Deep Convolutional Neural Network in Deformable Part Model for Face Detection

 Dataset: Face Detection Data Set and Benchmark (FDDB) [2] : contains 5171 faces in 2845 images with variation of background, illumination, face’s pose and appearance?. Results of stat[r]

The 1 st UTS-VNU Research School

Advanced Technologies for IoT Applications

Results

 Dataset: Face Detection Data Set and Benchmark (FDDB) [2] : contains 5171 faces in 2845 images with variation of background,

illumination, face’s pose and appearance Results of state-of-the-art methods are published in FDDB’s website

 Evaluation: We use standard evaluation protocol provided with dataset

References

[1] Felzenszwalb, P.F., Girshick, R.B., McAllester, D., Ramanan, D.: Object detection with discriminatively trained part-based models Pattern Analysis and Machine Intelligence, IEEE Transactions on 32 (2010)

[2] Jain, V., Learned-Miller, E.G.: Fddb: A benchmark for face detection in unconstrained set-tings UMass Amherst Technical Report (2010)

Conclusion

Our system reveals the fact that :

 structure learning and deep learning can be

integrated together to get the top performance

 DeepFaceDPM becomes new state-of-the-art in

Face detection area

Deep Convolutional Neural Network in Deformable Part Models

for Face Detection

Nguyen Dinh Luan, University of Science, VNU-HCMC, Ho Chi Minh city, Vietnam

Introduction

Face detection is not a new area BUT challenging:

- The variation of face’s pose, lighting conditions

- Does not solve completely with high precision

and speed

What is DPM?

a model use: HOG + latent SVM

+ provides parts and structure of object

- does not exploit high level features

What is CNN?

a type of learning with layers

+ meaningful features

- does not provide explicit relationship

between features

Abstract

Deformable Part Models (DPM) [1] and Convolutional Neural Network (CNN) are state-of-the-art approaches in object detection While DPM makes use of the general structure between parts and root models, CNN uses all information of input to create meaningful features These two types of characteristics are necessary for face detection Experimental results show that our method surpasses the highest result of existing methods for face detection on the standard dataset with 87.06% in true positive rate at 1000 number false positive images Our method sheds a light in face detection which is commonly regarded as a saturated area

Contributions

There are two key ideas

- new 4-5 part Face DPM model for

face detection

- new adaptive way of integrated

CNN into DPM called DeepFace DPM

Fig 1 Superiority of proposed method

First row: results detected by DPM

Second row: results detected by CNN

Third row: results detected by our method

Fig 2 Comparison with state-of-the-art techniques on FDDB dataset