Embedded Files
Idea
Idea
The idea of this project is from one of my friend, who use a smart ring bell to monitor who is downstairs. Although the camera has a reasonable physical resolution, the video quality still remains not as good enough as expected when we review that.
Compression of the video saves the size of the video that suitable for transfer through Internet. However, when we have the demand to watch the poor quality video especially when you want to get more detailed information, the blur picture becomes the biggest obstacle. The goal of this project is to enhance the quality of the blur video with the aid of deep learning.
Structure
Structure
Key Python Packages:
Key Python Packages:
- Theano (Numerical Computation Library )
- Lasagne (Deep Learning Framework )
- SciPy, Pillow (Imaging Library)
- OpenCV (Video Processing)
- FFmpeg (Audio Processing)
Background
Background
The processing part uses a combination of techniques from several papers. The main techniques for constructing a neural network list as follow：
- SRCNN, three-layer convolutional network trained to minimize per-pixel loss [1, 2, 3]
- ESPCN, efficient sub-pixel convolutional neural network [2]
- DRCN, deeply-recursive convolutional network [3]
- SRGAN, Super-Resolution Using a Generative Adversarial Network [4]
Google Sites
Report abuse