Methods and systems for facial recognition in outdoor video.

Abstract

This invention provides a method and system for recognizing faces in outdoor images. The method for recognizing faces in outdoor images includes: an RGB image generation step, which captures an RGB image of the face of a person to be identified outdoors; a 3D image generation step, which captures a real 3D image of the face of the person to be identified outdoors; a virtual overexposed image generation step, which generates one or more virtual overexposed images using an algorithm; a face image feature generation step, which calculates face image features based on the RGB image, the real 3D image, and one or more virtual overexposed images; and a face recognition step, which uses one or more trained artificial intelligence / neural network (AI / NN) models to perform face recognition on the person to be identified outdoors based on the face image features.

Description

Technical Field

[0001] This application relates to the technical field of identification methods and identification systems, and in particular to a method and system for identifying faces in outdoor images. Background Technology

[0002] With the rapid development of big data and machine learning technologies, existing facial recognition technologies have matured and are widely applied in various fields. For example, machine learning has achieved remarkable results in facial recognition technology. One reason for this is that facial recognition is a biometric identification technology based on physical appearance, and its non-invasive nature makes it popular with users. In existing technologies, facial recognition is further integrated with research in other fields, leading to a variety of applications. These include surveillance, security (e.g., system login, account security), and even entertainment (e.g., human-computer interaction, virtual reality).

[0003] In current technologies, mainstream machine learning architectures include Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Deep Neural Networks (DNNs). Most current research on face recognition uses CNNs as its foundation. Generally, the effectiveness of machine learning depends on the content and quantity of training data. Furthermore, in some extreme and complex situations, such as dim lighting or when the user is wearing makeup or disguise, the recognition rate may decrease.

[0004] Therefore, how to provide a method and system for recognizing faces in outdoor images that can solve the above problems is an important issue that the industry needs to consider. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and system for recognizing faces in outdoor images in order to improve the accuracy of face recognition in outdoor images.

[0006] According to one aspect of this application, a method for recognizing a face in outdoor images is provided, comprising:

[0007] The RGB image generation step is used to capture RGB images of the faces of people to be identified outdoors.

[0008] The three-dimensional image generation step involves capturing a real three-dimensional image of the face of the person to be identified outdoors.

[0009] The virtual overexposed image generation step uses an algorithm to generate one or more virtual overexposed images;

[0010] The facial image feature generation step involves calculating facial image features based on the RGB image, the real 3D image, and one or more virtual overexposed images; and

[0011] The facial recognition step utilizes one or more trained artificial intelligence / neural network (AI / NN) models to perform facial recognition actions on the person to be identified outdoors based on the facial image features.

[0012] In one embodiment, the virtual overexposed image generation step modifies the algorithm with a Gaussian Gain Model to obtain the one or more trained artificial intelligence / neural network (AI / NN) models.

[0013] The virtual overexposed image is a simulated image of sunlight shining on a face, obtained through a modified algorithm.

[0014] In one embodiment, the virtual overexposed image generation step includes:

[0015] Establish a Gaussian Gain Model;

[0016] Adjust the radius and intensity to correct the Gaussian gain model, and calculate the Gaussian gain.

[0017] The Gaussian gain is applied to the face image to simulate the image of sunlight shining on the face, thus obtaining the virtual overexposed image.

[0018] In one embodiment, the virtual overexposed image generation step further includes:

[0019] The step of randomly generating the Gaussian beam position involves first rotating the face image to a vertical direction and obtaining one or more simulated Gaussian beam positions on the face image before establishing the Gaussian gain model.

[0020] In one embodiment, the virtual overexposed image generation step further includes:

[0021] The face and facial feature position detection step first detects the position of the face and facial features sequentially before the step of randomly generating the Gaussian beam position.

[0022] In one embodiment, the virtual overexposed image generation step further includes:

[0023] The initial face image input step involves acquiring one or more initial face images before the position detection step of the face and the facial features.

[0024] In one embodiment, the virtual overexposed image generation step further includes:

[0025] The comprehensive stability test steps involve repeatedly applying the Gaussian gain to the face image to simulate the image of outdoor sunlight shining on the face from different angles, thereby obtaining the virtual overexposed image.

