Superpixels Filter

This algorithm computes an isotropic filtering of an image using a superpixel oversegmentation.

Input

The image or volume to be filtered and optionally a second reference with the same dimension (but not necessarily with the same spacing) from which the superpixel oversegmentation will be computed. If only one input is specified, it will also be used for the superpixel computation.

Output

The filtered float image/volume with the same size as the input.

Description

This algorithm consists in replacing all pixel values by the average within its corresponding superpixel (see documentation of the Superpixels Segmentation algorithm). A typical use case of this algorithm is to smooth and upsample a low-resolution pixelwise prediction map to the original image resolution.

In order to avoid artifacts due to the superpixel segmentation, the filter can be run Filter Iterations times with different random oversegmentations. The output will then be the average of the different runs.