In my opinon, no. You can estimate noise signal from the residuals, after data fitting.

Depends what do you mean by S/N ratio in FLIM and what do you consider as FLIM data in the first place. The noise in photon counting is governed by Poisson statistics, so the value of noise in each TCSPS bin depends on the number of photons in that bin. FLIM images can be constructed in several different ways - using mean lifetime (which is basically the integral of the TCSPC histogram), phasors (the Fourier transform of the TCSPC histogram), actual fluorescence lifetimes obtained by multi-exponential fitting (tail- or reconvolution fitting), ... . The error of the output image and the way how it can be estimated will then depend on the way how the output is obtained from the measured TCSPC histograms, so there is not a simple general answer to your question.

Radek Machan , Thank you for your detailed answer. However, the crux of my problem is exactly that I cannot use either of those methods. In my samples, I have discovered an artifact (similar in characteristics but still varying to a degree across samples regardless of makeup, conditions and acquisition settings) which permeates the sample in a way which affects the temporal distribution of each pixel's photon count. The distribution is complex, with the IRF being easily described as a wave-like portion of bands. Its histogram followed a skewed or a bimodal distribution which is hard to characterise or include in a reconvolution fitting equation.

Therefore, I decided perform pixel-by-pixel fits in which the starting and peak points of the photon curves would not be affected by this strange distribution. In this way I can export the fitted lifetime values for each pixel to an image, get the histogram, fit it to a model which would describe it well (I've found that Gamma variate does well in that regard) and get the peak value as the true average lifetime.

However, since photon counts of single pixels are too little (