视觉注意模型的低照度图像感兴趣区域检测
Region of Interest Detection in Low Light Images Using Visual Attention Models
唐菀 1刘鑫2
作者信息
- 1. 电子科技大学成都学院,四川 成都 611731
- 2. 中南大学计算机学院,湖南 长沙 410083
- 折叠
摘要
针对低照度图像质量较低导致的检测困难问题,提出一种基于视觉注意模型的低照度图像感兴趣区域检测方法.将图像由RGB色彩空间转换至HSV色彩空间,通过NSST获取图像的多个高通子带和一个低通子带;在高通子带中利用自适应阈值去噪法去噪V分量,在低通子带中采用多尺度Retinex增强V分量,再修正增强后图像S分量,将处理后图像转换回RGB色彩空间;依据视觉注意模型分别获取图像亮度显著值、色彩显著值和方向显著值,联合构建图像像素点特征向量,采用过渡滑动窗贝叶斯方法实现图像感兴趣区域检测.实验结果表明,所提方法的预处理效果更理想、错分率和误分率更低.
Abstract
Aiming at the difficulty of detection caused by low-light images,this paper put forward a method for detecting the regions of interest in low-light image based on visual attention model.First of all,the image was conver-ted from RGB color space to HSV color space,and then multiple high-pass sub bands and a low-pass sub-band were obtained by NSST.Moreover,adaptive threshold denoising method was adopted to denoise the V component in the high-pass sub-bands.Meanwhile,multi-scale Retinex was used to enhance the V component in the low-pass sub-band,and then correct the S component in the enhanced image.After that,the image was converted back to RGB color space.Furthermore,saliency value of brightness,color saliency value and directional saliency value were obtained re-spectively according to the visual attention model.Finally,feature vectors of image pixel were constructed jointly,and then the transition-sliding window Bayesian method was adopted to detect the region of interest in the image.The ex-perimental results show that the proposed method has more ideal preprocessing effect as well as lower the misclassifi-cation rate and the classification error rate.
关键词
视觉注意模型/低照度图像/感兴趣区域检测/过渡滑动窗贝叶斯Key words
Visual attention model/Low-light image/Detection for region of interest/Transition-sliding window Bayesian引用本文复制引用
出版年
2024
NETL
NSTL
万方数据