A modified Richardson–Lucy algorithm for single image with adaptive reference maps

• We introduce reference maps to help improve RL algorithm.
• The reference map is adaptively updated to obtain a more accurate edge description as a local constraint.
• Image priors are adopted as the regularization.
• Our approach makes a good balance between preserving image details and suppressing ringings.
• The proposed method has a good performance as well even if the blurred image contains rich texture.

In this paper, we propose a modified non-blind Richardson–Lucy algorithm using adaptive reference maps as local constraint to reduce noise and ringing artifacts effectively. The deconvolution process can be divided into two stages. In the first deblurring stage, the reference map is estimated from the blurred image and an intermediate deblurred result is obtained. And then the adaptive reference map is updated according to both the blurred image and the deblurred result of the first stage to produce a more accurate edge description, which is very helpful to suppress the ringing around edges. Gaussian image prior is adopted as the regularization to improve the standard Richardson–Lucy algorithm. Experimental results show that the presented approach could suppress the negative ringing artifacts effectively as well as preserve the edge information, even if the blurred image contains rich textures.