基于视觉感知的平面设计背景图像裁剪  

Graphic Design Background Image Cropping Based on Visual Perception

作　　者：姚锦 程时伟[1,2,3] 刘征 YAO Jin;CHENG Shi-wei;LIU Zheng(School of Computer Science,Zhejiang University of Technology,Hangzhou 310023,China;Zhejiang Provincial Key Laboratory of Integration of Healthy Smart Kitchen System,Ningbo 315336,China;China Academy of Art,Hangzhou 310000,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023 [2]浙江省健康智慧厨房系统集成重点实验室,浙江宁波315336 [3]中国美术学院,杭州310000

出　　处：《小型微型计算机系统》2023年第3期565-572,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61772308)资助;浙江省健康智慧厨房系统集成重点实验室项目(2020F04)资助。

摘　　要：在平面设计工作中,为了解决前景图像与背景画布大小的不匹配问题,设计师通常需要对背景图像进行裁剪,但已有的裁剪方法没有考虑到用户对裁剪后视觉效果的主观体验.为此,本文提出了一种基于视觉感知的平面设计背景图像裁剪方法,首先基于全卷积神经网络训练平面设计数据集,建立视觉显著性预测模型,对图像进行视觉显著性预测;然后基于眼动跟踪技术,利用获得的眼动跟踪数据来识别图像的重要区域;最后将上述两步的结果进行融合,得到建议裁剪区域.实验结果表明,该方法的图像裁剪结果比已有方法更能吸引用户的视觉注意,具有更好的主观体验,且裁剪效果在平均重叠率和边界位移误差等指标上均有一定提升,验证了该方法在具体平面设计工作中的有效性与实用性.During the graphic design, in order to address the mismatching problem of the size of foreground image and background image, designers usually have to crop the background image.But the existing cropping methods neglect the user′s subjective feeling after viewing the cropping result.To solve this problem, we proposed a graphic design background image copping methods based on visual perception.First, we used fully convolutional network to train the graphic design data set, and built a visual saliency prediction model.Then, we used the eye tracking data to discriminate the important region of the image.And the results of these two steps were then combined to obtain the proposing cropping region.The experiment result showed that our proposed methods could attract more user′s visual attention, and provide better user experience, and the cropping effect has a certain improvement in indicators such as the average overlapped ratio and the boundary displacement error, which validated the efficiency and practicability of our method in graphic design work.

关 键 词：视觉感知 眼动跟踪 视觉显著性 图像裁剪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]