摘要
为了快速准确在立体图像中检测感兴趣区域,提出结合视觉感知特性的立体图像感兴趣区域检测方法。分析人眼视觉敏感度以及视觉暂留等多个重要视觉特性,通过时空域联合分析受损立体图像,改进视点合成,有效抑制合成失真,增强立体图像视觉质量。将眼动仪数据以及实验标记获取的立体图像数据作为人工神经网络的输入样本,训练更加符合人类视觉特性的先验模型,将全部特征融合处理形成显著图。在显著图的基础上定位感兴趣区域,实现立体图像感兴趣区域检测。经实验测试结果证明,所提方法可以有效提升检测结果的准确性,同时能够有效降低检测时间。
In order to detect regions of interest in stereo images quickly and accurately, a method of detecting the regions of interest in stereo images was put forward based on visual perception characteristics. Firstly, many important visual characteristics such as human visual sensitivity and visual persistence were analyzed. And then, damaged stereo images were analyzed with the spatiotemporal domain. Moreover, viewpoint synthesis was improved to effectively suppress synthesis distortion, and thus to enhance the visual quality of stereo images. Furthermore, the eye tracker and stereo image data obtained from the experimental markers were used as input samples of the artificial neural network to train a priori model that was more in line with human visual features, and then all features were fused to form a salient map. Finally, the regions of interest were located on the salient map, thus realizing the detection of the region of interest in stereo images. Experimental results show that the proposed method can effectively improve the accuracy of detection results while effectively reducing the detection time.
作者
卢佳佳
蔡坚勇
LU Jia-jia;CAI Jian-yong(School of Computing and Information Science,Fuzhou Institute of Technology,Fuzhou Fujian 350506,China;Fujian Normal University,College of Photonic and Electronic Engineering,Fuzhou Fujian 350117,China)
出处
《计算机仿真》
北大核心
2023年第1期234-238,共5页
Computer Simulation
基金
福建省教育厅中青年教师教育科研项目(科技类)(JAT200896)。
关键词
视觉感知特性
立体图像
感兴趣区域
显著图
人工神经网络
Visual perception property
Stereoscopic image
Regions of interest
Salient map
Artificial neural network
