Fast Local Laplacian Filters: Theory and Applications

ACM Transactions on Graphics 2014
Mathieu Aubry, INRIA / ENS
Sylvain Paris, Adobe
Samuel W. Hasinoff, Google Inc.
Jan Kautz, University College London
Fr├ędo Durand, MIT

We provide sample results of our algorithm and comparison with some other relevant methods

Large detail enhancement

We provide a comparison of our local laplacian filter with two sets of parameters, our unnormalized bilateral filter, the guided filter [He et al. 2010], the domain transform [Gastal and Oliveira 2011], the adaptive manifolds [Gastal and Oliviera 2012] and the multiscale method of [Fattal et al. 2007] in the case of large detail enhancement.

Style transfer and gradient histograms

Click on the thumbnails to see sample results of our style transfer algorithm. These results were produced by a script that applied the same parameters to all the images, without any manual tweaking.

We provide the comparison with the methods of Bae et al. [2006] and Sunkavalli et al. [2010]. The results for our method are given after 4 iterations (thus after convergence), using the fast LLF algorithm with 60 levels for all images, and finishing by half an intensity transfer.

Additional results without comparison.

We additionally provide the gradient histograms comparison for a large number of transfers, showing that our method transfers them reliably.

HDR compression with the bilateral filter

We present some results of tone mapping and detail enhancement with the different versions of the bilateral filter: the standard version, the version from [Durand and Dorsey, 2002] and our unnormalized verion.