Reprojecting datasets as part of a pre-processing operation introduce distortions due to the warping. Moreover, datasets that are adjacent in their native space may be transformed to "tilted rectangles" once reprojected.
The stepped edges of these transformed rectangles may not have a perfect matching on the borders since rasters have discrete pixels.
Isolated transparent pixels (small dashed transparent lines) can be visible when mosaicking the reprojected datasets together.
Whenever having some partial overlapping on input datasets before the warping is not possible, it would be great to have a process filling these transparent stripes.
I have setup a transparencyFill raster process based on a custom JAI operation which scans the image (in case it contains some transparency) looking for horizontal/vertical 1 pixel-thickness stripes, and fill them with neighbour pixels (simple clone or average of the edge).
Once stable and well tested, I will move it to a separate JAI operation in JAI-EXT project.