Iris Image Acquisition Method, Iris Recognition Method and Device

Abstract

The present invention provides an iris image acquisition method, an iris recognition method and an apparatus. Among them, the acquisition method includes: obtaining a visible light image and a near-infrared image of an object to be acquired, and performing coordinate mapping to obtain a coordinate mapping relationship; performing face detection on the visible light image, and if a face detection result is obtained, cropping the iris image in the near-infrared image in combination with the coordinate mapping relationship; inputting the iris image into a deep convolutional network to obtain optimized image acquisition parameters including lighting parameters; obtaining optimized pan-tilt parameters by using reinforcement learning according to the face detection result; adjusting the pan-tilt of the acquisition device by using the optimized pan-tilt parameters, and acquiring a visible light face image and a near-infrared face image based on the optimized image acquisition parameters; performing face detection on the visible light face image and cropping the iris image from the near-infrared face image according to the face detection result of the visible light face image. Through the above solution, the quality of iris acquisition can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image acquisition and pattern recognition, and in particular, to an iris image acquisition method, an iris recognition method, and a device. Background Art

[0002] With the gradual maturity of biometric recognition technology, iris recognition technology has been widely used in identity management such as epidemic prevention and control security, public security criminal investigation, suspect screening, mine personnel attendance, and entry and exit inspection due to its advantages of high accuracy, good uniqueness, and strong anti-counterfeiting performance. However, with the complexity of the iris recognition environment, fixed-distance and close-range iris recognition can no longer meet the social needs, and long-distance iris recognition technology has become the development trend.

[0003] Iris acquisition is an important link in the iris recognition system. During close-range acquisition, due to the slow convergence of automatic exposure, the image is overexposed; during long-distance acquisition, due to the very small diameter of the iris and insufficient near-infrared light, it is impossible to obtain an iris image that meets the requirements of iris recognition. In addition, the position adjustment of the pan-tilt will directly affect the clarity of the iris. If the quality of the iris image is not good, it will lead to slow image acquisition speed, low recognition rate, poor user experience, and poor product usability.

[0004] Iris acquisition is a crucial step in the iris recognition system. However, it is not easy to acquire high-quality iris images. It requires a camera module that supports the acquisition of high-resolution images and good lighting conditions during acquisition to ensure that the acquired iris can be used for effective recognition.

[0005] The supplementary lighting method of a general iris acquisition device is continuous lighting or fixed-frequency pulsed lighting. An infrared light structure of one side or two sides in a left-right group or multiple groups is used to adjust the supplementary lighting intensity. However, this also brings problems such as high power consumption, poor heat dissipation, and uneven lighting. To reduce power consumption, an alternating left-right lighting scheme can be adopted. However, it is difficult for the lighting timing of the left and right infrared light sources to be synchronized with the exposure time of each frame of the image sensor, which will result in uneven infrared lighting during the exposure time of the corresponding image frame when the left and right light sources alternate, thus causing the phenomenon of bad image frames. Moreover, the adjustment of the pan-tilt is extremely likely to cause motion blur, resulting in a decrease in imaging clarity and unable to meet the scenario of iris recognition during user movement, with poor user experience.

[0006] Some iris acquisition methods and devices use two sets of light sources at different angles for supplementary lighting, and perform lighting enhancement according to distance prompts, achieving relatively good results within the range of 30 to 80 cm. However, during actual iris acquisition, it is easy for the position of the target person to move significantly before and during exposure, resulting in inaccurate focusing and exposure, and causing the iris image to be blurred. High-quality iris images are susceptible to light intensity, exposure parameters, gain, and focusing parameters. In addition, the appropriate imaging position and lighting scheme directly affect the quality of the iris image. Summary of the Invention

[0007] In view of this, the present invention provides an iris image acquisition method, an iris recognition method, and a device to improve the quality of iris acquisition.

[0008] To achieve the above object, the present invention is implemented by the following solutions:

[0009] According to one aspect of an embodiment of the present invention, an iris image acquisition method is provided, including:

[0010] Obtain a visible light image of the object to be acquired collected by a visible light image acquisition device and a near-infrared image of the object to be acquired collected by a near-infrared image acquisition device;

[0011] Perform coordinate mapping on the visible light image and the near-infrared image to obtain a coordinate mapping relationship;

[0012] Perform face detection on the visible light image, and obtain a face detection result when a face is detected;

[0013] According to the face detection result and the coordinate mapping relationship, extract an initial iris image from the near-infrared image;

[0014] Input the initial iris image into a pre-trained deep convolutional network image acquisition parameter tuning model to obtain tuned image acquisition parameters; the image acquisition parameters include lighting parameters, and the lighting parameters include infrared lighting parameters and / or visible light lighting parameters;

[0015] According to the face detection result and using a reinforcement learning pan-tilt parameter tuning model, obtain tuned pan-tilt parameters;

[0016] Use the tuned pan-tilt parameters to adjust the pan-tilt positions of the visible light image acquisition device and the near-infrared image acquisition device, and collect a visible light face image and a near-infrared face image of the object to be acquired based on the adjusted pan-tilt positions and the tuned image acquisition parameters;

[0017] Perform face detection on the visible light face image, and crop an iris image for iris recognition from the near-infrared face image according to the face detection result of the visible light face image.

[0018] In some embodiments, the lighting parameters include the lighting position and the lighting intensity; the image acquisition parameters further include the exposure time, the gain, and the focusing parameters; the pan-tilt parameters include moving left, moving right, rotating up, and rotating down.

[0019] In some embodiments, performing coordinate mapping on the visible light image and the near-infrared image to obtain a coordinate mapping relationship includes:

[0020] Perform coordinate mapping on the visible light image and the near-infrared image according to the position information of the visible light image acquisition device and the position information of the near-infrared image acquisition device to obtain a coordinate mapping relationship.

[0021] In some embodiments, the iris image acquisition method further includes:

[0022] In the case where no face is detected, re-acquire the visible light image of the object to be acquired re-acquired by the visible light image acquisition device and the near-infrared image of the object to be acquired re-acquired by the near-infrared image acquisition device.

[0023] In some embodiments, the face detection result of the visible light image includes the face detection box position information and the face key point position information;

[0024] Cropping the initial iris image in the near-infrared image according to the face detection result and the coordinate mapping relationship includes:

[0025] Crop the initial iris image in the near-infrared image according to the face key point position information in the face detection result and the coordinate mapping relationship;

[0026] Obtain the optimized pan-tilt parameters according to the face detection result and by using the reinforcement learning pan-tilt parameter optimization model, including:

[0027] Obtain the optimized pan-tilt parameters according to the face detection box position information in the face detection result and by using the reinforcement learning pan-tilt parameter optimization model.

[0028] In some embodiments, the iris image acquisition method is characterized in that it further includes: training a policy function to obtain a reinforcement learning pan-tilt parameter optimization model;

[0029] Training a policy function to obtain a reinforcement learning pan-tilt parameter optimization model includes:

[0030] Obtain a training sample set; wherein, the training samples in the training sample set include: the device spatial position environment corresponding to the visible light image and the true face detection frame of the corresponding visible light image.

[0031] Input the device spatial position environment corresponding to the visible light image in the training sample into the policy function, use the device spatial position corresponding to the visible light image as the environmental state in the policy function, use the adjustment method and amplitude of the pan-tilt head as possible actions, and use the change direction of the repetition rate of the face detection frame after executing the action relative to the true face detection frame before as the reward value to train the policy function, obtain the optimized value of the pan-tilt head parameters. After the repetition rate of the face detection frame after executing the action exceeds the set threshold, obtain the reinforcement learning pan-tilt head parameter optimization model according to the trained policy function.

[0032] In some embodiments, using the change direction of the repetition rate of the face detection frame after executing the action relative to the true face detection frame before as the reward value includes:

[0033] Use the method of representing the detection frame position by two pixel points on the diagonal of the detection frame to calculate the position of the face detection frame after executing the action, the position of the face detection frame before executing the action, and the position of the true face detection frame respectively.

