This image is very similar to the binary mask, the only difference is that for each pixel of the DSM that doesn't contain a measurement we store the euclidean distance to the closest pixel that actually does contain a valid measurement. Each pixel containing a valid measurement will therefore end up with the value 0. Close to tile borders the closest measured pixel could very well be in a neighboring tile. In order to find the closest measured pixel SURE will therefore search the extended DSM. The number of pixels used as extension on all four sides of the extended DSM is controlled via $DsmTempApronSize in controlDsm.txt. Values larger than $DsmTempApronSize close to the tile border of the distance mask could mean that the closest measured pixel is outside that extended DSM and the value in the distance mask is incorrect.
The left image is a DSM and the right image is the derived distance mask. The distance mask was scaled for viewing in a way that a distance of 15 pixels (and everything above) is white and a distance of 0 is black: