双参数Beta过程联合字典遥感图像超分辨

为了改善遥感图像超分辨重建(super-resolution reconstruction,SRR)效果,针对以往仅适用于单特征空间的稀疏字典超分辨算法,提出同时适用于两个特征空间的双参数Beta过程联合字典(Beta process joint dictionary,BPJD)遥感图像SRR方法。首先,根据遥感图像退化模型生成训练样本图像,并分别对高、低分辨率图像进行分块和Gibbs采样,生成字典训练样本。然后,依据BPJD,建立连接高、低分辨率遥感图像空间的双参数联合稀疏字典,将字典稀疏系数分解为系数权值和字典原子的乘积,依据字典原子指标训练和更新字典,得到高低分辨率联合字典映射矩阵。最后,进行遥感图像超分辨稀疏重构。实验结果表明:所提方法可自适应地缩小字典尺寸,能以更小尺寸的稀疏字典重建更高质量的超分辨遥感图像,重建结果图像的纹理细节信息更丰富,峰值信噪比和结构相似性度均有提高。

Remote sensing image super-resolution based on double parameters Beta process joint dictionary

An image super-resolution reconstruction (SRR) algorithm based on the double parameters Beta process joint dictionary (BPJD) is proposed to improve the image SRR effect. Compared with those sparse representation SRR algorithms that are only applicable to single feature space, the BPJD learning dictionary is obtained in coupled feature spaces. Firstly, a training sample image is obtained according to the remote sensing image degradation model, and high and low resolution images are respectively patched and Gibbs sampled to generate the dictionary training samples. Then, a double-parameter joint dictionary is established to connect the high-low resolution image space according to the BPJD model. In the Beta process, sparse coefficients are expressed as the multiplication of coefficient weights and dictionary atoms and the joint dictionary mapping matrix is obtained by training and updating. Finally, remote sensing image super-resolution reconstruction is performed in sparse representation. Experiment results show that the proposed method can reduce the size of the dictionary adaptively, and can reconstruct the higher quality super-resolution image with the smaller sparse dictionary. The remote sensing image SRR result contains more texture details, and peak signal-to-noise ratio and structural SIMilarity are increased objectively.