Inhomogeneous deconvolution in a biological images context

In this paper we present a new framework for correcting partially out-of-focus biological images. To evaluate depth variation, our method extracts similar objects of interest over the image. The depth is based on a defocus value computation for each salient point. We use an iterative algorithm to convert each defocus value into a Gaussian standard deviation. These measures are then interpolated over the whole image using thin plate spline transform, providing a regularized map. This array provides, for every spatial position, an evaluation of the standard deviation used in the final step for an inhomogeneous deconvolution based on the Richardson-Lucy algorithm. Experiments on real and simulated images validate the accuracy of such a method compare to traditional shift invariant deconvolution approaches.