[0034] Use the object detection algorithm to calculate the first repetition rate of the face detection frame before executing the action and the true face detection frame according to the position of the face detection frame before executing the action and the position of the true face detection frame.

[0035] Use the object detection algorithm to calculate the second repetition rate of the face detection frame after executing the action and the true face detection frame according to the position of the face detection frame after executing the action and the position of the true face detection frame.

[0036] Calculate the difference between the first repetition rate and the second repetition rate, and use the sign of this difference as the change direction of the repetition rate of the face detection frame after executing the action relative to the true face detection frame before, so as to obtain the reward value.

[0037] In some embodiments, the iris image acquisition method further includes: training the initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter optimization model.

[0038] Training the initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter optimization model includes:

[0039] Obtain a training sample set; wherein, the training samples in the training sample set include: iris images and known image acquisition parameters.

[0040] Input the iris images in the training samples into the initial deep convolutional network to obtain the predicted image acquisition parameters;

[0041] Calculate the loss based on the known image acquisition parameters and the predicted image acquisition parameters in the training samples, and feedback the calculated loss to the initial deep convolutional network to train the initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter tuning model.

[0042] According to another aspect of the embodiments of the present invention, there is provided an iris recognition method, including:

[0043] Collect the iris images of the object to be recognized by using the iris image acquisition method described in any of the above embodiments;

[0044] Use the collected iris images to recognize the identity of the object to be recognized.

[0045] In some embodiments, before using the collected iris images to recognize the identity of the object to be recognized, the method further includes:

[0046] Perform iris quality evaluation on the collected iris images. In the case where the iris quality evaluation result meets the iris recognition requirements, perform the step of using the collected iris images to recognize the identity of the object to be recognized. In the case where the iris quality evaluation result does not meet the iris recognition requirements, re-perform the step of collecting the iris images of the object to be recognized by using the iris image acquisition method described in any of the above embodiments; and / or,

[0047] Perform iris quality evaluation on the initial iris images obtained during the collection process of the iris image acquisition method. In the case where the iris quality evaluation result meets the iris recognition requirements, perform the step of using the collected iris images to recognize the identity of the object to be recognized.

[0048] In some embodiments, performing iris quality evaluation on the collected iris images includes:

[0049] Calculate the light intensity, clarity, and brightness distribution of the collected iris images;

[0050] Judge whether the iris quality evaluation result of the collected iris images meets the iris recognition requirements by judging whether the light intensity, clarity, and brightness distribution of the collected iris images meet the set requirements;

[0051] Performing iris quality evaluation on the initial iris images obtained during the collection process of the iris image acquisition method includes:

[0052] Calculate the light intensity, clarity, and brightness distribution of the initial iris images;

[0053] By judging whether the light intensity, clarity, and brightness distribution of the initial iris image meet the set requirements, it is determined whether the iris quality evaluation result of the initial iris image meets the requirements for iris recognition.

[0054] In some embodiments, calculating the light intensity, clarity, and brightness distribution of the acquired iris image includes:

[0055] The light intensity of the acquired iris image is obtained by calculating the average pixel value of the acquired iris image;

[0056] The acquired iris image is filtered using a Laplacian of Gaussian kernel, and the power value of the filtered iris image is used to calculate the clarity of the acquired iris image;

[0057] The brightness distribution of the acquired iris image is calculated using the pixel values and image size of the acquired iris image;

[0058] By judging whether the light intensity, clarity, and brightness distribution of the acquired iris image meet the set requirements, it is determined whether the iris quality evaluation result of the acquired iris image meets the requirements for iris recognition, including:

[0059] By judging whether the light intensity of the acquired iris image is greater than the set light intensity threshold, whether its clarity is greater than the set clarity threshold, and whether its brightness distribution meets the set brightness distribution requirements, it is determined whether the iris quality evaluation result of the acquired iris image meets the requirements for iris recognition;

[0060] Calculating the light intensity, clarity, and brightness distribution of the initial iris image includes:

[0061] The light intensity of the initial iris image is obtained by calculating the average pixel value of the initial iris image;

[0062] The initial iris image is filtered using a Laplacian of Gaussian kernel, and the power value of the filtered iris image is used to calculate the clarity of the initial iris image;

[0063] The brightness distribution of the initial iris image is calculated using the pixel values and image size of the initial iris image;

[0064] By judging whether the light intensity, clarity, and brightness distribution of the initial iris image meet the set requirements, it is determined whether the iris quality evaluation result of the initial iris image meets the requirements for iris recognition, including:

[0065] By judging whether the light intensity of the initial iris image is greater than the set light intensity threshold, whether its clarity is greater than the set clarity threshold, and whether its brightness distribution meets the set brightness distribution requirements, it is determined whether the iris quality evaluation result of the initial iris image meets the requirements for iris recognition.

[0066] According to another aspect of the embodiments of the present invention, there is also provided an iris image acquisition system, including:

[0067] A visible light image acquisition device for acquiring a visible light image of an object to be acquired;

[0068] A near-infrared image acquisition device for acquiring a near-infrared image of an object to be acquired;

[0069] A visible light fill light for performing visible light fill light on the object to be acquired according to visible light lighting parameters when acquiring the visible light image of the object to be acquired;

[0070] An infrared fill light for performing infrared fill light on the object to be acquired according to infrared lighting parameters when acquiring the near-infrared image of the object to be acquired;

[0071] A pan-tilt for supporting a visible light camera of the visible light image acquisition device, a near-infrared camera of the near-infrared image acquisition device, the visible light fill light, and the infrared fill light based on pan-tilt parameters;

[0072] An iris acquisition device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of the method in any of the above embodiments to acquire the visible light image and the near-infrared image, and optimize image acquisition parameters and the pan-tilt parameters; wherein, the lighting parameters in the image acquisition parameters include the visible light lighting parameters and the infrared lighting parameters.

[0073] In some embodiments, the infrared fill light includes multiple groups of infrared lights disposed around the near-infrared camera in the near-infrared image acquisition device, and the infrared lighting parameters include information on the groups of infrared lights; and / or,

[0074] The visible light fill light includes multiple groups of visible light lights disposed around the visible light camera in the visible light image acquisition device, and the visible light lighting parameters include information on the groups of visible light lights.

[0075] In some embodiments, the visible light image acquisition device and the near-infrared image acquisition device are VCM cameras, and the image acquisition parameters further include a focusing parameter, and the focusing parameter is the VCM drive current corresponding to the VCM distance of the VCM camera.

[0076] According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of the method in any of the above embodiments.

[0077] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in any of the above embodiments are implemented.

[0078] The iris image acquisition method, iris recognition method, iris image acquisition system, electronic device and computer-readable storage medium according to the embodiments of the present invention can optimize the lighting parameters by using a deep convolutional network to optimize the image acquisition parameters, so that the lighting can be quickly and appropriately performed during image acquisition, so that images can be acquired under uniform light and appropriate light intensity. In addition, the problem of slow automatic exposure convergence can be overcome by lighting. By using reinforcement learning to optimize the pan-tilt parameters, images can be quickly acquired at appropriate angles. In this way, iris images convenient for recognition can be quickly acquired in various lighting environments and various acquisition positions. Therefore, this solution can solve the problems existing in existing iris acquisition, such as slow exposure convergence, uneven infrared fill light, poor quality of iris images, and blurred iris images, thereby improving the quality and speed of iris acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts. In the drawings:

[0080] Figure 1 is a flowchart of an iris image acquisition method according to an embodiment of the present invention;

[0081] Figure 2 is a structural diagram of an iris image acquisition system according to an embodiment of the present invention;

[0082] Figure 3 is a flowchart of an iris image acquisition method according to a specific embodiment of the present invention;

[0083] Figure 4 is an example of an iris image acquisition process according to a specific embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0084] In order to make the objectives, technical solutions and advantages of the embodiments of the present invention clearer and more understandable, the following will further describe the embodiments of the present invention in detail with reference to the drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but not to limit the present invention.

