RECONSTRUCTION OF SPECKLED IMAGES USING BISPECTRA

The bispectrum of a signal or an image has useful properties, such as being zero for a Gaussian random process, retaining Fourier transform phase and magnitude information, and being insensitive to linear motion. These properties are used in reconstructing images contaminated by coherent speckle noise when the latter can be modeled as multiplicative noise. We used a logarithmic transformation to convert such speckle noise to a signal-independent, additive process, which is close to Gaussian when an integrating aperture is used. Bispectral reconstruction is then performed with multiple independent snapshots.