Haze is light which is reflected from particles in the atmosphere, between the subject of the photo/video and the camera.
Abstract
Haze is light which is reflected from particles in the atmosphere, between the subject of the photo/video and the camera. Haze causes change in the perceived color of the subject, loss of contrast which is dependent on the distance of the subject from the camera and loss of fine detail in the subject. Dehazing has much value in applications of computer vision, image restoration as well as aesthetic value. In this project we implemented a polarization-based method for dehazing of video, using a special dual-channel video camera. The code was implemented in MATLAB and works on pre-recorded video and in real-time in Simulink.
Polarization-based Dehazing
- Several dehazing methods exist – single image / multiple images, physics based / image enhancement based
- The polarization-based method (described by Prof. Yoav Shechner) is based on a physical haze model and requires two images taken through a polarizing filter
- A significant advantage – minimal computational resources
- The main complexity, in particular for video applications, is in the image acquisition process
- Dehazing video using this method requires two synchronized, aligned video streams, each taken through a different polarizer
Conclusions and possible future steps
- Dehazing algorithm was successfully implemented using the Dual Channel camera
- Video results although B/W show improvement in picture quality
- Real time operation achieved
Possible Future Steps
- System mobility is needed to test the camera for different sceneries
- Real time implementation using programming platform (C++, etc) is needed for 24 FPS and higher results
- Automatic parameter estimation will remove the scenery limitation from the algorithm