Detection of Small Local Intensity Changes in CCD Images with Nonuniform Illumination and Large Signal Dependent Noise

We demonstrate two efficient methods for detecting small local changes in intensity between two CCD images nonuniformly illuminated and corrupted by a signal-dependent noise. A pixel to pixel relative variation is computed to detect changes in intensity on the initial field. Due to the nonuniform illumination this method increases the noise in the computed image. However, this noise can be removed by taking into account its signal-dependence properties. We propose two filtering algorithms, both based on local properties of the pixels. The first is a new low-level algorithm based on adaptive thresholding. The second uses the wavelet transform and provides a multiscale vision of the significant changes in intensity. Both methods were successfully applied in neurobiology to detect the spatial distribution of depolarized patches of membrane during the excitation of single neurons in culture.