Diego
you are seeking for the white areas to be "NoData", to allow valid pixels to show through. There are two methods - 1) define nodata as 255 in the properties of the mosaic dataset, but note if these rasters are compressed (JP2, MrSID), some of the white values will be 254, 253 etc. due to compression, so you'll see undesirable white pixels around the edges. You can define multiple values as NoData but you may lose bright pixels in the interior of your image (e.g. white buildings, cars will have transparent "holes")
2) the other method is to build footprints (which defines the extents of valid pixels) and then set property "clip raster to its footprint = YES". For any normal mosaic, I recommend method 2, but in your case, the boundary is so complicated and convoluted, calculating (or manually entering) a proper footprint will be very difficult and will have so many vertices that performance will suffer.
Given the two complicated images you are working with, you are facing some serious limitations. The optimum implementation would have been to build both of your large rasters as Mosaic Datasets using the hundreds (thousands?) of original images. You'd have a lot more flexibility for color correction, adding/removing updated images, etc. As is, with two large mosaics with pixels "baked in" you have very little flexibility and I suspect will run into a lot of other headaches. Do you have the original, individual images? Building a mosaic dataset of thousands (or even millions) of individual images is not difficult...
Cody B.
