Empty gaps between rasters in Mosaic Dataset

[JerryTGroom]

I have several scenes collected by Maxar's Worldview-2 satellite that I am trying to pansharpen and stitch together for a forest-wide remote survey. Because of the color difference between these, I've also been asked to balance the coloration between raster. To get right to it, here is my workflow:
Download imagery from G-EGD (ortho all bands).
Create dedicated GDB
Create mosaic dataset (product definition = WV2)
Add rasters to mosaic dataset (raster type = WV2, Processing template pansharpen and multispectral, input folder)
Data filter = *.IMD; Calculate statistics, build pyramids

I've included a screenshot of the resulting mosaic for reference, which shows gaps where the rasters overlap.

If I add the rasters/IMD files individually, the overlap is significant and has no "dead-zone" gaps between imagery. I would consider this route, but in doing so, I lose the ability to pansharpen specific bands (Red, Yellow, and Coastal Blue) and more importantly, the ability to perform a color balance on the rasters. Only when adding rasters to the mosaic dataset do I get these empty slivers. I've tried rebuilding footprints, Build Overviews, and synchronizing the rasters, all to no effect.

If anyone has any suggestions or guidance on this, I look forward to hearing your pro-tricks!

Thanks much!

[tikola]

I do not have an answer and I am looking same thing - The Best Pratices in Mosaic. The thing you are seeing is that images has no data/empty areas and now you are bring in that empty data in mosaic process. You need things like calculate active area of images when bringing them to process - so that only valid data is included. Another solultion is to manage image order for mosaic but as all files might have equal no data areas it most likely just changes the location of slivers. 

So I would first try build seamlines or compute mosaic candidates geoprocessing to see does it solve the problem. In nutshell for satellites (if no cloud coverage for changing things) the good approach is to find most nadir images in process so that seamline gets build in a way that it follows the maximum difference to image center to minimize geometric errors in data. 

Another way I would like to learn is how to handle NODATA/real data in smart way. So that while bringing data in process simply ignores NoData areas which is more or less sama thing as my "compute active area" procesa above.

I have same learning undergoing. All samples tends to have perfect align of separate files but in real life all kind of tilt/sliver etc. issues exist as in your case and learning smartest ways to live with that is a thing I am learning now too.