Foveated Video Compression

A study and comparison between quantization based foveated video compression and interpolation based variable resolution technique.

Required dependencies

• python 3.6 or higher
• numpy
• opencv
• imutils
• pillow (PIL)
• argparse

How to run

• Make sure the required dependecies are installed

• Unzip the FoveatedCompression.zip

• Open terminal and cd to the extracted directory

1. VR compression:

• In the terminal write the command below and run.

    `python detect_faces_vr_compression.py -i image/test1.jpg`
  

• Here, image/test1.jpg is the location of the test image

• Press q to iterate through the output images

• the output images are saved in the root directory

2. Quantization foveated video compression:

• In the terminal write the command below and run.

    `python fovea_video.py -v video/test5.mp4`
  

• Here, "video/test5.mp4" is the location of the test video

• To change the quantization rate use the command below:

    `python detect_faces_video.py -v video/test5.mp4 -q 70`
  

  Here, -q is an optional argument.

• Press q to stop streaming and close all window

• the output frames are saved in the saved_frames folder

Results

VR Compression result

Color quantization Video Compression result

Output Comparison: Color quantization Video Compression vs VR Compression

Output Comparison: PSNR, SSIM, Compression ratio

Developers

Naimur Rahman Jeem

Subho Ghose

Hanming Li

Acknowledgements

The basic code structure is based on the example code of pyimagesearch.

If needed, please contact at rahmanje[at]ualberta[dot]ca, subho[at]ualberta[dot]ca, hanming[at]ualberta[dot]ca.