[0026] According to another aspect of this application, a method for recognizing faces in outdoor images is provided, comprising:

[0027] The virtual overexposed image generation step utilizes an algorithm to generate one or more virtual overexposed images; and

[0028] The steps for recognizing faces in outdoor images utilize one or more trained artificial intelligence / neural network (AI / NN) models to identify faces in one or more outdoor images.

[0029] The one or more trained artificial intelligence / neural network (AI / NN) models pre-store or continuously store the one or more virtual overexposed images, and the one or more virtual overexposed images serve as a training set. After being processed along with the images of faces in the one or more outdoor images, one or more facial image features are generated to pre-train or continuously train the one or more trained artificial intelligence / neural network (AI / NN) models.

[0030] According to another aspect of this application, a face recognition system is provided, suitable for electronic devices, and performs the face recognition method in outdoor images as described in any of the foregoing embodiments, the face recognition system comprising:

[0031] An artificial intelligence (AI) computing unit is used to perform the facial image feature generation step;

[0032] An RGB image capturing unit is used to perform the RGB image generation step and is coupled to the artificial intelligence (AI) computing unit;

[0033] A three-dimensional image capturing unit is used to perform the three-dimensional image generation step and is coupled to the artificial intelligence (AI) computing unit;

[0034] A virtual overexposed image generation unit, used to perform the virtual overexposed image generation step, and coupled to the artificial intelligence (AI) computing unit; and

[0035] The face recognition unit has one or more trained artificial intelligence / neural network (AI / NN) models built in to perform the face recognition step or the face recognition step in the outdoor image.

[0036] In one embodiment, the three-dimensional image capturing unit includes a depth sensing camera. Attached Figure Description

[0037] To make the above and other objects, features, advantages and embodiments of the present invention more readily understood, the accompanying drawings are described below:

[0038] Figure 1 This is a schematic diagram of a face recognition system according to an embodiment of the present invention.

[0039] Figure 2 One or more steps for generating the virtual overexposed image in the illustrated embodiment.

[0040] Figure 3 One or more steps for generating the virtual overexposed image in the illustrated embodiment.

[0041] Figure 4 One or more steps for generating the virtual overexposed image in the illustrated embodiment.

[0042] Figure 5 One or more steps for generating the virtual overexposed image in the illustrated embodiment.

[0043] Figure 6 One or more steps for generating the virtual overexposed image in the illustrated embodiment.

[0044] As is customary practice, the various features and elements in the figures are not drawn to scale, but rather in a manner designed to best represent the specific features and elements related to the present invention. Furthermore, similar elements and components are referred to by the same or similar element symbols across different figures.

[0045] Brief explanation of component symbols:

[0046] 1: Face recognition system 10: First image acquisition unit

[0047] 20: Second image acquisition unit; 30: Artificial intelligence computing unit

[0048] 40: Virtual overexposed image generation unit; 50: Image recognition unit

[0049] 300: Outdoor imaging; 310, 310a, 310b: Facial imaging.

[0050] 320, 330, 340: Facial features; 412, 412a: Virtual axes

[0051] 600: Gaussian gain; 600a, 700a, 700b, 700c: Feature region

[0052] 1310a: Simulated position of Gaussian beam 2310a: Simulated position of Gaussian beam Detailed Implementation

[0053] To gain a better understanding of the purpose, shape, structural features, and effects of the present invention, embodiments are described in detail below with reference to the accompanying drawings.

[0054] The following disclosure provides different embodiments or examples to establish different features of the provided subject matter. The specific examples of the components and arrangements described below are for the purpose of simplifying this disclosure and are not intended to constitute limitation; the size and shape of the elements are not limited by the scope or values ​​disclosed, but may depend on the manufacturing conditions of the elements or the desired characteristics. For example, the technical features of the invention are described using cross-sectional views, which are schematic diagrams of idealized embodiments. Therefore, differences in the shapes illustrated due to manufacturing processes and tolerances are foreseeable and should not be limiting.

