基于显著特征分类的立体图像重定向方法  

Stereo Image Retargeting Based on Salient Feature Classification

作　　者：黄悦铭 唐振华[1,2] HUANG Yueming;TANG Zhenhua(School of Computer,Electronics and Information,Guangxi University,Nanning 530004,China;Guangxi Key Laboratory of Multimedia Communication and Network Technology,Nanning 530004,China)

机构地区：[1]广西大学计算机与电子信息学院,广西南宁530004 [2]广西多媒体通信与网络技术重点实验室,广西南宁530004

出　　处：《无线电工程》2024年第2期267-275,共9页Radio Engineering

基　　金：广西自然科学基金(2021GXNSFAA220058)。

摘　　要：现有立体图像重定向方法对不同特征图像均采用相同的策略进行重定向操作,导致一些立体重定向图像出现信息丢失、形变扭曲或深度改变的情况。影响立体图像重定向结果质量的因素主要包括显著区域形状和可视深度的改变等。为了解决这些问题,提出一种基于图像显著特征分类的立体图像重定向方法,将图像分为无显著及有显著2类图像,结合立体智能剪裁方法及立体非均匀映射方法对不同特征图像采用不同的重定向策略以减少信息丢失及几何失真。通过利用显著区域与非显著区域的深度信息差异可以更好地保持显著图像的深度感。实验结果表明,提出方法在主观对比及客观指标评价中均取得了优于其他算法的效果。Stereo image retargeting algorithms often adopt the same strategy to retarget various images with different features,resulting in some serious retargeting distortions such as information loss,geometric distortion,and depth change.The factors that affect the quality of stereo image retargeting results mainly include the change of salient region shape and visual depth.To solve these problems,a stereo image retargeting method based on image salient feature classification is proposed.This method divides the image into two categories:non-salient region and salient region images and combines the stereo cropping method and the stereo warping method to adopt different retargeting strategies for various feature images to reduce information loss and geometric distortion.The depth information difference between the salient region and non-salient region is used,so the perceptive depth of the salient image can be better maintained.Experimental results show that the performance of the proposed method is superior to other algorithms in subjective comparison and objective quality assessment.

关 键 词：立体图像重定向 显著特征分类 立体智能剪裁 立体非均匀映射 深度信息差异 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]