Face Search SDK(1:N Face Recognition) - Server

We implemented face search SDK to perform 1:N face recognition`and 1:N face matching ondocker container.

This SDK demonstrates 1:N face recognition, face search SDK derived from KBY-AI's face recognition server SDK by implementing the functionalities to register face and search face from database(PostgreSQL). This SDK offers APIs to enroll face, to search face, to see database, to clear database. And every API can be customized by updating app.py file accordingly.

Features

• Face Detection

• 1:N Face Recognition

• Face Enrollment(Face Register)

• Face Search

• Pose Estimation

• 68 points Face Landmark Detection

• Face Quality Calculation

• Face Occlusion Detection

• Eye Closure Detection

• Mouth Opening Check

License

We offer lifetime licenses(perpetual) based on machine ID for Servers (Windows, Linux).

To request a license, please contact us:

Email: [email protected]

Share on WhatsAppWhatsApp.com

Justin H.Telegram

Join the KBY-AI Discord Server!Discord

System Requirements

1. Windows

• CPU: 2 cores or more (Recommended: 2 cores)

• RAM: 4 GB or more (Recommended: 8 GB)

• HDD: 4 GB or more (Recommended: 8 GB)

• OS: Windows 7 or later

• Architecture: x64

• Dependency: OpenVINO™ Runtime (Version: 2022.3), ncnn Runtime(20220721), Vulkan SDK Runtime(1.3.250)

2. Linux

• CPU: 2 cores or more (Recommended: 2 cores)

• RAM: 4 GB or more (Recommended: 8 GB)

• HDD: 4 GB or more (Recommended: 8 GB)

• OS: Ubuntu 20.04 or later

• Architecture: x64

• Dependency: OpenVINO™ Runtime (Version: 2022.3)

Postman Endpoints

To test the API, you can use Postman. Here are the endpoints for testing:

1. http://<your-base-url>/register This API enrolls face data from image base64 format and save it to database(PostgreSQL)

2. http://<your-base-url>/search This API seeks face similar to input face among database and returns enrolled image ID and similarity score.

3. http://<your-base-url>/user_list This API shows all data enrolled on database(PostgreSQL).

4. http://<your-base-url>/remove_all This API removes all data from database.

Initializing SDK

1. Step one

• First, obtain the machine code for activation and request a license based on the machine code.

machineCode = getMachineCode()
print("machineCode: ", machineCode.decode('utf-8'))

1. Step Two

• Next, activate the SDK using the received license.

setActivation(license.encode('utf-8'))

• If activation is successful, the return value will be SDK_SUCCESS. Otherwise, an error value will be returned.

1. Step Three

• After activation, call the initialization function of the SDK.

initSDK("data".encode('utf-8'))

• The first parameter is the path to the model.

• If initialization is successful, the return value will be SDK_SUCCESS. Otherwise, an error value will be returned.

Enums and Structure

1. SDK_ERROR

This enumeration represents the return value of the initSDK and setActivation functions.

2. FaceBox

This class represents the output of the face detection function that contains the detected face rectangle, liveness score, and facial angles such as yaw, roll, and pitch.

Main Functions

1. getMachineCode

First, obtain the machine code for activation and request a license based on the machine code.

machineCode = getMachineCode()
print("machineCode: ", machineCode.decode('utf-8'))

2. setActivation

Next, activate the SDK using the received license.

setActivation(license.encode('utf-8'))

If activation is successful, the return value will beSDK_SUCCESS. Otherwise, an error value will be returned.

3. initSDK

After activation, call the initialization function of the SDK.

initSDK("data".encode('utf-8'))

The first parameter is the path to the model.

If initialization is successful, the return value will be SDK_SUCCESS. Otherwise, an error value will be returned.

4. faceDetection

The Face SDK provides a single API for detecting faces, determining face orientation (yaw, roll, pitch), assessing face quality, detecting facial occlusion, eye closure, mouth opening, and identifying facial landmarks.

The function can be used as follows:

# Python code example
faceBoxes = (FaceBox * maxFaceCount)()
faceCount = faceDetection(image_np, image_np.shape[1], image_np.shape[0], faceBoxes, maxFaceCount)

// C++ code example
FaceBox* faceBoxes = (FaceBox*)malloc(sizeof(FaceBox) * maxFaceCount);
memset(faceBoxes, 0, sizeof(FaceBox) * maxFaceCount);

ret = faceDetection(image.data, image.cols, image.rows, faceBoxes, maxFaceCount);

This function requires 5 parameters.

• The first parameter: the byte array of the RGB image buffer.

• The second parameter: the width of the image.

• The third parameter: the height of the image.

• The fourth parameter: the FaceBox array allocated with maxFaceCount for storing the detected faces.

• The fifth parameter: the count allocated for the maximum FaceBox objects.

The function returns the count of the detected face.

5. templateExtraction

The SDK provides a function that enables the generation of templates from RGB data. These templates can be used for face verification between two faces.

The function can be used as follows:

# Python code example
faceBoxes = (FaceBox * maxFaceCount)()

...

templateExtraction(image_np1, image_np1.shape[1], image_np1.shape[0], faceBoxes[0])

// C++ code example

FaceBox* faceBoxes = (FaceBox*)malloc(sizeof(FaceBox) * maxFaceCount);
memset(faceBoxes, 0, sizeof(FaceBox) * maxFaceCount);

...

templateExtraction(image.data, image.cols, image.rows, faceBoxes[0]);

This function requires 4 parameters.

• The first parameter: the byte array of the RGB image buffer.

• The second parameter: the width of the image.

• The third parameter: the height of the image.

• The fourth parameter: the FaceBox object obtained from the faceDetection function.

If the template extraction is successful, the function will return 0. Otherwise, it will return -1.

6. similarityCalculation

The similarityCalculation function takes a byte array of two templates as a parameter.

# Python code example

similarity = similarityCalculation(faceBoxes1[0].templates, faceBoxes2[0].templates)

// C++ code example

FaceBox* faceBoxes = (FaceBox*)malloc(sizeof(FaceBox) * maxFaceCount);
memset(faceBoxes, 0, sizeof(FaceBox) * maxFaceCount);

...

float similarity = similarityCalculation(faceBoxes[0].templates, faceBoxes[1].templates);

It returns the similarity value between the two templates, which can be used to determine the level of likeness between the two individuals.

Default Thresholds

verifyThreshold = 0.67 

• If the similarity value between the two templates is high, it indicates that the faces are the same.