基于视觉感知的图像放大  

Human Visual Perception for Image Zooming

作　　者：杜宏伟 乔美丽 宋刚 张云峰[1,2] 包芳勋 DU Hongwei;QIAO Meili;SONG Gang;ZHANG Yunfeng;BAO Fangxun(School of Computer Science and Technology,Shandong University of Finance and Economics,Jinan 250014,China;Shandong Provincial Key Laboratory of Digital Media Technology,Jinan 250014,China;Shandong Provincial Academy of Educational Recruitment and Examination,Jinan 250011,China;School of Mathematics,Shandong University,Jinan 250100,China)

机构地区：[1]山东财经大学计算机科学与技术学院,济南250014 [2]山东省数字媒体技术重点实验室,济南250014 [3]山东省教育招生考试院,济南250011 [4]山东大学数学学院,济南250100

出　　处：《计算机工程与应用》2020年第9期204-212,共9页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61672018,No.61772309);国家自然科学基金-浙江两化融合联合基金(No.U1609218):山东省重点研发计划(No.2016GSF120013,No.2017GGX10109,No.2018GGX101013);山东省高等学校优势科研人才团队培养计划;山东省自然科学杰出青年基金(No.ZR2018JL022)。

摘　　要：为了有效保持图像纹理细节和边界结构,提出了一种基于视觉感知的图像放大算法。构造了一类C2连续的双变量有理函数插值模型。根据等值线生成原理,利用等值线方法将图像自适应地划分为边缘区域和平滑区域。基于图像的区域特征,平滑区域采用多项式模型插值,边缘区域采用有理模型插值。根据人眼对比敏感特性,对边缘区域进行参数优化,从而获得高质量插值图像。与其他经典的图像放大方法相比,该方法能有效保持图像的纹理细节和边缘结构,同时获得了较好的客观评价数据。In order to preserve image texture details and edge sharpness, a human visual perception adaptive image interpolation algorithm based on the rational function is proposed. A type of bivariate rational interpolation functions with adjustable parameters is constructed. The image edge regions can be detected adaptively by using isoline method. Selecting different parameters in the rational model, the smooth regions and edge regions are interpolated by polynomial interpolation and rational interpolation respectively. The optimal rational interpolation parameters can be obtained based on the contrast sensitivity enhancement of the human eyes. The objective image with higher resolution ratio than the original image is obtained by adaptive interpolation. Compared with the classical interpolation algorithm, the proposed method is highly competitive not only in PSNR and SSIM but also in visual effect. Experimental results illustrate that the proposed method offers a remarkable quality improvement both objectively and subjectively.

关 键 词：图像放大 自适应图像插值 区域划分 等值线方法 视觉感知 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]