[0085] It should be noted in advance that the descriptions of the following embodiments or examples, or the features mentioned therein, can be combined with the features in other embodiments or examples in the same or similar manner, or replace the features in other embodiments or examples, to form possible embodiments. Additionally, the term "including / containing" used herein refers to the existence of features, elements, steps, or components, but does not exclude the existence of one or more other features, elements, steps, or components.

[0086] With the development of iris acquisition technology, the distance of iris acquisition has evolved from a fixed distance to a range of 30 to 100 cm or even wider, laying the foundation for long-distance iris recognition. However, on the one hand, the intensity of infrared light attenuates with distance, resulting in phenomena such as overly high brightness in near-distance illumination images, insufficient lighting in far-distance lighting, weak lighting, and blurred iris textures. Increasing the number of light sources will also lead to an increase in power consumption. On the other hand, the adjustment process of the iris acquisition device (such as a pan-tilt head, lighting position) will also have a great impact on iris acquisition, causing uneven image brightness, motion blur, etc.

[0087] Existing automatic image exposure methods (adjusting exposure time and gain) not only have slow convergence, resulting in a large shift in the position of the target person during automatic exposure, inaccurate focusing and exposure, and obvious out-of-focus blur and motion blur in the image; when collecting at a long distance, increasing the exposure time easily leads to a large movement in the position of the target person before and during exposure, inaccurate focusing and exposure, blurred iris images, and further increasing the gain may increase noise, resulting in a decline in the quality of the iris image.

[0088] In view of the above problems, the present invention provides an iris image acquisition method to avoid the situation where the position of the target person has changed before and during exposure due to slow exposure convergence, resulting in inaccurate iris focusing and exposure. At the same time, it solves the problems of image blur and uneven brightness caused by inaccurate pan-tilt head adjustment and lighting position, so as to stably obtain high-quality iris images.

[0089] Figure 1 It is a flowchart of the iris image acquisition method according to an embodiment of the present invention. As Figure 1 shown, the iris image acquisition methods of these embodiments may specifically include the following steps S110 to step S180.

[0090] The specific implementation manners of steps S110 to S180 will be described in detail below.

[0091] Step S110: Obtain the visible light image of the object to be collected acquired by the visible light image acquisition device and the near-infrared image of the object to be collected acquired by the near-infrared image acquisition device.

[0092] In step S110, the visible light image acquisition device may be a visible light camera, or it may be a VCM camera (a camera with voice coil driven lens zoom). The near-infrared image acquisition device may be an infrared camera, or it may be a VCM camera. Obtaining the visible light image and the near-infrared image may refer to receiving the visible light image and the near-infrared image, or it may refer to acquiring the visible light image and the near-infrared image through the corresponding cameras.

[0093] Step S120: Perform coordinate mapping on the visible light image and the near-infrared image to obtain a coordinate mapping relationship.

[0094] Specifically, step S120 may include steps: S121, perform coordinate mapping on the visible light image and the near-infrared image according to the position information of the visible light image acquisition device and the position information of the near-infrared image acquisition device to obtain a coordinate mapping relationship.

[0095] In step S121, the position information of the visible light image acquisition device and the position information of the near-infrared image acquisition device can indicate the relative positions of the visible light image acquisition device and the near-infrared image acquisition device. Based on this, the corresponding positions of the object to be acquired in the visible light image and the near-infrared image can be obtained, that is, the coordinate mapping relationship between the visible light image and the near-infrared image is obtained.

[0096] In other embodiments, the visible light image and the near-infrared image can be recognized respectively through recognition algorithms to obtain the coordinate mapping relationship between the two.

[0097] Through step S120, it is convenient to perform face detection using the visible light image, and then according to the face position in the visible light image, information such as the face position in the near-infrared image can be found.

[0098] Step S130: Perform face detection on the visible light image. When a face is detected, a face detection result is obtained.

[0099] In step S130, existing algorithms can be used to perform face detection on the visible light image, and information related to the face position in the visible light image can be obtained. The obtained face detection result may include face frame position information and face key point position information. Among them, the face frame position information can be represented by the vertex positions of the face rectangle frame, such as the positions of two diagonal vertices. The face key point position information may include the positions of human eyes, the corners of the eyes, etc., based on which it is convenient to find the iris position in the near-infrared image.

[0100] In a further embodiment Figure 1The iris image acquisition method shown may further include the step: S190, in the case where no face is detected, re-acquire the visible light image of the object to be acquired re-acquired by the visible light image acquisition device and the near-infrared image of the object to be acquired re-acquired by the near-infrared image acquisition device. In this step S190, if no face is detected, it can be considered that there is no one in front of the lens at this time or the face deviates from the lens, and the image can be re-acquired until a face is detected in the re-acquired image.

[0101] Step S140: According to the face detection result and the coordinate mapping relationship, extract the initial iris image from the near-infrared image.

[0102] In this step S140, the initial iris image refers to the image of the iris region. According to the coordinate mapping relationship, the position corresponding to the region corresponding to the face detection result in the visible light image can be found in the near-infrared image, so that the iris region in the near-infrared image can be found and the iris image can be obtained. Specifically, the boundary of the iris in the near-infrared image can be found, and then the iris image can be extracted or cropped from it.

[0103] In the case where the face detection result of the visible light image includes the position information of the face key points, the above step S140, that is, according to the face detection result and the coordinate mapping relationship, extract the initial iris image from the near-infrared image, may specifically include the steps: S141, according to the position information of the face key points in the face detection result and the coordinate mapping relationship, extract the initial iris image from the near-infrared image.

[0104] In this step S141, the position information of the face key points may include the inner corner position and the outer corner position of the eyes, etc. The iris image can be obtained quickly and accurately according to the position information of the face key points.

[0105] Step S150: Input the initial iris image into the pre-trained deep convolutional network image acquisition parameter tuning model to obtain the tuned image acquisition parameters; the image acquisition parameters include lighting parameters, and the lighting parameters include infrared lighting parameters and / or visible light lighting parameters.

[0106] In step S150, if the image acquisition parameters include infrared lighting parameters, infrared fill light can be provided to the acquisition object when the infrared image acquisition device acquires a near-infrared image according to these parameters, so as to acquire a clearer iris image. If the image acquisition parameters include visible light lighting parameters, visible light fill light can be provided to the acquisition object when the visible light image acquisition device acquires a visible light image according to these parameters, so as to acquire a clearer face image. The lighting parameters may include corresponding information on lighting intensity and / or lighting position information. Then, the infrared lighting parameters may include corresponding information on infrared lighting intensity and / or infrared lighting position information, and the visible light lighting parameters may include corresponding information on visible light lighting intensity and / or visible light lighting position information.

[0107] Specifically, the lighting parameters may include lighting position and lighting intensity. For example, the infrared lighting parameters may include infrared lighting position and infrared lighting intensity, and the visible light lighting parameters may include visible light lighting position and visible light lighting intensity. The image acquisition parameters may also include exposure time, gain, focusing parameters, etc.

[0108] In step S150, optimizing the image acquisition parameters through the deep convolutional network image acquisition parameter optimization model can enable various image acquisition parameters to converge quickly, thereby improving the convergence speed. Since the image acquisition parameters include lighting parameters, even under poor lighting conditions, images with appropriate light intensity and brightness distribution can be acquired quickly. Additionally, fill light can overcome the problem of slow automatic exposure convergence. Moreover, the image acquisition parameters may include exposure time, so that the appropriate exposure time parameter can be quickly adjusted, thereby enabling an image with better light intensity to be obtained. The image acquisition parameters may include gain, which can quickly adjust the gain parameter to obtain a clearer image. The image acquisition parameters may include focusing parameters, which can adjust the focusing parameters of the camera (such as visible light image acquisition device / camera, infrared image acquisition device / camera), so that the parameters can be adjusted well relatively quickly, improving the user experience.

[0109] In step S150, the deep convolutional network image acquisition parameter optimization model can be a trained model. The deep convolutional network can be trained to learn and predict image acquisition parameters to achieve purposes such as clear images and uniform brightness.

