Face detection: mtcnn

The paper ：Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks
download ：https://arxiv.org/abs/1604.02878
Code ：https://github.com/kpzhang93/MT

Face detection is a problem in computer vision , That is, locate one or more faces in the photo .

Locating a face in a photo means finding the coordinates of the face in the image , The range of the face is divided by the bounding box around the face .

The face is dynamic , Its appearance is highly variable , Face detection becomes a difficult problem in computer vision . Such as face detection , Subject to its orientation or angle 、 Light level 、 clothing 、 Accessories 、 Hair color 、 Facial hair 、 Make up 、 Age and so on .

Face detection is an essential first step in face recognition system , Its purpose is to locate and extract face regions in the background .

There are two main methods of face recognition ：

• Feature-based approach , Use manual filters to search and detect faces . have access to OpenCV Library cascade classifiers (cascade classifiers) Conduct .
• Image based approach , Learn how to extract faces from the whole image as a whole . Can pass MTCNN Libraries use multitasking cascading CNN To achieve .

MTCNN Is a deep cascading multitasking framework . The framework is used to solve various postures 、 Lighting and shielding , Face detection and alignment in an unconstrained environment .

The framework takes advantage of the inherent correlation between detection and alignment to improve their performance . The framework uses cascaded architecture and three stages of a well-designed deep convolution network to predict human faces and landmark Location . Besides , A new online hard sample mining Strategy , Further improve the performance in practice .

notes ： Face alignment face alignment： It is a necessary preprocessing in many face applications , The purpose is to reduce the rotation of the input image (Rotation)、 translation (Translation)、 The zoom (Scale) Loss of accuracy due to change .

---

---

---

Reference material ：
Face detection MTCNN brief introduction
Face algorithm series ：MTCNN Detailed explanation of face detection
MTCNN working principle