摘要
复杂场景中的目标定位是目标检测和识别的重要过程,为了更好地对复杂场景中的目标进行定位,基于视觉的概率模型,提出了一种目标定位的新方法。区别于一般的区域分割和边缘检测方法,该方法首先通过建立平滑、纹理、阴影和杂乱等4种不同类型区域特性的概率模型,对场景中的前景和背景进行了概率分析;然后结合不同的尺度大小,标记出图像中显著度较高的目标区域;最后经过边缘轮廓的概率建模和连通性分析来提取完整目标区域。实验结果表明,该方法具有较好的鲁棒性和通用性,不仅符合人的视觉注意特性,而且具有一定的抗背景干扰能力。
Object localization in complex settings is the main process in vision task of detection and recognition. This paper presents a new method for object localization based on visual statistical probabilistic models. It is different from traditional ways of region segmentation and edge detection. First, the method can label the regions with high saliency through regional probabilistic models which were developed using the flat, texture, shading and clutter properties with different scales and then the whole object regions can be extracted by edge probabilistic models and connectivity discussion. Experiments show the approach has strong robustness and generality, because the results are in accord with visual attention properties and the ＂background＂ noise is lower.
出处
《中国图象图形学报》
CSCD
北大核心
2007年第7期1234-1242,共9页
Journal of Image and Graphics
基金
国家自然科学基金项目(60375011
60575028)
安徽省优秀青年科技基金项目(04042044)
"新世纪优秀人才支持计划资助"项目(NCET-04-0560)
关键词
视觉概率模型
区域分割
边缘检测
尺度
visual probabilistic models, region segmentation, edge detection, scale
