摘要
针对传统单目视觉显著性模型存在细节丢失,不适用于复杂场景等问题,提出一种基于立体视觉分析的显著性区域检测算法。首先,采用基于图的分割方法将图像分割成不同区域,根据颜色和视差以及空间相干性计算颜色复杂度和视差复杂度。其次,对两者进行显著性聚合,计算像素对比度从而得到区域对比度。最后,引用视差信息计算局部对比度后,进行融合归一化,获得显著图。该算法适用于背景纹理复杂的立体图像显著性区域检测,检测的显著图细节突出,边缘锐利。实验结果表明,该算法优于其他显著性算法,符合人类视觉机制,在立体图像数据集上获得了75%正确率和88%召回率。
The traditional saliency models of monocular vision have some deficiencies , such as the loss of detail and low performance in complex situation .Saliency region detection method based on stereo vision analysis was pro-posed.Firstly, the image was divided into different regions by the graph-based segmentation method, and color complexity and disparity complexity were completed according to color , disparity and spatial coherence.Secondly, pixel contrast was calculated to get the regional contrast by saliency aggregation .Finally, the local contrast was cal-culated on the basis of disparity information and all methods were aggregated and normalized to get saliency map . The algorithm is applied for saliency region detection of stereo images which have complex texture background .The saliency maps have outstanding detail and sharp edge .The experimental result shows that the proposed method is superior to other algorithms significantly and is consistent with human visual perception and achieves 75%precision and 88% recall ratio on the stereo image data set.
出处
《电子测量与仪器学报》
CSCD
北大核心
2015年第3期399-407,共9页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金(61175033)
中央高校基本科研业务费专项(WK2100100009
2012HGCX0001)资助项目
关键词
颜色
视差
立体视觉
显著性聚合
区域检测
显著图
color
disparity
stereo vision
saliency aggregation
region detection
saliency map
