室外光源光谱辐照度与K-means结合的单幅图像阴影检测

Outdoor Single Image Shadow Detection Based on Spectral and K-means

针对室外自然条件下单幅图像阴影检测困难,提出一种基于光谱辐照度的阴影检测算法。通过对户外光源的光谱特性进行分析,估算出阴影三色衰减模型(TAM)参数,由此得到阴影区域较暗的TAM图像。利用K-means方法将TAM图像分割为阴影区域和非阴影区域,结合中值滤波和形态学算子对阴影区域优化,成功提取出图像中的阴影部分。仿真表明该算法不需要复杂的特征学习过程,能够极大地提高运算速度;同时无需对图像校准以及获取任何先验知识,且可以用于相对复杂的真实场景中。

Single image shadow detection is difficult for outdoor natural conditions,concerning this issue a shadow detection algorithm is proposed based on spectral and K-means. The parameters of the shadow tricolor attenuation model( TAM) are estimated by analyzing the spectral characteristics of the outdoor light source,consequently,the darker TAM images are obtained. The shadow part of the image can be extracted successfully while K-means method is used to segment the TAM image into the shadow and non shadow regions combining with the optimal shadow area by the morphological operators. Simulation results show that this method does not need the complex learning process of feature operators and improves the speed of operation greatly. Meanwhile,the shadows for a single and uncalibrated image can be extracted without any prior knowledge,even though in complex scenes.