“…As the first step towards segmentation, a background subtraction (as described in 2.1) can be applied, here affecting only this step of the analysis. Second, default de-noising is performed by applying a Gaussian filter with a radius of 0.6 px [59] and by subtracting the estimated mean of the uniform Gaussian white noise, replacing the homogeneity analyzer proposed in [60,61], which is utilized to identify the empty portions of the image, by the Absolute Difference Mask (ADM) edge detector [62]. Candidate object locations are then detected using the ''à trous'' wavelet transform [30], followed by a filtering step in which only single-pixel local maxima detections are kept.…”