由于煤矿井下特殊的环境导致井下图像一般比较模糊和边缘特征比较弱,而目前基于字典学习的图像超分辨率重建算法通常是将全部图像块利用训练的单一字典进行重建,忽视了各图像块之间的差异性,不利于重建边缘不清晰的矿井图像。结合矿井图像特征提出一种边缘融合的多字典超分辨率图像重建方法,该算法根据各图像块的梯度统计信息将图像块进行分类并训练对应的字典库,重建时将不同字典重建的图像块融合成完整的高分辨率图像;此外为了提高图像的边缘信息,预处理阶段低分辨率图像进行边缘融合以增强边缘特征,重建的高分辨率图像利用学习的先验知识进行边缘融合以修正重建过程中出现的误差。实验表明,该算法的重建效果优于其它基于字典学习的超分辨率图像重建方法,能够很好地重建图像的边缘细节,并抑制重建过程中产生的重影和振铃效应,平均PSNR值提高1.19 d B。

Due to the special environment of coal mine,the images are generally blurred and have weak edges,and the image super-resolution reconstruction algorithm based on dictionary learning is usually that all blocks are reconstructed by single dictionary,ignoring the differences among them and being bad for reconstructing the mine images. Combined with the characteristics of mine images,this paper proposes a multi-dictionary learning image super-resolution method with edge fusion. In the method,all image blocks are classified according to the gradient statistics and dictionary librar- ies are trained. Finally,the image blocks reconstructed by different dictionaries are merged into a complete high resolu- tion image. In order to perfect the edge information,the preprocessing stage of the low-resolution image performs edge fusion to enhance their features. The high-resolution image reconstructed by dictionary learning uses the prior knowl- edge to fuse the edge information,thus correct the errors in the reconstruction process. The experiment shows that the effect of this method is improved compared with other super-resolution image reconstruction methods based on dictiona- ry learning,which can reconstruct the edge details of the image well and suppress the ghosting and ringing effect,and the value of average PSNR is increased by 1. 19 dB.