[0110] In some embodiments, Figure 1The iris image acquisition method shown may further include the steps of: S1100, training an initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter tuning model. This step S1100 may specifically include the steps of: S1101, obtaining a training sample set; wherein, the training samples in the training sample set include: iris images and known image acquisition parameters; S1102, inputting the iris images in the training samples into the initial deep convolutional network to obtain predicted image acquisition parameters; S1103, calculating a loss based on the known image acquisition parameters and the predicted image acquisition parameters in the training samples, and feeding back the calculated loss to the initial deep convolutional network to train the initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter tuning model.

[0111] In this step S1101, the known image acquisition parameters correspond to the image acquisition parameters to be output by this deep convolutional network image acquisition parameter tuning model. For example, it may include lighting parameters, and the lighting parameters may specifically include infrared lighting parameters and / or visible light lighting parameters. The known image acquisition parameters may also include exposure time, gain, and focusing parameters, etc. The lighting parameters may specifically include lighting position and lighting intensity. The infrared lighting parameters may include infrared lighting position and infrared lighting intensity, and the visible light lighting parameters may include visible light lighting position and visible light lighting intensity.

[0112] In this step S1103, for the lighting parameters, the loss can be calculated based on the difference between the known lighting position and the predicted lighting position; for the lighting parameters, the loss can also be calculated based on the difference between the known lighting intensity and the predicted lighting intensity; for the exposure time, the loss can be calculated based on the difference between the known exposure time and the predicted exposure time; for the gain, the loss can be calculated based on the difference between the known exposure time and the predicted exposure time; for the focusing parameters, the loss can be calculated based on the difference between the known focusing parameters and the predicted focusing parameters.

[0113] Step S160: Obtain the tuned pan-tilt parameters according to the face detection result and by using the reinforcement learning pan-tilt parameter tuning model.

[0114] In this step S160, the reinforcement learning pan-tilt parameter tuning model can be obtained from various reinforcement learning models, which can be a standard Markov decision process, a patterned reinforcement learning or a non-patterned reinforcement learning, and can be an active learning process or a passive learning process. For example, the reinforcement learning pan-tilt parameter tuning model can be a policy function.

[0115] The pan-tilt can be used to support various devices. One pan-tilt can support multiple devices, or different devices can be adjusted using different pan-tilts. Among them, the devices that can be adjusted by the pan-tilt include visible light image acquisition devices, infrared image acquisition devices, and can also include visible light floodlights, infrared floodlights, etc. The adjustable pan-tilt parameters can include moving left, moving right, rotating up, rotating down, etc. By moving left or right, the face image can be centered horizontally in the image and is not easily distorted. By rotating up or down, it can cooperate with the face looking up or looking down. In other embodiments, it can include moving left and moving right, or rotating up and rotating down, or can include rotating left and rotating right. By giving the pan-tilt parameters through reinforcement learning, it is possible to quickly adjust to a suitable image acquisition angle, improving the user experience.

[0116] When the face detection result of the visible light image includes the position information of the face detection frame, the above step S160, that is, according to the face detection result and using the reinforcement learning pan-tilt parameter tuning model to obtain the tuned pan-tilt parameters, specifically may include the steps: S161, according to the position information of the face detection frame in the face detection result and using the reinforcement learning pan-tilt parameter tuning model to obtain the tuned pan-tilt parameters. In this way, better pan-tilt parameters can be learned based on the position of the face frame to capture better images.

[0117] In some embodiments, the reinforcement learning pan-tilt parameter tuning model can be obtained by training a certain reinforcement learning model, such as a decision function. Figure 1 The iris image acquisition method shown may further include the step: S1110, training the policy function to obtain the reinforcement learning pan-tilt parameter tuning model. This step S1110 specifically may include the steps: S1111, obtaining a training sample set; wherein, the training samples in the training sample set include: the device space position environment corresponding to the visible light image and the true face detection frame of the corresponding visible light image; S1112, inputting the device space position environment corresponding to the visible light image in the training sample into the policy function, using the device space position corresponding to the visible light image as the environmental state in the policy function, using the adjustment method and amplitude of the pan-tilt as possible actions, and using the change direction of the repetition rate of the face detection frame after executing the action relative to the true face detection frame before as the reward value to train the policy function to obtain the tuning value of the pan-tilt parameters. After the repetition rate of the face detection frame after executing the action exceeds the set threshold, the reinforcement learning pan-tilt parameter tuning model is obtained according to the trained policy function.

[0118] In this embodiment, the real face detection box can be obtained through manual annotation. One visible light image corresponds to one acquisition environment. Therefore, the device spatial position environment corresponding to the visible light image (such as the relative positions of the visible light image acquisition device and the infrared image acquisition device, and may also include the relative positions between the two of them and the visible light spotlight and the infrared spotlight respectively) can be used as the state of the environment in the policy function, and the actions that the pan-tilt can perform (such as the rotation direction and amplitude) can be used as the execution actions.

[0119] In the above step S1112, the change direction of the repetition rate of the face detection box after the execution action relative to that before the execution action is used as the reward value. More specifically, it may include the steps: S11121, using the method of representing the detection box position with two pixel points on the diagonal of the detection box, calculate the positions of the face detection box after the execution action, the position of the face detection box before the execution action, and the position of the real face detection box respectively; S11122, using the object detection algorithm, calculate the first repetition rate of the face detection box before the execution action and the real face detection box according to the positions of the face detection box before the execution action and the real face detection box; S11123, using the object detection algorithm, calculate the second repetition rate of the face detection box after the execution action and the real face detection box according to the positions of the face detection box after the execution action and the real face detection box; S11124, calculate the difference between the first repetition rate and the second repetition rate, and use the sign of this difference as the change direction of the repetition rate of the face detection box after the execution action relative to that before the execution action, so as to obtain the reward value.

[0120] In this step S11121, the position of the face detection box can be expressed as:

[0121] δ w =δ×(x 2 -x 1 ),δ h =δ×(y 2 -y 1 ),δ∈[0,1]

[0122] Wherein, δ w represents the width of the new face box, δ represents the coefficient, x 1 and y 1 represent the abscissa and ordinate of the first vertex of the new face box respectively, x 2 and y 2 represent the abscissa and ordinate of the second vertex of the new face box opposite to the first vertex respectively, and δ h represents the height of the new face box.

[0123] The reward value can be expressed as γ t= sign(IoU(b', g) - IoU(b, g)), where γ t represents the reward value, IoU(b, g) is the repetition rate (first repetition rate) of the face bounding box in the current frame and the true face bounding box, and IoU(b', g) is the repetition rate (second repetition rate) of the face bounding box in the next frame and the true face bounding box. Among them, b’ represents the face bounding box in the current frame, g represents the true face bounding box, and b represents the face bounding box in the next frame. The reward value γ t is a sign function γ t ∈ {0, 1}.

[0124] Step S170: Adjust the pan-tilt positions of the visible light image acquisition device and the near-infrared image acquisition device by using the optimized pan-tilt parameters, and acquire the visible light face image and the near-infrared face image of the object to be acquired based on the adjusted pan-tilt positions and the optimized image acquisition parameters.

[0125] In this step S170, the optimized pan-tilt parameters can be transmitted to the adjustment device of the pan-tilt to adjust the rotation action and amplitude of the pan-tilt. The optimized image acquisition parameters can be transmitted to the corresponding devices. For example, the exposure time, gain, and focusing parameters can be transmitted to the corresponding cameras (visible light image acquisition device, infrared image acquisition device) to adjust the corresponding cameras. The lighting parameters can be transmitted to the lighting lamp or lighting device to adjust the lighting position and / or intensity.

[0126] Step S180: Perform face detection on the visible light face image, and crop the iris image for iris recognition from the near-infrared face image according to the face detection result of the visible light face image.

[0127] In this step S180, after acquiring the appropriate face images through the foregoing steps, the iris image can be cropped from the near-infrared face image, which can be used for iris recognition or further evaluation.

[0128] In addition, an embodiment of the present invention further provides an iris recognition method, which can be used to perform iris recognition on the iris image obtained by the iris image acquisition method according to any of the foregoing embodiments.

