摘要
提出了一种用神经网络鉴别退化图像的模糊类型的方法。由于采用不同降质方法得到退化图像的频谱差异较大,以此作为判别依据,用概率神经网络实现了对四种模糊类型:离焦,矩形,运动和高斯模糊的鉴别。根据神经网络的鉴别结果决定点扩散函数的初始估计值,可大大地提高盲解恢复算法的复原质量和系统点扩散函数的估计精度,扩大了算法的实用范围。
An original solution of the Dlur identification problem was presented, Because the certain blur leads the specific distortion of the image Fourier spectrum amplitude. According to this, a neural network based on probability was used for the blur identification. Four types of blur were considered: gaussian, rectangular, motion and defoeus ones. Using neural network' s output as the preliminary estimation of the PSF in the blind deeonvolution algorithm can significantly improve restored images' quality and the precision of the PSF. This method also expands the application of blind deconvolution algorithms.
出处
《光学技术》
EI
CAS
CSCD
北大核心
2006年第1期138-140,共3页
Optical Technique
关键词
神经网络
模糊
PSF
盲解卷积
频域
neural network
blur
PSF
blind deconvolution
frequency domain