Volume Filter Volume Filter icon

Volume Filter includes several options for smoothing or transforming volume data. See also: Hide Dust, vop, mask, segment

There are several ways to start Volume Filter, a tool in the Volume Data category (including from the Volume Viewer Tools menu).

Filter type:

Clicking Filter processes the current set of data in Volume Viewer. If the current set is the result of a previous application of the same type of filtering, it will simply be updated to reflect the new settings (the original data are re-filtered with the new settings). Otherwise, a new volume data set will be created and made the current set. The new data set can be saved to a file with Volume Viewer.

Clicking Options reveals additional settings (clicking the close button on the right hides them again):

Close dismisses the Volume Filter dialog; Help opens this manual page in a browser window.

TECHNICAL NOTES

Gaussian filtering. Convoluting the data with a Gaussian function improves the ratio of signal to noise but reduces resolution. It is fastest for data sizes that are powers of 2, and can be very slow when insufficient memory is available. It produces a a map with 32-bit floating point values and uses negligible additional memory. It may be helpful to limit the input to just a subsample or subregion of the original data. Although it uses a fast Fourier transform calculation method, it does not use map periodicity. Values outside the map boundaries are treated as zero.

Laplacian filtering. The Laplacian operation is a sum of second derivatives. Laplacian filtering is useful for edge detection but amplifies noise, so it may be necessary to perform smoothing such as Gaussian filtering beforehand. Finite differences v(i-1)-2*v(i)+v(i+1) along each axis are used, and voxels at the edge of the box are set to zero.

Fourier transform. Only the magnitudes of the complex Fourier components are included in the new data set; the phases are discarded and the constant component is set to zero. The box containing the Fourier transform (with axes in units of reciprocal space) is centered on the original data and scaled to have the same total volume. Some properties of the original data are evident from the Fourier transform. High-frequency components are near the edges of the box, low-freqency components near the center. Volume data is typically oversampled (voxel size two to three times smaller than the actual data resolution) and this causes the Fourier transform to have nonzero values only in the middle half or third of its bounding box. The missing wedge in electron microscope tomograms can also be seen. Spikes radiating along the principal axes in the Fourier transform are caused by nonperiodicity of the original data.


UCSF Computer Graphics Laboratory / June 2014