[0129] The iris recognition methods of these embodiments may include the steps: S210, acquiring the iris image of the object to be recognized by using the iris image acquisition method according to any of the foregoing embodiments; S220, recognizing the identity of the object to be recognized by using the acquired iris image.

[0130] Further, in the iris recognition method described in the above embodiments, before the above step S220, that is, before recognizing the identity of the object to be recognized using the acquired iris image, the method may further include the steps: S230, performing iris quality evaluation on the acquired iris image; in the case where the iris quality evaluation result meets the iris recognition requirements, performing the step of recognizing the identity of the object to be recognized using the acquired iris image (i.e., step S220); in the case where the iris quality evaluation result does not meet the iris recognition requirements, re-performing the step of acquiring the iris image of the object to be recognized using the iris image acquisition method described in any of the above embodiments (i.e., step S210).

[0131] In a specific embodiment, in the above step S230, performing iris quality evaluation on the acquired iris image may specifically include the steps: S231, calculating the light intensity, sharpness, and brightness distribution of the acquired iris image; S232, determining whether the iris quality evaluation result of the acquired iris image meets the iris recognition requirements by determining whether the light intensity, sharpness, and brightness distribution of the acquired iris image meet the set requirements.

[0132] More specifically, in the above step S231, calculating the light intensity, sharpness, and brightness distribution of the acquired iris image may specifically include the steps: S2311, obtaining the light intensity of the iris image by calculating the average pixel value of the acquired iris image; S2312, filtering the acquired iris image using a Laplacian of Gaussian kernel, and calculating the sharpness of the acquired iris image using the power value of the filtered iris image; S2313, calculating the brightness distribution of the acquired iris image using the pixel values and image size of the acquired iris image.

[0133] The above step S232, that is, determining whether the iris quality evaluation result of the acquired iris image meets the iris recognition requirements by determining whether the light intensity, sharpness, and brightness distribution of the acquired iris image meet the set requirements, may specifically include the steps: S2321, determining whether the iris quality evaluation result of the acquired iris image meets the iris recognition requirements by determining whether the light intensity of the acquired iris image is greater than the set light intensity threshold, whether its sharpness is greater than the set sharpness threshold, and whether its brightness distribution meets the set brightness distribution requirements.

[0134] In these embodiments, by further performing quality judgment on the iris image acquired by the embodiments of the present invention, the quality of the iris image can be further ensured, thereby further improving the accuracy of iris recognition.

[0135] In some other embodiments, for the iris recognition method described in the above embodiments, before the above step S220, that is, before identifying the identity of the object to be recognized by using the acquired iris image, the method may further include the step: S2302, performing iris quality evaluation on the initial iris image obtained during the acquisition process of the iris image acquisition method. When the iris quality evaluation result meets the iris recognition requirements, execute the step of identifying the identity of the object to be recognized by using the acquired iris image (i.e., step S220). Additionally, when the requirements are not met, the remaining steps in the acquisition process of the iris image acquisition method may be continued to obtain the acquired iris image.

[0136] In a specific embodiment, in the above step S2302, performing iris quality evaluation on the initial iris image obtained during the acquisition process of the iris image acquisition method may specifically include the steps: S23021, calculating the light intensity, clarity, and brightness distribution of the initial iris image; S23022, determining whether the iris quality evaluation result of the initial iris image meets the iris recognition requirements by judging whether the light intensity, clarity, and brightness distribution of the initial iris image meet the set requirements.

[0137] More specifically, in the above step S23021, calculating the light intensity, clarity, and brightness distribution of the initial iris image may specifically include the steps: S23121, obtaining the light intensity of the iris image by calculating the mean pixel value of the initial iris image; S23221, filtering the initial iris image by using a Laplacian of Gaussian kernel, and calculating the clarity of the initial iris image by using the power value of the filtered iris image; S23321, calculating the brightness distribution of the initial iris image by using the pixel values and image size of the initial iris image.

[0138] In the above step S23022, determining whether the iris quality evaluation result of the initial iris image meets the iris recognition requirements by judging whether the light intensity, clarity, and brightness distribution of the initial iris image meet the set requirements may specifically include the steps: determining whether the iris quality evaluation result of the initial iris image meets the iris recognition requirements by judging whether the light intensity of the initial iris image is greater than the set light intensity threshold, whether its clarity is greater than the set clarity threshold, and whether its brightness distribution meets the set brightness distribution requirements.

[0139] In these embodiments, during the iris image acquisition process, quality assessment is performed on the initially acquired iris image. If the requirements are met, it can be directly used for iris recognition without having to complete the entire iris image acquisition process, thereby avoiding filtering out usable iris images and improving the speed of iris recognition.

[0140] For example, the light intensity can be expressed as:

[0141]

[0142] Among them, L represents the light intensity, I(x, y) represents the pixel value at the position of the image (x, y), and w and h respectively represent the width and height of the image.

[0143] The sharpness can be expressed as:

[0144]

[0145]

[0146]

[0147] x ∈ [1, 5,.....w], y ∈ [1, 5,.....h],

[0148] Among them, SHARPNESS represents the sharpness, power represents the power value, I F (x, y) represents the output after filtering the image I(x, y). w and h respectively represent the width and height of the image I(x, y). W F and H F respectively represent the width and height of the filtered image I F (x, y). i and j represent integer variables.

[0149] Based on the same inventive concept as the iris image acquisition method shown in Figure 1 , the embodiment of the present invention also provides an iris image acquisition system as described in the following embodiments. Since the principle of solving problems by this iris image acquisition system is similar to that of the iris image acquisition method, the implementation of this iris image acquisition system can refer to the implementation of the iris image acquisition method, and the repeated parts will not be elaborated.

[0150] Figure 2 is a schematic structural diagram of an iris image acquisition system according to an embodiment of the present invention. Refer to Figure 2 , the iris image acquisition system of these embodiments may include: a visible light image acquisition device 310, a near-infrared image acquisition device 320, a visible light fill light 330, an infrared fill light 340, a pan-tilt 350, and an iris acquisition device 360. The visible light image acquisition device 310 and the near-infrared image acquisition device 320 can both be cameras, such as VCM cameras. The visible light fill light 330 and the infrared fill light 340 can both be LED lights. The iris acquisition device 360 can be a device capable of implementing the iris image acquisition method described in any of the above embodiments.

[0151] The visible light image acquisition device 310 is used to acquire the visible light image of the object to be acquired.

[0152] A near-infrared image acquisition device 320 for acquiring near-infrared images of an object to be acquired.

[0153] A visible light fill light 330 for performing visible light fill light on an object to be acquired according to visible light lighting parameters when acquiring visible light images of the object to be acquired.

[0154] An infrared fill light 340 for performing infrared fill light on an object to be acquired according to infrared lighting parameters when acquiring near-infrared images of the object to be acquired.

[0155] A pan-tilt 350 for supporting a visible light camera of the visible light image acquisition device, a near-infrared camera of the near-infrared image acquisition device, the visible light fill light, and the infrared fill light based on pan-tilt parameters.

[0156] An iris acquisition device 360 includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the iris image acquisition method described in any of the above embodiments are implemented to acquire the visible light image and the near-infrared image, and optimize image acquisition parameters and the pan-tilt parameters; wherein, the lighting parameters in the image acquisition parameters include the visible light lighting parameters and the infrared lighting parameters.

[0157] In a further embodiment, the infrared fill light may include multiple groups of infrared lights disposed around the near-infrared camera in the near-infrared image acquisition device, and the infrared lighting parameters include information of the groups of infrared lights; and / or, the visible light fill light may include multiple groups of visible lights disposed around the visible light camera in the visible light image acquisition device, and the visible light lighting parameters include information of the groups of visible lights. Among them, lights at different positions can be controlled by different signals, so that lighting at different positions can be achieved. Lighting according to the lighting parameters can accurately and quickly perform lighting, can shorten the lighting time, reduce power consumption, and improve the quality of image acquisition.

