌的掌心点C的步骤之后；所述以掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪的步骤之前，还包括下述步骤：
    根据所述中指指尖D、食指指根A和小指指根B计算所述图像中手掌的倾斜度；
    根据所述倾斜度，对所述图像进行旋转。
  20. 根据权利要求18所述的非易失性可读存储介质，其特征在于，在所述以所述掌心点C为中心，以至少包含所述食指指根A和所述小指指根B的圆形或矩形对所述图像进行裁剪的步骤中，包括下述步骤：
    以所述食指指根A和所述小指指根B之间的连线AB为矩形的其中一条边；
    以所述食指指根A或所述小指指根B为一个端点，以所述掌心点C为中心，以连线AC或者BC长度的2倍为矩形的对角线对所述图像进行裁剪。
PCT/CN2019/103179 2019-05-15 2019-08-29 手掌图像裁剪方法、装置、计算机设备及存储介质 Ceased WO2020228181A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910401330.4 2019-05-15
CN201910401330.4A CN110287775B (zh) 2019-05-15 2019-05-15 手掌图像裁剪方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020228181A1 true WO2020228181A1 (zh) 2020-11-19

Family

ID=68002134

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/103179 Ceased WO2020228181A1 (zh) 2019-05-15 2019-08-29 手掌图像裁剪方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN110287775B (zh)
WO (1) WO2020228181A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112836597A (zh) * 2021-01-15 2021-05-25 西北大学 基于级联并行卷积神经网络的多手姿态关键点估计方法
CN114694242A (zh) * 2020-12-25 2022-07-01 华为技术有限公司 Ai识别方法、电子设备和摄像头
CN115273282A (zh) * 2022-07-26 2022-11-01 宁波芯然科技有限公司 一种基于掌静脉识别的车门解锁方法
CN116188903A (zh) * 2023-01-29 2023-05-30 北京达佳互联信息技术有限公司 一种图像处理模型的训练方法、图像处理方法及相关设备
WO2025139304A1 (zh) * 2023-12-29 2025-07-03 广电运通集团股份有限公司 手掌区域检测方法和装置

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111523435A (zh) * 2020-04-20 2020-08-11 安徽中科首脑智能医疗研究院有限公司 一种基于目标检测ssd的手指检测方法、系统及存储介质
CN111814535B (zh) * 2020-05-20 2024-05-07 平安科技（深圳）有限公司 掌纹图像的识别方法、装置、设备及计算机可读存储介质
CN113743169B (zh) * 2020-05-29 2023-11-07 北京达佳互联信息技术有限公司 手掌平面检测方法、装置、电子设备及存储介质
CN113780201B (zh) * 2021-09-15 2022-06-10 墨奇科技（北京）有限公司 手部图像的处理方法及装置、设备和介质
CN114170620B (zh) * 2021-11-05 2025-03-14 深圳市鸿合创新信息技术有限责任公司 手掌位置确定方法、装置、电子设备及存储介质
CN120821369B (zh) * 2025-07-02 2026-04-21 贵州理工学院 一种自然人机交互的手语识别方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120051605A1 (en) * 2010-08-24 2012-03-01 Samsung Electronics Co. Ltd. Method and apparatus of a gesture based biometric system
CN104573615A (zh) * 2013-10-24 2015-04-29 华为技术有限公司 掌纹采集方法及装置
CN107016323A (zh) * 2016-01-28 2017-08-04 厦门中控生物识别信息技术有限公司 一种手掌感兴趣区域的定位方法及装置
CN108427942A (zh) * 2018-04-22 2018-08-21 广州麦仑信息科技有限公司 一种基于深度学习的手掌检测与关键点定位方法
CN109472213A (zh) * 2018-10-16 2019-03-15 平安科技（深圳）有限公司 掌纹识别方法、装置、计算机设备和存储介质

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10192098B2 (en) * 2016-09-09 2019-01-29 MorphoTrak, LLC Palm print image matching techniques

