Fast Bilateral Filter

Fast O(1) implementation of the non-linear bilateral and the joint (or cross) bilateral filter.

Description

MATLAB implementation of the fast O(1) bilateral filter described in the following papers:

[1] K.N. Chaudhury, D. Sage, and M. Unser, "Fast O(1) bilateral filtering using trigonometric range kernels," IEEE Trans. Image Processing, vol. 20, no. 11, 2011.

[2] K.N. Chaudhury, "Acceleration of the shiftable O(1) algorithm for bilateral filtering and non-local means," arXiv:1203.5128v1.

Also included is the MATLAB implementation of joint bilateral filtering.

• MATLAB release: MATLAB 7.10 (R2010a)

[Kunal Chaudhury]

Ben: Yes, you can modify the code for volume data as well. The dimension of the data is immaterial.

[Ben]

A quick question before trying it: does the code support 3D image?

[Kunal Chaudhury]

Yes, it can introduce artifacts in the output if the actual T is much larger than 255. This could well be the case near edges
when processing images corrupted by noise
with large sigma. A better guess in this case would be T + C(sigma), where C(sigma) is a reasonable upper bound on the intensity variation induced by noise.

[PeeVee]

Does setting T to max (255 in my case) decrease the accuracy of the final result?

[Kunal Chaudhury]

One can always use a good guess of the local dynamic range. The 'maxfilter' is not required in this case. Note, however, that the local dynamic range can be much smaller than 255(the global gynamic range) for a grayscale image, and in some cases, it might be difficult to make a good guess of T (eg, if there's a lot of noise).

[PeeVee]

Most of the time is spent on maxfilter for large image sizes. Is it really necessary, or can we just set T=255?

[Kunal Chaudhury]

Thanks Derek. Bug fixed.

[Derek]

Thanks! One bug: change line 66 in maxFilter.m to
R(m - k + 1) = max( R(m - k + 2), template(m - k + 1, jj));

otherwise will get error if width<height

[Curtis]

Nice effort!