摘要
针对现有的基于卷积神经网络的图像超分辨率算法参数较多、计算量较大、训练时间较长、图像纹理模糊等问题,结合现有的图像分类网络模型和视觉识别算法对其提出了改进。在原有的三层卷积神经网络中,调整卷积核大小,减少参数;加入池化层,降低维度,减少计算复杂度;提高学习率和输入子块的尺寸,减少训练消耗的时间;扩大图像训练库,使训练库提供的特征更加广泛和全面。实验结果表明,改进算法生成的网络模型取得了更佳的超分辨率结果,主观视觉效果和客观评价指标明显改善,图像清晰度和边缘锐度明显提高。
An improved image super-resolution algorithm based on the convolutional neural network is proposed to overcome many problems such as more parameters,a large amount of calculation,longer training time and fuzzy texture combined with the present image classification network model and visual recognition algorithms.The proposed algorithm adjusts the convolution kernel size to reduce parameters in the original three layers of convolutional neural network.Pool layers are added to reduce the dimension and decrease the computational complexity.The learning rate and size of input sub-blocks are improved to reduce the training time.The training database is expanded to provide extensive and comprehensive characteristics.Experimental results show that the proposed algorithm achieves good super-resolution results,and the subjective visual effect and objective evaluation indices are both improved obviously.The image resolution and edge sharpness are enhanced significantly.
出处
《光学学报》
EI
CAS
CSCD
北大核心
2017年第3期96-104,共9页
Acta Optica Sinica
基金
国家自然科学基金(61471272)
湖北省自然科学基金(2016CFB499)
关键词
显微
图像超分辨率
深度学习
卷积神经网络
卷积核参数
microscopy
image super-resolution
deep learning
convolutional neural network
parameters of convolution kernels
