基于NSPT域PCNN-SC的图像超分辨率重建

结合多尺度变换和脉冲耦合神经网络(PCNN)的 优点,将空域的稀疏编码(SC)算法通过非 下采样金字塔变换(NSPT)转换应用到频率域进行重建,并结合改进的PCNN对SC算法中的 最优系数 采用一种新的方法进行获取,进而提出一种新的超分辨率重建(SRR)模型与算法。首先将 图像进行三次B 样条 放大处理,然后采用NSPT对图像进行多尺度分解得到高低频子带系数。对于低频子带,运 用PCNN-SC 完成重建；对高频子带,将其与预测高分辨(HR)图像的特征图像运用改进的PCNN经图像融合 完成重建。最后 通过逆NSPT得到HR图像。实验表明,本文算法在主观视觉效果和客观数据都获 得了较好的效果。

Image super-resolution reconstruction method based on pulse coupled neural net work-sparse coding and nonsubsampled pyramid transform

A novel super-resolution reconstruction (SRR) algorithm is proposed based on nonsubsampled pyramid transform (NSPT) and pulse coupled neural network.Firstly,the image preprocessed by three-degree B-Spline interpolation is decomposed by NSPT to obtain its low frequency sub-band coeffi cients and high frequency sub-band coefficients.For the low frequency coefficients,a new reco nstruction method obtained by PCNN-SC is presented.For the high frequency coeff icients,the reconstruction is completed by fusioning them with the high resolu tion (HR) feature image through improved PCNN.Finally,the HR image is obtained by inverse NSPT The experimental results show that the proposed method outperforms other classical SRR algorithms in terms of b oth visual quality and objective evaluation.