| This software calculates the distance transform of a binary image using nonlinear distances. Providing a binary image with object pixels shown by 0 and background pixels shown by 255, the program creates an image where the value at a pixel will be proportional to the sum of distances of that pixel to all object pixels with a Gaussian proportionality term. The summation process has an averaging effect and reduces the effect of noise. As the standard deviation of the Gaussian is increased, the effect of noise decreases; however, that reduces local shape details as well. If noise is not present Euclidean distance transform is preferred over the Gaussian distance transform because it is faster. If noise is present, the Gaussian distance transform is preferred over the Euclidean distance transform because it can reduce the effect of noise. Select the standard deviation of the Gaussian proportion to the noise level in the image. An example comparing Euclidean and Gaussian distance transforms is given below. |
(a) (b)
(c) (d)
(e) (f)
| Fig. 1. (a) A binary image containing a circle. (b) The same binary image with 5 added random points. (c) Euclidean distance transform of (a). (d) Euclidean distance transform of (b).(e) Gaussian distance transform of (a). (f) Gaussian distance transform of (b). |
To obtain a license for this software, follow this link =>
| Gaussian distance transform |
| Image Registration and Fusion Systems |
 | |  |
