Parameters identification for blur image combining motion and defocus blurs using BP neural network

Motion blur and defocus blur are two usual blurs of image degradation. Identifying the motion blur length, angle and the defocus blur radius accurately is the prerequisite of the subsequent image restoration. The identification of blur parameters in images combining the two blurs is studied in this paper in the Fourier transform domain. Radon transform is used to identify motion blur angle. Then we estimate the motion blur length and defocus blur radius of the coupling blurred image with BP neural network. The input of BP network is the sum of the amplitudes in the central region of the spectrum. Before summing, we apply median filter to the spectrum to keep or strengthen the helpful variables and cut down the interfering ones. The filters use 1×5 neighborhoods and 5×1 neighborhoods separately. The amplitude sums for identifying motion blur length and defocus blur radius are different. The former is the result of summing the amplitudes horizontally, while the latter comes after summing the amplitudes on the concentric circles from the inside out. We also apply data normalization to the sums before inputting them to the BP neural network. The simulation experiments show that our novel scheme is correct and efficient.