Image restoration using steerable pyramids

On this page, we present image restoration results of our own developed technique, compared with standard methods for image restoration. The method we propose is a hybrid method based on steerable pyramids for noise reduction and blur estimaion (our method doesn't require knowledge about the Point Spread Function or PSF) and the Richardson-Lucy (RL) deconvolution method for deblurring, which we will call SPERRiL (Steerable Pyramid-based Estimation and Regularization of Richardson-Lucy restoration). In SPERRiL, we can choose to apply regularization every iteration, or once every n iterations.

Synthetic Test Image

True synthetic testimage	Degraded (blur + Poisson noise)	Unregularized RL	RL + slight postblur
	SPERRiL, n = 1	SPERRiL, n = 3	SPERRiL, n = 8

Real Confocal Image

Raw confocal image	Unregularized RL	RL + slight postblur
SPERRiL, n = 1	SPERRiL, n = 2	SPERRiL, n = 3