Title : A Fusion Face Recognition Approach based on 7-Layer Deep Learning Neural Network
Author : Jianzheng Liu, Chunlin Fang, and ChaoWu
Research Object : Face Recognition
- Why Face
- Why Face Recognition
Available methods and Advantages and Disadvantages
- Lu et al Method (Discriminative Multi-Manifold Analysis/DMMA)
- (+) Training is just only one sample face image. Most of Face Recognition training methods aren’t.
- Yang et al Method (Fast 1-minimization algorithm)
- (+) Speed and Scalability
- (+) new solution based on classical convex optimization framework (Augmented Lagrangian Methods)
- (+) Viable solution to real-world & time-critical
- Liao et al (Aligment-free approach)
- (+) Partial face recognition
- (+) no face aligment to recognize by eye coordinate or any fiducial points
- Wagner et al (Simple face recognition system)
- High degree of robustness
- Stability to illumination variation, image misalignment & partial occlusion.
Problem with selected method
- Fusion Feature
- MHI Algorithm for psychological face expression.
- Aligned Gray Image (Face Detection result with Haar Cascade method) and interpolated to 100×100 image.
- Problem of Fusion Feature that have about 20.000 feature generated, still difficult to classified by casual machine learning.
Solution for selected method
- Deep Learning Neural Network (7-Layer). Have advantages to reduce dimension code from high dimensional data to low dimensional codes that called auto-encoder in 6 first layer with DBN (Deep Belief Net). DBN is component of Deep Learning.