基于混合范式的评标记录影像重建方法  

Super-resolution Reconstruction Based on L1 and L2 Mixed Norm

作　　者：焦学瑞 张德州 李俊伟 于怀德 

机构地区：[1]浪潮通用软件有限公司

出　　处：《信息技术与标准化》2023年第11期44-49,共6页Information Technology & Standardization

摘　　要：为实现评标过程有据可查的目标,解决评标记录影像图片模糊、不清晰、分辨率低的问题,提出了一种基于L1与L2混合范式的超分辨率重建算法,该算法采用一种通用的代价函数来平衡L1范式与L2范式之间的拟合度,解决图像超分辨率重建过程中L1(一阶)范数法模型估计误差大和L2(二阶)范数法估计算子对图像灰度值异常点敏感、算法抗噪声能力差的缺陷,并结合图像的平均梯度来自适应确定影响函数的阈值。实验测试结果表明:此方法与采用单一的L1或L2范式估计算子相比,图像平均梯度分别平均提高了1.5倍和1.3倍,信息熵和对比度也得到了提高,图像视觉效果明显改善,有效解决阳光采购评标留档影像质量问题。For Sunshine Procurement,in order to achieve the documented goal of the evaluation process and solve the problems of blurred,unclear and low resolution images in the evaluation records.This paper presented a super-resolution algorithm based on L1 and L2 norm.The algorithm adapts a general local cost function to balance the fitting degree between L1 and L2 norm and determines the threshold of the influence function adaptively based on image average gradient.Solve the shortcomings like the L1 norm has bigger estimation error,the L2 norm is very sensitive to the outliers of the pixel,and poor noise immunity in the super-resolution regularization process.The experimental results show that the average gradient of the image reconstructed by the proposed algorithm is increased to 1.5 times and 1.3 times of that of the traditional super-resolution regularization algorithm based on single L1 norm or L2 norm averagely,the amounts of information and the contrast are also improved,the visual quality of the image improved significantly.Effectively solving the problem of image quality in the evaluation and filing of Sunshine Procurement.

关 键 词：超分辨率重建 L1、L2范式 正则化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]