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120051605A1 (en) * 2010-08-24 2012-03-01 Samsung Electronics Co. Ltd. Method and apparatus of a gesture based biometric system
CN104573615A (zh) * 2013-10-24 2015-04-29 华为技术有限公司 掌纹采集方法及装置
CN107016323A (zh) * 2016-01-28 2017-08-04 厦门中控生物识别信息技术有限公司 一种手掌感兴趣区域的定位方法及装置
CN108427942A (zh) * 2018-04-22 2018-08-21 广州麦仑信息科技有限公司 一种基于深度学习的手掌检测与关键点定位方法
CN109472213A (zh) * 2018-10-16 2019-03-15 平安科技（深圳）有限公司 掌纹识别方法、装置、计算机设备和存储介质

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114694242A (zh) * 2020-12-25 2022-07-01 华为技术有限公司 Ai识别方法、电子设备和摄像头
CN114694242B (zh) * 2020-12-25 2024-06-04 华为技术有限公司 Ai识别方法、电子设备和摄像头
CN112836597A (zh) * 2021-01-15 2021-05-25 西北大学 基于级联并行卷积神经网络的多手姿态关键点估计方法
CN112836597B (zh) * 2021-01-15 2023-10-17 西北大学 基于级联并行卷积神经网络的多手姿态关键点估计方法
CN115273282A (zh) * 2022-07-26 2022-11-01 宁波芯然科技有限公司 一种基于掌静脉识别的车门解锁方法
CN115273282B (zh) * 2022-07-26 2024-05-17 宁波芯然科技有限公司 一种基于掌静脉识别的车门解锁方法
CN116188903A (zh) * 2023-01-29 2023-05-30 北京达佳互联信息技术有限公司 一种图像处理模型的训练方法、图像处理方法及相关设备
WO2025139304A1 (zh) * 2023-12-29 2025-07-03 广电运通集团股份有限公司 手掌区域检测方法和装置

Also Published As

Publication number Publication date
CN110287775A (zh) 2019-09-27
CN110287775B (zh) 2023-09-22

Similar Documents

Publication Publication Date Title
WO2020228181A1 (zh) 手掌图像裁剪方法、装置、计算机设备及存储介质
CN108205655B (zh) 一种关键点预测方法、装置、电子设备及存储介质
US11868523B2 (en) Eye gaze classification
US12046012B2 (en) Face detection method, apparatus, and device, and storage medium
US10043308B2 (en) Image processing method and apparatus for three-dimensional reconstruction
US12051273B2 (en) Method for recognizing actions, device and storage medium
WO2021012526A1 (zh) 人脸识别模型的训练方法、人脸识别方法、装置、设备及存储介质
CN111062263B (zh) 手部姿态估计的方法、设备、计算机设备和存储介质
CN110765860A (zh) 摔倒判定方法、装置、计算机设备及存储介质
WO2020233010A1 (zh) 基于可分割卷积网络的图像识别方法、装置及计算机设备
WO2020019926A1 (zh) 特征提取模型训练方法、装置、计算机设备及计算机可读存储介质
WO2022037541A1 (zh) 图像处理模型训练方法、装置、设备及存储介质
CN114186632A (zh) 关键点检测模型的训练方法、装置、设备、存储介质
CN111695458B (zh) 一种视频图像帧处理方法及装置
CN111414946B (zh) 基于人工智能的医疗影像的噪声数据识别方法和相关装置
CN111738357A (zh) 垃圾图片的识别方法、装置及设备
US20210216913A1 (en) Prediction method, apparatus, and system for performing an image search
US20130236068A1 (en) Calculating facial image similarity
CN110555526B (zh) 神经网络模型训练方法、图像识别方法和装置
CN113065593A (zh) 模型训练方法、装置、计算机设备和存储介质
WO2021017289A1 (zh) 在视频中定位对象的方法、装置、计算机设备及存储介质
US20250148767A1 (en) Training method for image processing network, and image processing method and apparatus
CN111915480A (zh) 生成特征提取网络的方法、装置、设备和计算机可读介质
WO2025185480A1 (zh) 一种特征提取模型的训练方法、应用方法和相关装置
EP4724915A1 (en) Computer memory access for machine learning models

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: 19928684

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 14.01.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19928684

Country of ref document: EP

Kind code of ref document: A1