Inpainting algorithm using wavelet transforms

How to reconstruct a missing background portion?

Inpainting is the process of reconstructing lost or deteriorated parts of images and videos, using information within them. The missing area of the image is called region of inpainting, and it must be filled so that the structure and the texture in reconstructed image remains coherent.

The project proposed here aims to create an algorithm capable of reconstructing the background of an image due to the removal of a portion of it. This is achieved through the wavelet decomposition at three levels and specialising the linear prediction for each sub-band decomposition. Following empirical choices you have chosen to use the wavelet decomposition with Haar filters.

Essential operations in the execution of the algorithm are:

  • choice of the image;
  • Selection by the user, portion of the image that has to be removed and then rebuild;
  • Decomposition of the image through the wavelet transform;
  • Linear prediction and personalised independent of each sub-band;
  • Inverse transformation in order to recompose the image and display the result.

To evaluate the results of the algorithm we chosen to remove a known area and predict that; the content and the result was compared, in terms of PSNR, with the original. The predictions interpolation types and the results are shown below.

 

Take a look also to the project presentation (link – ITA) or download the MATLAB src.