基于稀疏度自适应的视觉图像三维清晰重构  

Three-Dimensional Clear Reconstruction of Visual Images Based on Adaptive Sparsity

作　　者：石磊[1] 马丽茵 SHI Lei;MA Li-yin(North Minzu University,NingxiaYinchuan750021,China)

机构地区：[1]北方民族大学,宁夏银川750021

出　　处：《计算机仿真》2021年第3期139-142,共4页Computer Simulation

基　　金：北方民族大学设计艺术学院2018年度校级一般科研项目(2018XYYSY02)。

摘　　要：针对当前方法重构视觉图像时,存在峰值信噪比低、重构时间长和图像分辨率低的问题,提出基于稀疏度自适应的视觉图像三维清晰重构方法,利用图像光度信息和几何信息划分图像,按照纹理类别和边缘类别对图像进行分类,在图像组类别和噪声水平的基础上训练自适应字典,根据字典获得图像非局部相似先验和稀疏表示,结合建立变分模型对图像进行去噪处理。对去噪后的图像进行奇异值分解字典训练,利用稀疏度自适应正则化正交匹配算法对分解后的图像重建,完成视觉图像的三维清晰重构。仿真结果表明,所提方法的峰值信噪比高、重构时间短、图像分辨率高。Currently,the methods of visual image reconstruction have the disadvantages of low PSNR,long reconstruction time and low image resolution.Therefore,this paper puts forward a method of three-dimensional reconstruction of visual image based on sparsity adaptation.Based on the photometric and geometric information,the image was divided.The images were classified by texture and edge categories.The adaptive dictionary was trained based on image group category and noise level.According to the dictionary,the image non-local similarity prior and sparse representation were obtained.The image was denoised by combining with the foundation of variational model.The denoised image was trained by singular value decomposition dictionary.The decomposed image was reconstructed via using the sparse adaptive regularization orthogonal matching algorithm,thus the three-dimensional clear reconstruction of the visual image was completed.Simulation results show that the method has high peak signal to noise ratio and resolution,and short reconstruction time.

关 键 词：稀疏度 图像重构 图像去噪 学习字典 追踪匹配 

分 类 号：TN911.7[电子电信—通信与信息系统]