For motion blur and defocus blur

we propose a robust bispectrum-based method to estimate the blurred parameters in noisy images. At first we calculate the bispectrum of a standard test image. Then we use curve fitting to get the function relations between statistics characteristic in bispectrum and degraded parameters. Finally

the BP neural network

which is trained by the aforementioned function relations

can accomplish the blurred parameters recognition in other noisy image. The experimental results show our method is effective in certain ranges. When SNR is 25dB，the tolerance of blurred parameters recognized is less than 0.5 pixel.