convolution: Convolution

Module for convolution on CPU/GPU.

class Convolution(shape, kernel, axes=None, mode=None, ctx=None, devicetype='all', platformid=None, deviceid=None, profile=False, extra_options=None)

Bases: silx.opencl.processing.OpenclProcessing

A class for performing convolution on CPU/GPU with OpenCL.

convolve(array, output=None)

Convolve an array with the class kernel.

Parameters

• array – Input array. Can be numpy.ndarray or pyopencl.array.Array.

• output – Output array. Can be numpy.ndarray or pyopencl.array.Array.

gaussian_kernel(sigma, cutoff=4, force_odd_size=False)

Generates a Gaussian convolution kernel.

Parameters

• sigma – Standard Deviation of the Gaussian curve.

• cutoff – Parameter tuning the truncation of the Gaussian. The higher cutoff, the biggest the array will be (and the closest to a “true” Gaussian function).

• force_odd_size – when set to True, the resulting array will always have an odd size, regardless of the values of “sigma” and “cutoff”.

Returns

a numpy.ndarray containing the truncated Gaussian function. The array size is 2*c*s+1 where c=cutoff, s=sigma.

Nota: due to the quick decay of the Gaussian function, small values of the

“cutoff” parameter are usually fine. The energy difference between a Gaussian truncated to [-c, c] and a “true” one is

erfc(c/(sqrt(2)*s))

so choosing cutoff=4*sigma keeps the truncation error below 1e-4.