摘要
衣着颜色是行人最显著的表观特征,在视频监控场景中极易受到光照变化的影响.为此,笔者提出了一种基于多尺度光照估计和层次化分类的衣着颜色识别方法.首先,提出一种多尺度局部反射统计的光照估计模型,通过该模型实现对偏色图像的光照矫正;其次,为了精确地识别衣着颜色,设计基于融合多颜色空间特征的层次化分类器;最后,在校园监控场景采集4 998张行人衣着图像(晚上2 052张,白天2 946张)进行对比实验.实验结果表明,该方法能有效提高监控视频中衣着颜色识别准确率且至少提高12.5%.
Clothing color of pedestrian, as the most salient appearance features, was easily contaminated by illuminance variation in video surveillance. To this end, this paper proposed a clothing color recognition method based on multi-scale illumination estimation and hierarchical categorization. Firstly, this paper presented a multi-scale illumination estimation model, which employed local reflection statistics, to correct colors of the color-biased image. Secondly, a hierarchical classifier based on multiple color space features was designed to accurately recognize clothing colors. Finally, extensive experiments on the newly created dataset, including 4 998 image samples (2 052 for evening, 2 946 for daytime) recorded in campus, suggested that the proposed method outperforms other methods at least 12.5 % on accuracy.
出处
《安徽大学学报(自然科学版)》
CAS
北大核心
2016年第6期24-30,共7页
Journal of Anhui University(Natural Science Edition)
基金
国家863计划资助项目(2014AA015104)
国家自然科学基金资助项目(61502006
61502003)
关键词
衣着颜色
光照估计
层次化分类
视频监控
偏色图像
clothing color
multi-scale illuminance estimation
hierarchical categorization
video surveillance
color-biased image