摘要
在编辑合成图像时,前景区域与背景区域由于分属不同来源,外观通常存在较大差异,不能呈现真实的视觉效果.图像协调化主要消除合成图像的不真实感,通过将背景域的外观信息迁移到前景域中,使人眼尽可能少地感受差异.现有方法在将背景信息进行转移时,能触及到的背景范围有限,信息提取不充分使得图像协调效果不佳.同时,前景细节丢失会影响图像协调的整体效果.基于以上问题,提出内卷特征提取和前景细节增强模块,进行图像协调.在iHarmony4数据集上试验结果表明,该方法能生成细节更充分、真实感更强的图像效果.
Abstract
When editing composite images,it can be challenging to achieve a realistic visual effect because of the sig-nificant differences in appearance between the foreground and background areas.Image harmonization aims to address this by transferring the appearance information of the background domain to the foreground domain,minimizing the perception difference of the human eye.However,existing methods often have limitations in their ability to extract information from the background and can result in poor harmonization effects.This paper proposes an inner roll feature extraction and fore-ground detail enhancement module for image harmonization.Experimental results on the iHarmony4 dataset demonstrate this method can generate images with sufficient details and stronger realism.
维普
万方数据