[0055] Furthermore, spatial relative terms, such as “below,” “under,” “lower than,” “above,” and “higher than,” are used to easily describe the relationship between the elements or features shown in the accompanying drawings. In addition, spatial relative terms include not only the directions depicted in the drawings but also the different orientations of the elements during use or operation.

[0056] First, it should be noted that, in view of the problems of the prior art, the invention described in the embodiments disclosed in this specification generates virtual overexposed images through algorithms, trains a deep learning model, and improves its accuracy in recognizing faces in outdoor images.

[0057] Therefore, one embodiment of the present invention provides a method for recognizing a face in outdoor images. This method includes: an RGB image generation step, a three-dimensional image generation step, a virtual overexposed image generation step, a face image feature generation step, and a face recognition step. The RGB image generation step is used to capture an RGB image of the face of a person to be identified outdoors. The three-dimensional image generation step is used to capture a real three-dimensional image of the face of the person to be identified outdoors. The virtual overexposed image generation step generates one or more virtual overexposed images using an algorithm. The face image feature generation step calculates a face image feature based on the RGB image, the real three-dimensional image, and the one or more virtual overexposed images. The face recognition step uses one or more trained artificial intelligence / neural network (AI / NN) models to perform face recognition on the person to be identified outdoors based on the face image features.

[0058] In addition, another embodiment of the present invention provides a method for recognizing faces in outdoor images. This method includes a virtual overexposed image generation step and a face recognition step in the outdoor images. The virtual overexposed image generation step uses an algorithm to generate one or more virtual overexposed images. The face recognition step in the outdoor images uses one or more trained artificial intelligence / neural network (AI / NN) models to recognize faces in one or more outdoor images. It should be specifically noted that the one or more trained AI / neural network (AI / NN) models pre-store or continuously store the one or more virtual overexposed images, and these virtual overexposed images serve as a training set. Together with the faces in the one or more outdoor images, they are used to calculate one or more face image features to pre-train or continuously train the one or more trained AI / neural network (AI / NN) models.

[0059] Furthermore, another embodiment of the present invention provides a face recognition system. This system is applicable to electronic devices and executes the face recognition method in outdoor images described in any embodiment of the present invention. The face recognition system includes an artificial intelligence (AI) computing unit, an RGB image capturing unit, a three-dimensional image capturing unit, a virtual overexposed image generation unit, and a face recognition unit. The AI ​​computing unit is used to perform the face image feature generation step. The RGB image capturing unit is used to perform the RGB image generation step and is coupled to the AI ​​computing unit. The three-dimensional image capturing unit is used to perform the three-dimensional image generation step and is coupled to the AI ​​computing unit. The virtual overexposed image generation unit is used to perform the virtual overexposed image generation step and is coupled to the AI ​​computing unit. In addition, the face recognition unit has one or more trained artificial intelligence / neural network (AI / NN) models built in to perform the face recognition step or the face recognition step in the outdoor image.

[0060] Please refer to Figure 1 , Figure 1 A schematic diagram of a face recognition system according to an embodiment of the present invention is shown.

[0061] like Figure 1 As shown, in an embodiment of the present invention, the face recognition system 1 is applicable to an electronic device and performs the face recognition method in outdoor images described in any embodiment of the present invention (as detailed below). The face recognition system 1 includes an artificial intelligence (AI) computing unit 30, an RGB image capturing unit 20, a three-dimensional image capturing unit 10, a virtual overexposure image generation unit 40, and a face recognition unit 50.

[0062] In an embodiment of the present invention, the artificial intelligence (AI) computing unit 30 is used to perform the facial image feature generation step.

[0063] In an embodiment of the present invention, the RGB image capturing unit 20 is used to perform the RGB image generation step and is coupled to the artificial intelligence (AI) computing unit 30.

[0064] In an embodiment of the present invention, the three-dimensional image capturing unit 10 is used to perform the three-dimensional image generation step and is coupled to the artificial intelligence (AI) computing unit 30.

[0065] In an embodiment of the present invention, the virtual overexposed image generation unit 40 is used to perform a virtual overexposed image generation step and is coupled to the artificial intelligence (AI) computing unit 30.