[0158] In some embodiments, the visible light image acquisition device and the near-infrared image acquisition device may be VCM cameras, and the image acquisition parameters may further include focusing parameters, and the focusing parameters may be VCM drive currents corresponding to VCM distances of the VCM cameras. Among them, the VCM distance may refer to the distance that the voice coil motor in the VCM camera (voice coil camera) moves, or may refer to the object distance, or may refer to the image distance. The VCM drive current refers to the drive current for driving the voice coil motor to reach the corresponding distance. The corresponding relationship table between the VCM distance and the VCM drive current can be obtained through calibration, and during use, a convolutional network can be trained to learn to look up the table to learn and predict the VCM drive current to obtain the focusing parameters.

[0159] In addition, an embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the iris image acquisition method or the iris recognition method described in any of the above embodiments are implemented.

[0160] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the iris image acquisition method or the iris recognition method described in any of the above embodiments are implemented.

[0161] The above method will be described below in conjunction with a specific embodiment. However, it should be noted that this specific embodiment is only for better explaining the present application and does not constitute an improper limitation to the present application.

[0162] Figure 3 It is a schematic flowchart of the iris image acquisition method according to a specific embodiment of the present invention. Refer to Figure 3 This iris image acquisition method is implemented based on deep reinforcement learning and can adopt a closed-loop control method of online learning to ensure the accuracy of control parameters. The iris image acquisition method may include the following steps S1 to S7. Pan-tilt: A movable device that can be in four states of "up rotation", "down rotation", "left rotation" (left shift), and "right rotation" (right shift). The pan-tilt includes a visible light camera, a fill light, a near-infrared camera, and multiple groups of infrared lights surrounding the near-infrared camera.

[0163] Step S1: Visible light and near-infrared image coordinate mapping.

[0164] Since the sizes of the visible light image and the near-infrared image are different, the position coordinates of the two are first mapped, and then the corresponding near-infrared image can be cropped from a certain position of the visible light image.

[0165] Step S2: Pan-tilt, exposure, and lighting adjustment.

[0166] The acquisition device adjusts the pan-tilt position according to the parameters given by deep reinforcement learning, and adjusts the exposure parameters, focusing parameters, and lighting position, so as to guide the acquisition device to quickly reach the appropriate imaging position and set accurate image acquisition parameters for image acquisition.

[0167] It should be noted that existing supplementary lighting methods all adopt continuous lighting on one side or both sides, which easily causes phenomena such as increased power consumption and uneven brightness. The present embodiment adopts a new supplementary lighting device and supplementary lighting method. Among them, in this supplementary lighting device, multiple groups of infrared lamps can be arranged around the camera. This supplementary lighting method can determine the turning-on situation of infrared lamps at which positions according to the position and distance of the human face and the brightness distribution of the iris image. Multiple groups of infrared lights around the iris camera can turn on the lights at corresponding positions according to the corresponding parameters.

[0168] Step S3: Visible light face detection.

[0169] The visible light face detection algorithm adopted can be a deep learning network model trained with labeled face data. The input is, for example, a visible light image with a size of 480*640 pixels, and the output is, for example, the coordinates of the face detection box and the coordinates of the face key points.

[0170] Step S4: Deep reinforcement learning parameter tuning.

[0171] Deep reinforcement learning parameter tuning can include pan-tilt parameter tuning and image acquisition parameter tuning. Among them, the pan-tilt parameters can include the amplitudes of 5 states of the pan-tilt, namely "tilt up", "tilt down", "pan left", "pan right", and "stationary". The image acquisition parameters can include light intensity, exposure time, gain, focusing parameters, and lighting position. Among them, the focusing parameters can be a pre-calibrated "VCM (voice coil motor) distance-current comparison table", and the lighting position can be the infrared lamp turning-on situation determined according to the image brightness distribution.

[0172] Regarding deep reinforcement learning parameter tuning:

[0173] Reinforcement learning is a learning method based on trial and error, and it iteratively learns through the reward values feedback from the environment. Deep reinforcement learning adopts the structure of reinforcement learning plus a neural network. Compared with the value-based method, the biggest advantage of directly outputting actions by deep reinforcement learning is that it can select actions within a continuous interval and has better convergence.

[0174] The VGG-M model of the convolutional neural network can be adopted. The fully connected layer of the last layer of the network can be the policy function. That is, deep reinforcement learning parameter tuning includes two branches, the VGG-M model of the convolutional neural network and the policy function, which share the face detection results for training or prediction. The former outputs parameters such as exposure and lighting, and the latter outputs the pan-tilt position.

[0175] Through the approximation representation of the reinforcement learning network, the policy function is trained by the stochastic policy gradient method. The pan-tilt head can execute the best action according to the optimal policy function and determine the movement state of the pan-tilt head based on the target position of the previous frame. The face position box can be represented by two pixel points (the upper left corner and the lower right corner). The face position box l can be expressed as: l = [x 1 , y 1 , x 2 , y 2 , where x 1 and y 1 represent the abscissa and ordinate positions of the first vertex of the face position box, and x 2 and y 2 represent the abscissa and ordinate positions of the second vertex of the face position box opposite to the first vertex. After the pan-tilt head executes an action on the current face box, the position of the new face box obtained can be expressed as:

[0176] δ w = δ × (x 2 - x 1 ), δ h = δ × (y 2 - y 1 ), δ ∈ [0, 1]

[0177] where δ w represents the width of the new face box, δ represents the coefficient, x 1 and y 1 represent the abscissa and ordinate of the first vertex of the new face box, and x 2 and y 2 respectively represent the abscissa and ordinate of the second vertex of the new face box opposite to the first vertex, and δ h represents the height of the new face box.

[0178] During the training phase, the pan-tilt head will receive a reward value after executing an action. During the actual testing phase, the pan-tilt head will not receive a reward value but follow the policy function.

[0179] where the reward value γ t can be defined as:

[0180] γ t = sign(IoU(b', g) - IoU(b, g))

[0181] where IoU(b, g) is the repetition rate of the face box in the current frame and the true face box, and IoU(b', g) is the repetition rate of the face box in the next frame and the true face box. Among them, b’ represents the face box in the current frame, g represents the true face box, and b represents the face box in the next frame. The reward value γ t is a sign function γ t∈{0,1}, if the reward value feedback after the pan-tilt executes an action is +1, then continue to strengthen this action; if the reward value is -1, then punish this action, and gradually execute the action, for example, until IoU(b',g) > 0.95 to execute the stop action.

[0182] Regarding the training of the policy function:

[0183] Learn the policy function π=(a t |s t ; θ) in the last fully connected layer of the VGG-M model. Regard a single visible light face image as the environment and the pan-tilt as the agent. The pan-tilt reaches a suitable imaging position through a series of actions. Use s t to represent the state of the environment, a t to represent possible actions, r t to represent the immediate reward value. Therefore, a Markov decision process can be described as τ={...,(s t ,a t ,r t ),(s t+1 ,a t+1 ,r t+1 )...}. The purpose of training the policy function is to guide the pan-tilt to know which action or actions (such as "turn up", "turn down", "turn left", "turn right", "stay still") are the optimal choices on the premise of knowing the environmental state. Training the policy function requires an intermediate function, namely the state value function V(s t ; θ v ), which can be expressed as:

[0184] V(s t ; θ v ) = E[R t |s t = s],

[0185] where θ v represents the network parameters in the learning process, E represents the mathematical expectation, s represents the value of the environmental state s t . s t represents the target state at time t. R t represents the future total discounted reward function, which can be expressed as:

[0186]

[0187] where k represents the number of iterations in the training process, γ k represents the reward function of the action from t to t + k, and r t+k represents the reward value of the action from t to t + k.

[0188] Apply the stochastic gradient descent method to update the parameters θ and θ v, can be expressed as:

[0189]

[0190] θ v ←θ v -α(R t -V(s t ))

[0191] Among them, θ represents the network parameter, represents taking the gradient with respect to θ, α represents the action variable, β represents the hyperparameter, and θ v represents the network parameter under the state value v. H represents entropy and can be expressed as:

[0192]

[0193] Among them, a represents the action variable, s represents the state variable of the environment, a t represents executing the t action, and s t represents the state variable of the environment when executing the t action. The hyperparameter β controls the strength of entropy regularization. Stop iterating until the total discounted future reward function R t no longer increases.

