基于奇异值分解的图像超分辨率重建

在图像获取过程中,由于环境的变化以及成像传感器自身硬件条件的限制,获得的图像往往含有噪声并且分辨率较低.为了提高图像分辨率来满足实际应用的需求,充分考虑了自然图像中图像块的自相似性,并采用奇异值分解方法对图像块的相似性进行度量,将其作为权重与非局部均值方法相结合,实现了单幅图像的超分辨率重建.为了减少计算量,在计算图像块的相似度之前,先对图像块的均值进行统计并引入一个阈值,对均值绝对差小于阈值的图像块进行相似度估计.为了对算法有效性进行验证,文中采用基于误差敏感性的峰值信噪比来对重建图像的质量进行度量.仿真结果表明算法在提高图像分辨率的同时有效的抑制了噪声,并且很好的保持了图像的细节信息.

Singular value decomposition based image super resolution reconstruction

In the process of image acquirement,as the changing of environment and the constrain of sensors,acquired images always have noise and low resolution.In order to improve the resolution of degraded images for reality application,we fully considered the self-similarity of image patches,and adopt a singular value decomposition based method to evaluate the similarity of patches,and take it as the weight combined with the non-local means to reconstruct the high resolution images.For reducing the complexity of computation,we give a statistic of the mean value of patches and bring in a threshold,only the patches which have absolute difference less than the threshold are chosen to compute the weight.To verify the effectiveness of the proposed method,the error sensitive method peak signal-noise-ratio is adopted to evaluate the quality of the restored image.Simulation results show that our method can improve the resolution well,restrain the noise,and keep the details.