gpeyre/2012-siims-nl-ac

M. Jung, G. Peyré, L. D. Cohen. Nonlocal Active Contours. SIAM Journal on Imaging Sciences, 5(3), pp. 1022–1054, 2012.

This Matlab package contains the source code to reproduce the figure of the article:

M. Jung, G. Peyré, L. D. Cohen. Nonlocal Active Contours. SIAM Journal on Imaging Sciences, 5(3), pp. 1022–1054, 2012.

========= main_nonlocal_seg.m

Non-local Active Contours, Two Non-local models: "un-normalized" (NL-U) / " normalized" (NL-N), using:

• (1) L2-distance: pixel values (intensity, color, Gabor coefficient)
• (2) Wasserstein distance: texture
• (3) Motion distance using motion profiles

========= main_other_methods.m

Implements:

• GAC,
• CV,
• IAC (pixel based)

Copyright (c) 2011 Miyoun Jung

2012-siims-nl-ac's Issues

Wrong Calculation of the Weights at the Function `compute_l2_dist_gray`

At line 39 of the function you calculate the weights of the Gaussian Kernel:

K = GauK(q1+1-length(selx):q1,q1+1-length(sely):q1);

While this tricks works well on the left side of the image, it doesn't work on its right side.
Imagine you are 2 pixels left to the boundary on the right side.
The selection of the pixel would be from far left until the center + 2 on the right of the center of the Gaussian.
Yet the selection of the weights from the Gaussian would be 2 left to the peak, the peak and then down till the boundary.

Multiplying them means multiplying them in the wrong order which creates the wrong weights.

The Meaning of the `lambda` Parameter in the Code

Hello,
Saw that you include the lambda parameter in the code yet it is no where to be found on the article.
Could elaborate on its meaning?

Thank You.