[0066] In addition, in embodiments of the present invention, the face recognition unit 50 is equipped with one or more trained artificial intelligence / neural network (AI / NN) models to perform a face recognition step or a face recognition step in an outdoor image.

[0067] Next, please refer to the following in order. Figures 2-6 , Figure 2 One or more steps for generating a virtual overexposed image in the illustrated embodiment are shown. Figure 3 One or more steps for generating a virtual overexposed image in the illustrated embodiment are shown. Figure 4 One or more steps for generating a virtual overexposed image in the illustrated embodiment are shown. Figure 5 One or more steps for generating a virtual overexposed image in the illustrated embodiment are shown. Figure 6 One or more steps for generating a virtual overexposed image in the illustrated embodiment are shown.

[0068] In an embodiment of the present invention, the method for recognizing a face in an outdoor image 300 includes: an RGB image generation step, a three-dimensional image generation step, a virtual overexposed image generation step, a face image feature generation step, and a face recognition step. The RGB image generation step is used to capture an RGB image of the face of the person to be identified outdoors. The three-dimensional image generation step is used to capture a real three-dimensional image of the face of the person to be identified outdoors. The virtual overexposed image generation step generates one or more virtual overexposed images using an algorithm. The face image feature generation step calculates face image features based on the RGB image, the real three-dimensional image, and the one or more virtual overexposed images. The face recognition step uses one or more trained artificial intelligence / neural network (AI / NN) models to perform face recognition on the person to be identified outdoors based on the face image features.

[0069] In an embodiment of the present invention, the virtual overexposed image generation step includes: an initial face image input step, in which one or more initial face images are first obtained for subsequent face and facial feature position detection steps.

[0070] Figure 2 As shown, in an embodiment of the present invention, the virtual overexposed image generation step includes: a face and facial feature position detection step, which sequentially detects the positions of face image 310 and facial features 320, 330, and 340 in the outdoor image 300. Next, a step of randomly generating a Gaussian beam position is performed.

[0071] Figures 3-4 As shown, in an embodiment of the present invention, the virtual overexposed image generation step includes: randomly generating Gaussian beam positions, rotating the virtual axis 412 of the face image 310 to the vertical direction (i.e., the virtual axis 412 of the face image 310a is located in the vertical direction), and obtaining one or more simulated Gaussian beam positions 1310a (i.e., the upper half of the face position) and 2310a (i.e., the lower half of the face position) on the face image 310a. Then, the step of establishing a Gaussian gain model continues.

[0072] Figure 5-6 As shown, in an embodiment of the present invention, the virtual overexposed image generation step includes: establishing a Gaussian Gain Model; adjusting the radius and intensity to correct the Gaussian Gain Model, and calculating the Gaussian Gain; applying the Gaussian Gain 600 to the face image 310a to simulate the image of sunlight shining on the face, thereby obtaining the virtual overexposed image, i.e., face image 310b. For example, feature regions 600a, 700a, 700b, and 700c are the areas of the face illuminated by sunlight.

[0073] In other words, in the embodiments of the present invention, the virtual overexposed image generation step corrects the algorithm with a Gaussian Gain Model to obtain one or more trained artificial intelligence / neural network (AI / NN) models, wherein the virtual overexposed image is an image simulating sunlight shining on a face obtained through the corrected algorithm.

[0074] Figure 2 and Figure 6 As shown, in an embodiment of the present invention, the virtual overexposed image generation step further includes: a comprehensive stability test step, in which the Gaussian gain is repeatedly applied to the face image to simulate the image of outdoor sunlight shining on the face at different angles, thereby obtaining the virtual overexposed image.

[0075] Finally, it should be reiterated that the embodiments of the present invention provide a face recognition system applicable to electronic devices, and perform the face recognition method in outdoor images described in any of the foregoing embodiments. The face recognition system includes: an artificial intelligence (AI) computing unit for performing the face image feature generation step; an RGB image capturing unit for performing the RGB image generation step and coupled to the AI ​​computing unit; a three-dimensional image capturing unit for performing the three-dimensional image generation step and coupled to the AI ​​computing unit; a virtual overexposure image generation unit for performing the virtual overexposure image generation step and coupled to the AI ​​computing unit; and a face recognition unit, which incorporates one or more trained artificial intelligence / neural network (AI / NN) models to perform the face recognition step or the face recognition step in the outdoor image.

