PLP biology suggests that this is normal, you should use only the signal from the contact surface of two type of cells if i understand well your problem..

Hard to help without seeing and playing with the data myself, but a good start is to open the two channels in ImageJ, smooth or otherwise decrease resolution significantly, make channels binary (black/white) at some threshold that makes sense, and save this data. Now use this simplified data set to test co-localization functions and various math functions until you have built a strategy you're happy with.  Now redo the processing on the original files, retaining slightly more detail, then slightly more again, and do iterations of varying levels of data simplification until you find a balance.  Be exact with how you describe the processing in the methods section (possibly have a supplemental figure demonstrating the results of each step on example data).

Best of luck!

p.s. I should note I'm aware you might not like this answer, but simplification makes sense to me in this case because you're essentially labeling tubing around some wiring within that tubing, so the argument can be made that measuring co-localization at high resolution and magnification is not going to accurately capture what you're after ("is this tube around these wires?").

this is really difficult to answer, because of many influences, but perhaps the following publications might help you to:

a) get an idea about quantifying protein levels

b) how to calculate distances and how to make corrections

if you look at more of my papers you might get a better idea even

so here the list:

for quantification:

Counting nucleosomes in living cells with a combination of fluorescence correlation spectroscopy and confocal imaging. Weidemann T, Wachsmuth M, Knoch TA, Müller G, Waldeck W, Langowski J. J Mol Biol. 2003 Nov 21;334(2):229-40.

Analyzing intracellular binding and diffusion with continuous fluorescence photobleaching. Wachsmuth M, Weidemann T, Müller G, Hoffmann-Rohrer UW, Knoch TA, Waldeck W, Langowski J. Biophys J. 2003 May;84(5):3353-63.

Trichostatin A-induced histone acetylation causes decondensation of interphase chromatin. Tóth KF, Knoch TA, Wachsmuth M, Frank-Stöhr M, Stöhr M, Bacher CP, Müller G, Rippe K. J Cell Sci. 2004 Aug 15;117(Pt 18):4277-87. Epub 2004 Aug 3.

Detection of NGF-receptors TrkA and p75NTR in human tumor cell lines and effect of NGF on the growth characteristic of the UT-7/EPO cell line. Westphal G, van den Berg-Stein S, Braun K, Knoch TA, Dümmerling M, Langowski J, Debus J, Friedrich E. J Exp Clin Cancer Res. 2002 Jun;21(2):255-67.

so and for high resolution stuff and quantification:

Light optical precision measurements of the active and inactive Prader-Willi syndrome imprinted regions in human cell nuclei. Rauch J, Knoch TA, Solovei I, Teller K, Stein S, Buiting K, Horsthemke B, Langowski J, Cremer T, Hausmann M, Cremer C. Differentiation. 2008 Jan;76(1):66-82. Epub 2007 Nov 26.

The 3D structure of the immunoglobulin heavy-chain locus: implications for long-range genomic interactions. Jhunjhunwala S, van Zelm MC, Peak MM, Cutchin S, Riblet R, van Dongen JJ, Grosveld FG, Knoch TA, Murre C. Cell. 2008 Apr 18;133(2):265-79. doi: 10.1016/j.cell.2008.03.024.

Super-resolution imaging reveals three-dimensional folding dynamics of the β-globin locus upon gene activation. van de Corput MP, de Boer E, Knoch TA, van Cappellen WA, Quintanilla A, Ferrand L, Grosveld FG. J Cell Sci. 2012 Oct 1;125(Pt 19):4630-9. doi: 10.1242/jcs.108522. Epub 2012 Jul 5.