异构融合网络下超采样图像细节增强仿真  

Simulation of Super Sampling Image Detail Enhancement in Heterogeneous Fusion Network

作　　者：邹小花[1] 夏容[1] ZOU Xiao-hua;XIA Rong(Science and Technology College,Nanchang Hangkong University,Nanchang Jiangxi 332020,China)

机构地区：[1]南昌航空大学科技学院,江西南昌332020

出　　处：《计算机仿真》2021年第11期499-503,共5页Computer Simulation

基　　金：基于特征建模的陶瓷产品三维设计与可视化国家基金(61562063)。

摘　　要：针对超采样图像缩放至较小程度分辨率降低,导致原始图片场景中的细节信息丢失等问题,提出了基于异构融合网络的超采样图像细节增强方法。在异构融合网络中,通过双三次插值法采集训练样本,建立一个奇异值算法模型稀疏表示超采样图像;通过差值算法得获取图像初始超分辨率,利用块匹配搜索匹配像素对应点,根据奇异值阈值算法找出匹配最佳的图像,对此做加窗处理后增强超采样图像细节。仿真结果表明,峰值信噪比较高,且结构相似度值接近于1,对图像有很好的细节增强效果。When the super sampling image is zoomed to a small extent,its resolution is reduced,resulting in the loss of detail information in original scene.Therefore,a method to enhance the details of super sampling image based on heterogeneous fusion network was proposed.In heterogeneous converged network,the training samples were collected by bicubic interpolation method,and then a model of singular value algorithm was built to sparse the super sampling image.Moreover,the difference algorithm was used to get the initial super-resolution of image,and the corresponding points of the matched pixels were searched by the block matching.Finally,the best matching image was found by the singular threshold algorithm,so that the details were enhanced after by adding the window.Simulation results show that the peak signal-to-noise ratio is high,and the value of structure similarity is close to 1,so the proposed method has good detail enhancement effect on the image.

关 键 词：异构融合网络 超采样图像 细节增强 差值算法 奇异值算法 超分辨率 

分 类 号：TP391.2[自动化与计算机技术—计算机应用技术]