[0076] In an embodiment of the present invention, the three-dimensional image capturing unit includes a depth sensing camera.

[0077] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for recognizing a face in an outdoor image, characterized in that, Comprising: an RGB image generating step for capturing an RGB image of a face of an outdoor person to be identified; a three-dimensional image generating step for capturing a real three-dimensional image of the face of the outdoor person to be identified; a virtual overexposed image generating step for generating one or more virtual overexposed images by using an algorithm; a face image feature generating step for calculating a face image feature according to the RGB image, the real three-dimensional image, and the one or more virtual overexposed images; and a face recognition step for performing a face recognition action on the outdoor person to be identified according to the face image feature by using one or more trained artificial intelligence / neural network (AI / NN) models, wherein the virtual overexposed image generating step corrects the algorithm by a Gaussian Gain Model to obtain the one or more trained artificial intelligence / neural network (AI / NN) models; wherein the virtual overexposed image is an image simulated by the corrected algorithm that the sunlight is irradiated on the face. 2.The method of claim 1, wherein, The virtual overexposed image generating step comprises: establishing a Gaussian Gain Model; adjusting a radius and an intensity to correct the Gaussian Gain Model and calculating a Gaussian Gain; applying the Gaussian Gain to a face image to simulate an image that the sunlight is irradiated on the face to obtain the virtual overexposed image. 3.The method of claim 2, wherein, The virtual overexposed image generating step further comprises: a step of randomly generating a Gaussian beam position, before establishing the Gaussian Gain Model, the face image is first converted to a vertical direction and one or more Gaussian beam simulation positions are obtained on the face image. 4.The method of claim 3, wherein, The virtual overexposed image generating step further comprises: a step of detecting positions of a face and a facial feature, before the step of randomly generating a Gaussian beam position, positions of the face and the facial feature are sequentially detected. 5.The method of claim 4, wherein, The virtual overexposed image generating step further comprises: an initial face image input step, before the step of detecting positions of the face and the facial feature, one or more initial face images are obtained. 6.The method of claim 5, wherein, The virtual overexposed image generating step further comprises: a comprehensive stability test step, the Gaussian Gain is repeatedly applied to the face image to simulate an image that the sunlight at different angles is irradiated on the face to obtain the virtual overexposed image.

7. A method for recognizing a face in an outdoor image, characterized by, Comprising: a virtual overexposed image generating step for generating one or more virtual overexposed images by using an algorithm; and ​ The face recognition step in the outdoor image uses one or more trained artificial intelligence / neural network (AI / NN) models to recognize the face image in one or more outdoor images; The one or more trained artificial intelligence / neural network (AI / NN) models pre-store or continuously store the one or more virtual overexposed images, which are used as a training set to calculate one or more face image features along with the face image in the one or more outdoor images to pre-train or continuously train the one or more trained artificial intelligence / neural network (AI / NN) models. The virtual overexposed image generation step uses a Gaussian Gain Model to modify the algorithm to obtain the one or more trained artificial intelligence / neural network (AI / NN) models. The virtual overexposed image is an image simulated by the modified algorithm, in which sunlight is simulated to shine on the face.

8. A face recognition system suitable for an electronic device and performing the face recognition method in the outdoor image according to any one of claims 1 to 7, the face recognition system comprising: An artificial intelligence (AI) computing unit configured to perform the face image feature generation step; An RGB image capturing unit configured to perform the RGB image generation step and coupled to the artificial intelligence (AI) computing unit; A three-dimensional image capturing unit configured to perform the three-dimensional image generation step and coupled to the artificial intelligence (AI) computing unit; A virtual overexposed image generation unit configured to perform the virtual overexposed image generation step and coupled to the artificial intelligence (AI) computing unit; and A face recognition unit configured to perform the face recognition step or the face recognition step in the outdoor image, and built-in the one or more trained artificial intelligence / neural network (AI / NN) models.

9. The face recognition system of claim 8, wherein, The three-dimensional image capturing unit comprises a depth sensing camera.

Images