基于颜色属性的车辆阴影去除方法  

VEHICLE SHADOW REMOVAL METHOD BASED ON COLOUR ATTRIBUTE

作　　者：徐少飞[1] 刘政怡[1] 

机构地区：[1]安徽大学计算机科学与技术学院,安徽合肥230601

出　　处：《计算机应用与软件》2016年第9期203-207,共5页Computer Applications and Software

基　　金：安徽省科技攻关计划科技强警专项资金项目(1301b042020);高等学校博士学科点专项科研基金联合课题(20133401110009)

摘　　要：颜色属性,即语言颜色标签,可以表示世界中的所有颜色。视频帧图像中的阴影的颜色属性为黑色,因此提出一种基于颜色属性的车辆阴影去除方法。利用概率潜在语义模型学习颜色属性,建立颜色名概率字典,并实现对视频帧图像的颜色名映射,将表征阴影的黑色区域二值化为背景,而非阴影区域二值化为前景。同时,将二值化图与背景差分图进行"与"操作,去除阴影,再去除阴影外边缘的噪声。最后,对其进行先膨胀再连通域填充处理,以得到去除阴影后的车辆目标。实验证明,该方法在一定场景下可以很好地去除阴影,获得相对完整的运动车辆目标。Colour attributes,i, e. ,the language colour labels,can be used to represent all the colours in the world. The colour attribute of shadow in video frames is the black ,therefore we propose a colour attribute-based vehicle shadow removal method. We study the colour attrib- utes using probabilistic latent semantics model, establish colour names probability dictionary, and realise colour names mapping on video frames image. After binarisation operation, the black regions representing the shadows are set as the background, and the non-shaded regions are set as the foreground. At the same time, we process ＂AND＂ operation on the binarisation graph and background difference graph to remove shadows and then remove the noises on edges outside the shadows. Finally, we deal with the AND graph in operations of dilation first followed by connected domains filling to get the target vehicle without shadows. Experiments show that the method can remove shadows well under certain scenarios, and can get a relatively intact moving vehicle target.

关 键 词：颜色属性 概率潜在语义分析模型 阴影 颜色名映射 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]