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

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。

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 converted 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 subband,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 respectively 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 experimental results show that the proposed method has more ideal preprocessing effect as well as lower the misclassification rate and the classification error rate.