多先验约束正则化模型图像复原方法

Image restoration method based on multi-prior constraint regularization model

图像在成像过程中经常会产生模糊与噪声等多种复杂的降质问题,针对同时存在模糊与噪声的图像复原问题,提出一种多先验约束正则化模型图像复原方法。图像组稀疏表示能够很好的去除图像模糊,图像自相似性的非局部均值理论能够很好的抑制图像噪声,因此引入组稀疏约束项与非局部均值自相似约束项,构建一种新的图像复原模型,根据图像组稀疏与非局部均值自相似性先验知识求解清晰图像。实验选择不同自然模糊影像、真实遥感影像以及水下模糊影像,并对比了组稀疏算法、盲去卷积算法、非局部均值算法与比值稀疏约束算法。针对含有不同类型模糊与噪声影像,本算法实验结果PSNR值平均提高2.17 dB、2.08 dB、2.14 dB,充分说明本算法在去除图像模糊的同时抑制了图像噪声,达到提高图像质量的目的。

A variety of complex degradation problems such as blur and noise are often generated in the image imaging process.In order to solve the problem of image restoration with both blur and noise,a multi-prior constraint regularization model image restoration method was proposed.Image group sparse representation can effectively remove the image blur,good image self-similarity of nonlocal average theory can well restrain image noise.Therefore,a set of sparse constraint items and nonlocal average self-similar constraints terms are introduced to construct a new image restoration model according to the image group sparse and nonlocal average self-similarity of image groups.Different natural blur images,real remote sensing images and underwater blur images were selected in the experiment,and the group sparse algorithm,blind deconvolution algorithm,non-local mean algorithm and ratio sparse constraint algorithm were compared.The experimental results show that PSNR values of the proposed algorithm increase by 2.17 dB,2.08 dB and 2.14dB on average for images with different types of blur and noise,which fully demonstrates that the proposed algorithm can suppress image noise while removing image blur and achieve the purpose of improving image quality.