[0194] Regarding the deep convolutional network model:

[0195] (1) According to the iris acquisition device, set different light intensities, exposure parameters, gains, focusing parameters, and lighting strategies, and acquire iris images with a size of 1920*1080 pixels, so as to prepare positive and negative sample training sets, validation sets, and test sets with labels.

[0196] (2) From the test set prepared in (1), select lightweight deep learning networks such as (AlexNet, MobileNet, etc.), connect multiple fully connected layers after the convolutional layer, and perform multi-task training on images of different qualities.

[0197] (3) Apply the model trained in (2) to actual iris acquisition.

[0198] Step S5: Iris acquisition refers to cropping a near-infrared iris image with a size of 1920*1080 pixels according to the face key points given by face detection.

[0199] Step S6: Iris image quality evaluation is to verify the quality of the currently acquired iris image. By calculating the light intensity, brightness distribution, and clarity of the iris image, it is judged whether the current iris image can be used for iris recognition.

[0200] (1) Calculation of light intensity:

[0201]

[0202] Among them, L represents the light intensity, I(x, y) represents the pixel value at the position (x, y) of the image, and w and h respectively represent the width and height of the image.

[0203] (2) Calculation of sharpness:

[0204] The Laplacian of Gaussian can be used to filter the image, the power value is obtained according to the filtering result, and the sharpness is deduced according to the obtained power value.

[0205] Filtering the image can be expressed as:

[0206]

[0207] x ∈ [1, 5,.....w], y ∈ [1, 5,.....h],

[0208] Among them, I(x, y) represents the image, and I F (x, y) represents the output after filtering the image I(x, y), and w and h respectively represent the width and height of the image I(x, y).

[0209] The power value power is obtained according to the image filtering result, which can be expressed as:

[0210]

[0211] Among them, W F and H F respectively represent the width and height of the filtered image I F (x, y).

[0212] The sharpness SHARPNESS is obtained according to the power value, which can be expressed as:

[0213]

[0214] (3) Calculation of brightness distribution:

[0215]

[0216] Among them, Q represents the brightness distribution, N i is the number of pixels with the value i in the image, and S is the size of the image.

[0217] If each index is greater than the specified threshold, it can be used for the iris recognition process, save the current iris image, and perform iris recognition. If each index is less than the specified threshold, it cannot be used for the iris recognition process, and continue to collect the next frame of image.

[0218] Step S7: Iris recognition means that a high-quality iris image that meets the recognition requirements is collected, and the iris recognition process can be carried out.

[0219] Figure 4 This is an example of the iris image acquisition process in a specific embodiment of the present invention. Refer to Figure 4 , within the shooting field of view 102, the face detection module 103 continuously detects faces. If no face is detected, the detection continues. If a face is detected, a near-infrared iris image of size 1920*1080 can be cropped from the near-infrared image according to the key point coordinates of the face detection. The face image and the iris image are sent to the deep reinforcement learning model for parameter tuning, and different iris images 101 can be obtained. After obtaining the optimal adjustment parameters, the pan-tilt can place the face at the center position of the imaging area according to the pan-tilt adjustment parameters. The acquisition device sets the accurate light intensity, exposure time, gain, focusing parameters, and lighting position according to the iris image adjustment parameters to obtain the iris image 104 after parameter tuning. The iris quality evaluation gives the scores of each index of the current image based on the light intensity, brightness distribution, and clarity of the current image. If the scores of each index are less than the set threshold, the current iris image is discarded and the next acquisition is performed. If the scores of each index are greater than the set threshold, the current iris image is output, and then the iris recognition process is carried out.

[0220] In this embodiment, through the parameter tuning of deep reinforcement learning and deep neural network (pan-tilt parameters, light intensity, exposure parameters, gain, focusing parameters, and lighting position), combined with the supplementary lighting device and the supplementary lighting method, high-quality iris image acquisition can be achieved. This acquisition method has the advantages of fast exposure convergence, accurate focusing and exposure, high-quality iris images, and greatly improved user experience, and can adapt to iris acquisition and recognition at different distances. The method of this embodiment combines the light intensity, exposure parameters, gain, and focusing parameters, as well as a suitable imaging position and lighting scheme, greatly improving the iris quality.

[0221] In summary, the iris image acquisition method, iris recognition method, iris image acquisition system, electronic device, and computer-readable storage medium according to the embodiments of the present invention can optimize the lighting parameters by using a deep convolutional network to optimize the image acquisition parameters, so that lighting can be quickly and appropriately performed during image acquisition, and thus images can be acquired under uniform light and appropriate light intensity. In addition, the problem of slow automatic exposure convergence can be overcome by lighting. By using reinforcement learning to optimize the pan-tilt parameters, images can be quickly acquired at appropriate angles. In this way, iris images convenient for recognition can be quickly acquired in various lighting environments and various acquisition positions. Further, the image acquisition parameters may further include exposure time, gain, and focusing parameters. By using a convolutional network to obtain the exposure time, an image with a more appropriate image light intensity can be obtained. By using a convolutional network to obtain the gain, a clearer image can be quickly obtained. By using a convolutional network to obtain the focusing parameters, rapid focusing can be performed at various acquisition distances to capture clear images. Therefore, this solution can solve the problems existing in existing iris acquisition, such as slow exposure convergence, uneven infrared supplementary lighting, poor iris image quality, and blurred iris images, thereby improving the quality and speed of iris acquisition and enhancing the user experience.

[0222] In the description of this specification, the descriptions referring to terms such as "one embodiment", "a specific embodiment", "some embodiments", "for example", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. The step sequences involved in each embodiment are used to schematically illustrate the implementation of the present invention, and the step sequences are not limited and can be appropriately adjusted as needed.

[0223] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0224] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general purpose computers, special purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0225] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0226] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0227] The specific embodiments described above further elaborate on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. An iris image acquisition method, characterized in that, it includes: Obtain the visible light image of the object to be acquired collected by the visible light image acquisition device and the near-infrared image of the object to be acquired collected by the near-infrared image acquisition device; Perform coordinate mapping on the visible light image and the near-infrared image to obtain a coordinate mapping relationship; Perform face detection on the visible light image, and in the case of detecting a face, obtain a face detection result; According to the face detection result and the coordinate mapping relationship, extract the initial iris image from the near-infrared image; Input the initial iris image into the pre-trained deep convolutional network image acquisition parameter tuning model to obtain the tuned image acquisition parameters; the image acquisition parameters include lighting parameters, and the lighting parameters include infrared lighting parameters and / or visible light lighting parameters; According to the face detection result and using the reinforcement learning pan-tilt parameter tuning model, obtain the tuned pan-tilt parameters; Use the tuned pan-tilt parameters to adjust the pan-tilt positions of the visible light image acquisition device and the near-infrared image acquisition device, and based on the adjusted pan-tilt positions and the tuned image acquisition parameters, acquire the visible light face image and the near-infrared face image of the object to be acquired; Perform face detection on the visible light face image, and according to the face detection result of the visible light face image, extract the iris image for iris recognition from the near-infrared face image.

2. The iris image acquisition method according to claim 1, characterized in that, The lighting parameters include lighting position and lighting intensity; the image acquisition parameters further include exposure time, gain, and focusing parameters; the pan-tilt parameters include moving left, moving right, rotating up, and rotating down.

3. The iris image acquisition method according to claim 1, characterized in that, Performing coordinate mapping on the visible light image and the near-infrared image to obtain a coordinate mapping relationship includes: According to the position information of the visible light image acquisition device and the position information of the near-infrared image acquisition device, perform coordinate mapping on the visible light image and the near-infrared image to obtain a coordinate mapping relationship.

4. The iris image acquisition method according to claim 1, characterized in that, It further includes: In the case of not detecting a face, re-obtain the visible light image of the object to be acquired re-collected by the visible light image acquisition device and the near-infrared image of the object to be acquired re-collected by the near-infrared image acquisition device.

