摘要
目前,深度学习已经在图像超分辨率重建上表现出不错的性能,但是对某些纹理细节还原度不高。针对此问题,提出了基于卷积神经网络的纹理的超分辨率重建算法。首先用梯度算子提取图像的纹理特征,再将图像按照纹理进行分类,最后用卷积神经网络对同一类别的样本集进行超分辨率重建。实验证明,该算法能够恢复一定的纹理信息,而且对同类纹理的重建结果优于已有算法。
At present,deep learning performs well in super resolution reconstruction. However,some texture information is not reconstructed well enough. To solve this problem,we proposed a texture super-resolution reconstruction algorithm based on convolution neural network（CNN）. We firstly extract texture features using gradient operator and classify the images according to their texture features. And then,for the same category of image samples,a super-resolution reconstruction model is built based on CNN. The experiments show that this algorithm can restore some texture information very well,and the results are better than other algorithms.
出处
《微型机与应用》
2017年第20期57-60,共4页
Microcomputer & Its Applications
关键词
图像处理
卷积神经网络
超分辨率重建
纹理特征
image processing
convolutional neural network(CNN)
super-resolution reconstruction
texture feature
作者简介
蒋雪（1993-）,女,硕士,主要研究方向：深度学习、图像超分辨率.
韩芳（1981-）,通信作者,女,博士,副教授,主要研究方向：神经动力学、智能算法、深度学习等.E-mail：yadiahan@163.com.
