摘要： 两步自适应字典学习的超分辨率算法易受插值图像影响而导致图像模糊。针对该问题,提出一种改进交叉分辨率自适应字典学习算法。根据自然图像的冗余性,即不同分辨率图像依然有相似的图像块,直接以低分辨率图像作为字典学习对象。为了弥补单帧图像作为字典学习的不足,采用镜像图像进行字典学习,以产生容量更大的字典。输入低分辨率图像,利用新的字典通过稀疏表示获得高分辨率图像,采用峰值信噪比(PSNR)和结构相似性度量(SSIM)评估重建效果。实验结果表明,与立方插值、SUSR、MSS、HLSR算法相比,提出算法的重建图像纹理保留得最好,图像效果更加丰富自然,且算法运行速度较快,在多数情况下具有最高的PSNR值和SSIM值。

关键词: 超分辨率, 自适应字典学习, 镜像图像, 峰值信噪比, 结构相似性度量

Abstract: As the issue of the sensitivity to interpolated image in the SuperResolution(SR) algorithm of two-step adaptive dictionary learning,causing blurred images,an improved algorithm is proposed,named intersected-resolution adaptive dictionary learning algorithm.According to the redundancy of natural images,that is images of different resolution still have similar patches,the low-resolution image are used as the learning object of the dictionary directly.In order to solve the insufficiency of using a single image for dictionary learning,mirror images are used to produce a bigger dictionary.A low-resolution image is input and a high-resolution image is obtained by the sparse representation with the new dictionary.The Peak Signal to Noise Ratio(PSNR) and Structural Similarity (SSIM) measure are used to evaluate the effect of reconstruction.Experimental results show that,compared with cubic interpolation,SUSR,MSS,HLSR algorithms,the reconstructed image’s texture of the proposed algorithm can be reserved better.The image effect is more abundant and natural.And the running speed is faster.It has the highest PSNR value and SSIM value in most cases.

Key words: Super Resolution(SR), adaptive dictionary learning, mirror image, Peak Signal to Noise Ratio(PSNR), Structural Similarity(SSIM) measure

中图分类号: 

• TP391