Algorithm 1: Fingerprint segmentation

Data: Volume V, Mask M (optional)    Result: Surface S    Def. $f_w$ := region scaling factor (hardware dependent)    Def. $w_N$ := initial search region size (hardware dependent)    Def. $L_N$ := set of initial kernel sizes (scan size and resolution dependent)  1   Copy the full volume scan V from CPU memory to $V_0$ located in GPU memory    ▹Intensity roll-off compensation:  2   Store the average intensities of the en-face slices $V_0(:,:,z)$ as a vector $A\left(z\right)$  3   Run a linear regression on $A\left(z\right)$ to measure the deviation $\delta \left(z\right)$ from the average  4   Set $V_0(x,y,z)\leftarrow V_0(x,y,z)-\delta \left(z\right)\forall x,y,z$ to normalize the intensity profile    ▹Volume pyramid and contrast enhancement:  5   Generate $N>0$ additional versions of $V_0$ where $V_{n+1}$ is the result of downsampling $V_n$ by a factor of two along all three dimensions (note: for upsampling and downsampling, the GPU’s texture mapping unit is used in linear interpolation mode)