Am working on face recognition in video, i want to start work on deep learning based face recognition for better accuracy.
kindly suggest me if i use deep learning based method will it improve the accuracy?
currently am using modified version LBP, ELBP, SIFT, Local Directional based feature extraction.
Basically, deep learning works better in face images (and videos). It may give you a better representation of features. There are plenty works in computer vision using deep learning. You should refer to more papers in CVPR, ICCV, ECCV, etc and see how they use deep learning to address their specific problem. And then modified it to use in your problem.
Maybe you should first start learning CNN.
Some work has been done with Cartesian Genetic programming in this field.
http://link.springer.com/chapter/10.1007%2F978-1-4614-6846-2_3#page-1
Something like that could be useful for your case study. In this paper the authors perform a method to represent ten thousand classes of faces by using Deep Convolutional Networks.
http://www.cv-foundation.org/openaccess/content_cvpr_2014/papers/Sun_Deep_Learning_Face_2014_CVPR_paper.pdf
Basically, deep learning works better in face images (and videos). It may give you a better representation of features. There are plenty works in computer vision using deep learning. You should refer to more papers in CVPR, ICCV, ECCV, etc and see how they use deep learning to address their specific problem. And then modified it to use in your problem.
Maybe you should first start learning CNN.
Deep learning works better but there are other methods also like using HOG3D descriptor and color histogram. These can give better spacio-temporal features. For more details you can go through CVPR papers.
You can go through below paper:
"A. Klaser,M. Marszałek,and C. Schmid. A spatio-temporal descriptor based on 3d-gradients. In BMVC, 2008"
Wavelet network theory could be used particularly with beta wavelet - fast wavelet transform . Refer the paper titled, " A novel apporach for face recognition based on fast learning algorithm and wavelet network theory" Mourad Zaied et al. International journal of Wavelets, Multiresolution Information Processing, 09,923,2011.
I would suggest to follow facenet architecture .
Below is the paper URL .
https://arxiv.org/abs/1503.03832
Dear Mohan Raj,
Read my articles about face recognition, please. May be useful.
https://www.researchgate.net/profile/Shafagat_Mahmudova/publications
Regards, Shafagat
The principle behind deep learning can be applied to most of the problem domains. In the context of the problem stated by you, I believe you would still need to use a face detection mechanism, let it be HOG, SIFT, Haarscale detector etc.. The 'deep' learning mechanism would be needed to either to give feature vector representation (unsupervised learning) or labels (supervised learning) to the image frames. These feature vectors can either be learnt using a deep learning mechanism or can be hand crafted like the HOG or LoG or anything. So, principally, if you do not want to hand craft these features, use deep learning (unsupervised) after extracting a face from the video frame or otherwise split the video into frames, extract face images, apply HOG/LoG etc. and extract feature vectors and then feed it to a multi-layered network giving it a label for recognition. Let me know if this helps, I can perhaps give you a better response as well based on your specific questions.
Currently deep learning is the top algorithm in computer vision problems including face recognition. In "Labeled Faces in the Wild LFW" benchmark, the methods use deep learning are in the top of the benchmark result page in terms of accuracy. However, since you are recognizing faces in videos, face poses is still a problem using deep learning. You need to use some other algorithm first to estimate and correct poses first pefore you apply deep learning.
Article Unconstrained Facial Recognition Systems: A Review
Hi!
Using deep learning would definitely improve the accuracy as indicated in the other answers.
You may refer to the following articles for a better understanding of how deep learning can leverage very large datasets of faces, learn rich and compact representations and outperform its compeers:
Best
Deeksha
Deep learning is one of the most novel ways to improve face recognition technology. The idea is to extract face from images with faces. Such facial will be unique for different faces. And training of a deep neural network is the most optimal way to perform this task.
For more information, we have:
https://machinelearningmastery.com/introduction-to-deep-learning-for-face-recognition/
https://www.sciencedirect.com/science/article/abs/pii/S1077314219301183
https://www.hitechnectar.com/blogs/deep-learning-face-recognition/
- Badges
- Science topic