一种融合图像区域属性特征的运动阴影消除算法

针对传统C1C2C3彩色不变性特征模型中分母只考虑值最大的颜色信息导致阴影消除结果不准确及局部二值模式(local binary pattern,LBP)易受噪声扰动的缺陷,在C1C2C3模型中采用两个颜色通道的均值作为分母以及在LBP模型中加入抗噪声因子的方法分别对以上两个模型进行改进,提高了两个模型的阴影检测准确率.为了有效融合各种基于属性阴影消除方法的优势,引入LBP纹理复杂度测量函数,根据LBP纹理复杂度分区域融合文中改进的两个模型,实验结果表明,改进的算法提高了运动阴影消除准确率.

Moving cast shadow removing algorithm based on blocked texture and color feature fusion

For the traditional shadow model of C1C2C3 ,only one color channel, which is the largest channel value in the denominator , was selected to count for moving cast shadow removing. Thus, the result is not accurate. And the shadow model of LBP was apt to be confused by noises. In order to solve the problem effectively, we improved the C1C2C3 model by using the average of the two channel value in the denominator and added a divisor for resisting noises in LBP. Furthermore, we brought in the LBP texture measure to calculate the coefficient of texture complexity in the neighbor field of center pixel. Then whether the improved C1C2C3 model was been adopted to replace the improved LBP model or not was determined by the coefficient of texture complexity to enhance the accuracy of the result. Experimental results show that the improved shadow removing algorithm performs better.