Fisheye Window

Thanks to the study

This is snippet program implements dewarped image extraction algorithm exactly as presented in this research paper:

"High Quality Image Correction Algorithm With Cubic Interpolation for Fish-eye Lens" by Mori Takahiro et al.

CAUTION: Main part of this paper - high quality image correction - is not handled here.

This snippet is...

There are many information found on Web regarding getting panorama image from fisheye image, but these are limited to show off each technique. My snippet shows exact code for getting image from your fisheye source image freely regarding:

• Angle
• Zooming

You can get images as if you are shitting in the center of fisheye image, and looking around. I call this as "window" of the fisheye image, so the class name is FishEyeWindow. Let's see what you can get below.

Sample

Source FishEye image

Sample Results

Parameters

Angle	Parameter
left	alpha=-270.000000, beta=135.000000, theta=270.000000, zoom=0.300000
center	alpha=-270.000000, beta=0.000000, theta=270.000000, zoom=0.300000
right	alpha=-270.000000, beta=-135.000000, theta=270.000000, zoom=0.300000
up	alpha=0.000000, beta=90.000000, theta=0.000000, zoom=0.300000
down	alpha=0.000000, beta=-90.000000, theta=0.000000, zoom=0.300000

How to use

Once you start python snippet like this:

$ python fisheye360.py sample.png

Program will stay running waiting for your control.

Hit followings to move your view:
  'r' or 'f' to zoom
  'g' or 't' to rotate alpha
  'h' or 'y' to rotate beta
  'j' or 'u' to rotate theta
  's' to save current view to ./result.png

Hit ESC to exit.

Requirements

Opencv 2.x would be ok, or try 3.x if you see anything wrong.

Design detail

The FishEyeWindow class is designed based on:

• The center of source image is the center.
• Top/bottom or left/right part of image can be cut, as long as the center is the center.
• Small workaround done for hiding an issue - orthant exchange; explained below.

Regarding orthant issue

If I follow simply as the original math, the result image has problem:

• What should be on the orthant I is on the II, and the II on the I.
• Tha same thing for III and IV.

Then I implemented simple workraound to exchange orthant as you can find in the source.

No copyright, no license

This is just have interpreted math equations into python code, feel free to use.