Pyvox is designed to work directly with multi-dimensional image data (up to 8 dimensions) in signed integer, unsigned integer, real floating point, and complex floating point formats from 1 to 16 bytes long. Currently supported operations on such arrays include:

• Reading or writing image files in all supported formats, both big and little endian. For data portability, external files are always written in some specified external format (e.g. big-endian 2-byte two's complement integer or big-endian 4-byte IEEE 754 float), and converted to or from the native format as necessary.
• Image arithmetic, including boolean operators, comparison, transcedental functions, table lookup, and min/max.
• Image resampling to new coordinate systems.
• The morphological operations erode and dilate.
• Univariate and bivariate histograms.
• K-means and nearest neighbor classification.
• Object extraction, where an object is defined as a maximal connected set of voxels.
• Convolution and linear filtering.
• A basic set of matrix operations.
• Fast Fourier transforms.
• Interactive image viewing along any coordinate axis with intensity windowing and selection of data format (which is also useful for determining the format of a unknown image file).
• Interactive and automated image registration.
• Interactive modification of images, especially for manual correction of not-quite-correct automatically masked and segmented images.
• Interactive editing of ROI maps.

Additional capabilities will be added as determined by the needs of Pyvox users. Some areas that are currently under consideration include:

• Improved masking and segmentation algorithms, including handling of shading and partial volume effects.
• Tools for validating brain masking and segmentation algorithms.