Gradient Vector Flow

Image gradient allows calculating image gradient in a specified direction [1] . It is a preprocessing step for active contours algorithms. This option opens a dialog window like the one shown below:

Different algorythms are supported:

• GVF: calculates gradient using gradient vector flow algorithm;
• General GVF: calculates gradient using generalized gradient vector flow algorithm;
• Differential gradient: calculates gradient substracting values of neighbouring pixels;
• Sobel operator:calculates gradient by convolving filter mask with matrix consisting of image pixels

During the estimation procedure the following parameters can be set:

• Gradient width: represents distance (in pixels) beween two points, the intensity difference which defines the gradient value;
• Iterations: are the number of iterations performed during GVF and GGVF calculation;
• Smoothing parameter (m): is the regularization parameter governing the tradeoff between the first and the second integral term. Smoothing parameters should be set according to the amont of noise present in the image: the higher the noise the bigger the value.
• Time step (D): is representing the time lenght for each iteration.

In order to guaratee the algorithm convergence, the smoothing parameter and the time step should satisfy the following expression: m<-1.36*Dt+0.22. Therefore, the Convergence restriction box should be enabled.

Reference

[1] Chenyang Xu, Jerry L. Prince. Snakes, Shapes and Gradient Vector Flow, Tansactions on Image Processing, March 1998, p. 359-369.