Noise histogram regularization for iterative image reconstruction algorithms.

We derive a regularization term for iterative image reconstruction algorithms based on the histogram of the residual difference between a forward-model image of a given object estimate and noisy image data. The term can be used to constrain this residual histogram to be statistically equivalent to the expected noise histogram, preventing overfitting of noise in a reconstruction. Reconstruction results from simulated imagery are presented for the cases of Gaussian and quantization noise.