Face Feature Extraction Service

---

Introduction

The Face Feature Extraction Service (Face Server) is a standalone backend that runs on a CentOS server. It exposes an HTTP API: you upload a photo containing a face, and the service returns a cropped face thumbnail and the corresponding feature vector (feature value).

This is useful when you manage user photos on a PC or cloud backend and need to prepare data for VF-series DejaOS face devices (Linux-based; not Android face terminals). Typical workflows:

• Upload a large high-resolution photo (e.g. several MB); receive a small cropped face image (~20 KB) for faster download to devices.
• Convert photos into 1024-byte feature vectors and push only the vectors to devices — less bandwidth and no need to store portrait photos on the device.

For concepts such as feature vectors and registration, see Facial Recognition Development Overview.

---

Photo Requirements

Uploaded registration photos should meet the following requirements (same as on-device photo registration):

Item	Requirement
Face pixels	Min 80, max 512
File size	50–300 KB
Image dimensions	Width 512–800, height 640–1024
Format	JPG
Content	Everyday / lifestyle photo; no beauty filters; no heavy makeup; no occlusion such as sunglasses, scarves, or masks

---

Deployment Overview

Item	Requirement
OS	CentOS 7.x
Install path	/root/face (fixed)
HTTP port	9001 (must be open)
Compatible devices	VF-series face products (DejaOS / Linux), excluding Android models

After installation, start the crop/feature service with run.sh under /root/face/bin/. On first launch you may need a License file (server ID is shown in the console; contact your administrator). See the package README for environment variables (LD_LIBRARY_PATH, FACEIDSDKPATH) and startup details.

Download: FaceServer_dejaos.zip

---

HTTP API

Endpoint: POST http://<server-ip>:9001/dxdop/webadmin/face/clipper

Request: multipart/form-data with field file — the image to process.

Response (success example):

{
  "code": 1,
  "msg": "裁剪成功",
  "data": {
    "faceB64": "/9j/4AAQS....",
    "faceFeature": "6gcF8uFUNjcwC.....AAA=="
  }
}

Field	Description
faceB64	Base64-encoded cropped face image (thumbnail)
faceFeature	Base64-encoded feature vector for device registration / 1:N matching

Use the cropped image or feature value on the device according to your application (see Face Registration and Recognition).

---

Test Client (Windows)

The installation package includes a Windows desktop test tool under bin/test/ for batch processing folders of photos without writing your own HTTP client.

1. Start the Face Server on CentOS (see package README).
2. Configure data/photo_tool/config with your server IP:

   {
     "batchUrl": "http://<IP>:9001/dxdop/webadmin/face/clipper",
     "locale": "zh_CN"
   }

3. Run photo_tool.exe, set input/output folders, and click Process. Cropped images and feature files are written to the directories you choose; failures appear in the failure list for retry.

---

Related Topics

• Facial Recognition Development Overview — feature vectors, registration, recognition
• Example: Face Registration and Recognition — on-device registration and matching