5. The iris image acquisition method according to claim 1, characterized in that, The face detection result of the visible light image includes face detection frame position information and face key point position information; According to the face detection result and the coordinate mapping relationship, extracting the initial iris image from the near-infrared image includes: According to the face key point position information in the face detection result and the coordinate mapping relationship, extract the initial iris image from the near-infrared image; According to the face detection result and using the reinforcement learning pan-tilt parameter tuning model, obtaining the tuned pan-tilt parameters includes: According to the face detection frame position information in the face detection result and using the reinforcement learning pan-tilt parameter tuning model, obtain the tuned pan-tilt parameters.

6. The iris image acquisition method according to claim 5, characterized in that, further comprising: training a policy function to obtain a reinforcement learning pan-tilt parameter tuning model; Training a policy function to obtain a reinforcement learning pan-tilt parameter tuning model includes: Obtaining a training sample set; wherein, the training samples in the training sample set include: the device space position environment corresponding to the visible light image and the true face detection frame of the corresponding visible light image; Inputting the device space position environment corresponding to the visible light image in the training sample into the policy function, using the device space position corresponding to the visible light image as the environmental state in the policy function, using the adjustment method and amplitude of the pan-tilt as possible actions, and using the change direction of the repetition rate of the face detection frame after executing the action relative to the true face detection frame before as the reward value to train the policy function, obtaining the tuning value of the pan-tilt parameters. After the repetition rate of the face detection frame after executing the action exceeds the set threshold, obtain the reinforcement learning pan-tilt parameter tuning model according to the trained policy function.

7. The iris image acquisition method according to claim 6, characterized in that, Using the change direction of the repetition rate of the face detection frame after executing the action relative to the true face detection frame before as the reward value, including: Using the method of representing the detection frame position by two pixel points on the diagonal of the detection frame, calculating the position of the face detection frame after executing the action, the position of the face detection frame before executing the action, and the position of the true face detection frame respectively; Using an object detection algorithm, calculating the first repetition rate of the face detection frame before executing the action and the true face detection frame according to the position of the face detection frame before executing the action and the position of the true face detection frame; Using an object detection algorithm, calculating the second repetition rate of the face detection frame after executing the action and the true face detection frame according to the position of the face detection frame after executing the action and the position of the true face detection frame; Calculating the difference between the first repetition rate and the second repetition rate, and using the sign of the difference as the change direction of the repetition rate of the face detection frame after executing the action relative to the true face detection frame before, so as to obtain the reward value.

8. The iris image acquisition method according to claim 5, characterized in that, further comprising: training an initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter tuning model; Training an initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter tuning model includes: Obtaining a training sample set; wherein, the training samples in the training sample set include: iris images and known image acquisition parameters; Inputting the iris images in the training samples into the initial deep convolutional network to obtain predicted image acquisition parameters; Calculating the loss according to the known image acquisition parameters and the predicted image acquisition parameters in the training samples, and feeding back the calculated loss to the initial deep convolutional network to train the initial deep convolutional network to obtain a pre-trained deep convolutional network image acquisition parameter tuning model.

9. An iris recognition method, characterized in that, comprising: Collect the iris image of the object to be identified by using the iris image acquisition method according to any one of claims 1 to 8; Identify the identity of the object to be identified by using the collected iris image.

10. The iris recognition method according to claim 9, characterized in that, before identifying the identity of the object to be identified by using the collected iris image, the method further includes: performing iris quality evaluation on the collected iris image, and when the iris quality evaluation result meets the iris recognition requirements, performing the step of identifying the identity of the object to be identified by using the collected iris image; when the iris quality evaluation result does not meet the iris recognition requirements, re-performing the step of collecting the iris image of the object to be identified by using the iris image acquisition method according to any one of claims 1 to 8; and / or, performing iris quality evaluation on the initial iris image obtained during the acquisition process of the iris image acquisition method, and when the iris quality evaluation result meets the iris recognition requirements, performing the step of identifying the identity of the object to be identified by using the collected iris image.

11. The iris recognition method according to claim 10, characterized in that, performing iris quality evaluation on the collected iris image includes: calculating the light intensity, clarity, and brightness distribution of the collected iris image; judging whether the iris quality evaluation result of the collected iris image meets the iris recognition requirements by judging whether the light intensity, clarity, and brightness distribution of the collected iris image meet the set requirements; performing iris quality evaluation on the initial iris image obtained during the acquisition process of the iris image acquisition method includes: calculating the light intensity, clarity, and brightness distribution of the initial iris image; judging whether the iris quality evaluation result of the initial iris image meets the iris recognition requirements by judging whether the light intensity, clarity, and brightness distribution of the initial iris image meet the set requirements.

12. The iris recognition method according to claim 11, characterized in that, calculating the light intensity, clarity, and brightness distribution of the collected iris image includes: obtaining the light intensity of the iris image by calculating the mean value of the pixel values of the collected iris image; filtering the collected iris image by using a Laplacian of Gaussian kernel to obtain an image filtering result, and obtaining the power value of the filtered iris image according to the image filtering result, and calculating the clarity of the collected iris image by using the power value; calculating the brightness distribution of the collected iris image by using the pixel values and the image size of the collected iris image; judging whether the iris quality evaluation result of the collected iris image meets the iris recognition requirements by judging whether the light intensity, clarity, and brightness distribution of the collected iris image meet the set requirements includes: judging whether the iris quality evaluation result of the collected iris image meets the iris recognition requirements by judging whether the light intensity of the collected iris image is greater than the set light intensity threshold, whether its clarity is greater than the set clarity threshold, and whether its brightness distribution meets the set brightness distribution requirements; calculating the light intensity, clarity, and brightness distribution of the initial iris image includes: The light intensity of the iris image is obtained by calculating the mean value of the pixel values of the initial iris image; The initial iris image is filtered using a Laplacian of Gaussian kernel to obtain an image filtering result, and the power value of the filtered iris image is obtained according to the image filtering result. The sharpness of the initial iris image is calculated using the power value; The brightness distribution of the initial iris image is calculated using the pixel values and the image size of the initial iris image; By determining whether the light intensity, sharpness, and brightness distribution of the initial iris image meet the set requirements, it is determined whether the iris quality evaluation result of the initial iris image meets the iris recognition requirements, including: By determining whether the light intensity of the initial iris image is greater than the set light intensity threshold, whether its sharpness is greater than the set sharpness threshold, and whether its brightness distribution meets the set brightness distribution requirements, it is determined whether the iris quality evaluation result of the initial iris image meets the iris recognition requirements.

13. An iris image acquisition system, characterized in that, it includes: A visible light image acquisition device for acquiring a visible light image of an object to be acquired; A near-infrared image acquisition device for acquiring a near-infrared image of an object to be acquired; A visible light fill light for performing visible light fill light on the object to be acquired according to visible light lighting parameters when acquiring the visible light image of the object to be acquired; An infrared fill light for performing infrared fill light on the object to be acquired according to infrared lighting parameters when acquiring the near-infrared image of the object to be acquired; A pan-tilt for supporting the visible light camera of the visible light image acquisition device, the near-infrared camera of the near-infrared image acquisition device, the visible light fill light, and the infrared fill light based on pan-tilt parameters; An iris acquisition device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 8 to acquire the visible light image and the near-infrared image, and optimize the image acquisition parameters and the pan-tilt parameters; wherein, the lighting parameters in the image acquisition parameters include the visible light lighting parameters and the infrared lighting parameters.

14. The iris image acquisition system according to claim 13, characterized in that, The infrared fill light includes multiple groups of infrared lights arranged around the near-infrared camera in the near-infrared image acquisition device, and the infrared lighting parameters include information about the groups of infrared lights; and / or, The visible light fill light includes multiple groups of visible lights arranged around the visible light camera in the visible light image acquisition device, and the visible light lighting parameters include information about the groups of visible lights.

15. The iris image acquisition system according to claim 13, characterized in that, The visible light image acquisition device and the near-infrared image acquisition device are VCM cameras, and the image acquisition parameters further include a focusing parameter, and the focusing parameter is the VCM drive current corresponding to the VCM distance of the VCM camera.

16. An electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the steps of the method according to any one of claims 1 to 12.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 12.

Images