# US 2019/0377930 A1 · Method and System for Face Recognition Via Deep Learning

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0377930 A1
- Document ID: US20190377930A1
- Status: Published application
- Kind: A1
- Application number: 16/177,952
- Filed: 2018-11-01
- Published: 2019-12-12
- Earliest claimed: 2018-06-11
- Assignees: Zkteco USA LLC
- Applicants: Zkteco USA, LLC
- Inventors: Shukai CHEN; Feiyang TONG
- Web page: https://patentgrep.com/patent/US20190377930A1

## CPC

- G06N 3/08: Learning methods
- G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques}
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/09: Supervised learning
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/82: using neural networks
- G06V 40/1347: {Preprocessing; Feature extraction}
- G06V 40/1365: {Matching; Classification}
- G06V 40/172: {Classification, e.g. identification}
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

Embodiments of the present application disclose a deep learning-based face identification method and a related product. The method is applied to an electronic apparatus. The method includes: acquiring an aligned face image, and scaling the face image at a preset ratio to obtain a target image; extracting a pixel matrix of the target image, inputting the pixel matrix to the neural network model, and executing multilayer computing to obtain a computing result; and calculating a cosine similarity degree between the computing result and a facial template vector, wherein comparison is successful if the cosine similarity degree is greater than a preset threshold. The embodiments of the present application help improve the identification success rate and the identification speed.

## Claims

### Claim at position 1

1-6. (canceled)

### Claim 7

A deep learning-based face identification method comprising:
acquiring an aligned face image;
scaling the face image at a preset ratio to obtain a target image;
extracting a pixel matrix of the target image;
inputting the pixel matrix to a neural network model; and executing multilayer computing to obtain a computing result; and
calculating a cosine similarity degree between the computing result and a facial template vector, wherein comparison is successful if the cosine similarity degree is greater than a preset threshold.

### Claim 8

An electronic apparatus for deep learning-based face identification, the electronic apparatus comprising:
an acquisition unit configured to:
acquire an aligned face image, and
scale the face image at a preset ratio to obtain a target image;
an extraction unit configured to:
extract a pixel matrix of the target image, and
input the pixel matrix to a preset trained neural network model, and execute multilayer forward computing to obtain a computing result; and
a calculation unit configured to calculate a cosine similarity degree between the computing result and a facial template vector, wherein comparison is successful if a matching value is greater than the cosine similarity degree.

### Claim 9

(canceled)

### Claim 10

A non-transitory machine-readable medium having information recorded thereon for performing deep learning-based face identification, wherein the information, when read by the machine, causes the machine to perform the following:
acquiring an aligned face image;
scaling the face image at a preset ratio to obtain a target image;
extracting a pixel matrix of the target image;
inputting the pixel matrix to a neural network model; and executing multilayer computing to obtain a computing result; and
calculating a cosine similarity degree between the computing result and a facial template vector, wherein comparison is successful if the cosine similarity degree is greater than a preset threshold.

### Claim 11

An apparatus for face identification comprising:
a memory; and
a processor configured to implement a neural network model including:
a first convolutional layer,
a second convolutional layer,
a third convolutional layer,
a first bottleneck network,
a second bottleneck network,
a third bottleneck network,
a fourth bottleneck network,
a first pooling layer,
a second pooling layer,
a third pooling layer, and
a fourth pooling layer, wherein
the first convolutional layer, the first bottleneck network, the first pooling layer, the second bottleneck network, the second convolutional layer, the second pooling layer, the third bottleneck network, the third convolutional layer, the third pooling layer, the fourth bottleneck network and the fourth pooling layer are connected in series sequentially, and wherein
the first convolutional layer, the first bottleneck network, the first pooling layer, the second bottleneck network, the second convolutional layer, the second pooling layer, the third bottleneck network, the third convolutional layer, the third pooling layer, the fourth bottleneck network and the fourth pooling layer are densely connected according to a connection policy.

### Claim 12 (depends on 11)

The apparatus of claim 11, wherein the neural network model further comprises six global pooling layers, three first global poolings in the six global poolings are used for pooling a feature map output by the first convolutional layer to obtain three first pooling results, and inputting the three first pooling results to the second bottleneck network, the third bottleneck network and the fourth bottleneck network respectively, so as to perform feature fusion on the three first pooling results and other input data of the second bottleneck network, the third bottleneck network and the fourth bottleneck network; two second global poolings in the six global poolings are used for pooling a feature map output by the first bottleneck network to obtain two second pooling results and inputting the two second pooling results to the third bottleneck network and the fourth bottleneck network respectively, so as to perform feature fusion on the two second pooling results and other input data of the third bottleneck network and the fourth bottleneck network; one third global pooling in the six global poolings is used for pooling a feature map output by the second bottleneck network to obtain one third pooling result and inputting the third pooling result to the fourth bottleneck network, so as to perform feature fusion on the third pooling result and other input data of the fourth bottleneck network.

### Claim 13 (depends on 11)

The apparatus of claim 11, wherein the first bottleneck network, the second bottleneck network, the third bottleneck network and the fourth bottleneck network comprise 3, 4, 6, and 3 sub-bottleneck networks (sub-bottlenecks) respectively, and the respective sub-bottlenecks in the first bottleneck network, the second bottleneck network, the third bottleneck network, and the fourth bottleneck network are connected in series sequentially.

### Claim 14 (depends on 13)

The apparatus of claim 13, wherein any sub-bottleneck network structure comprises a first group convolutional layer (i.e., first group cony), a second group convolutional layer (i.e., second group cony), a first depth convolutional layer (first depth-wise cony), a superposition layer, and a channel switching network layer;
the first group cony, the first depth-wise cony, the second group cony, the add, and the shuffle are connected in series sequentially;
the first group cony has a convolution kernel size of [1*1*256] and a step size of 1, the second group cony has a convolution kernel size of [3*3*256] and a step size of 1, and the first depth-wise cony has a convolution kernel size of [1*1*64] and a step size of 1; and
input data and output data of any of the sub-bottlenecks are superposed at the superposition layer, and then the superposed data is input to the channel switching network layer.

### Claim 15 (depends on 11)

The apparatus of claim 11, wherein the first convolutional layer has a convolution kernel size of [3*3*64] and a step size of 2, the second convolutional layer has a convolution kernel size of [1*1*64] and a step size of 1, and the third convolutional layer has a convolution kernel size of [1*1*96] and a step size of 1.

### Claim 16 (depends on 11)

The apparatus of claim 11, wherein the first pooling layer, the second pooling layer and the third pooling layer are all average pooling, the fourth pooling layer is global pooling, and the first pooling layer, the second pooling layer and the third pooling layer each have a convolution kernel size of [3*3] and a step size of 2.