基于改进模拟退火的神经网络降质图像恢复

Degraded image restoration using neural network based on improved simulated annealing

随着科学技术的发展,图像处理技术已经成为科学研究不可或缺的强有力工具,而图像恢复是图像处理中非常重要的一环。传统基于模拟退火算法的神经网络降质图像恢复方法,为了避免退火过程过早收敛,对温度的降低不得不慢慢进行,这样导致算法运行时间太长。采用改进的含回火过程模拟退火算法降温,实验表明该改进算法求解时间比传统的方法有了很大的提高,图像的恢复效果也较令人满意,比传统的逆滤波、维纳滤波方法具有更好的峰值信噪比。

With the development of science and technology, the image processing technology already has become the scientific research indispensable powerful tool, but the image restoration is a very important link in the image processing. To avoid early convergence, degraded image restoration using neural network based on traditional simulated annealing has to decrease the temperature slowly, as a result, the algorithm has a long runtime. An improved simulated annealing algorithm which include a backfiring process is adopted. The experiments show that this algorithm is effective for image restoration and greatly reduces runtime. It has a higher PSNR than restoration algorithms of inverse filter and wiener filter and strengthens the quality of image restoration.