局部遮挡下的目标识别算法

Target identification algorithm under partial occlusion

规格化互相关算法是用得较普遍的目标识别方法,但是当目标区域被局部遮挡时,该方法通常不能正确定位目标。提出了一种新的基于选择互相关系数的目标识别算法用于搜索有局部遮挡的目标区域。算法分两步进行:用增量互相关算法计算出模板图和场景图的增量图像,比较二者增量图像的一致性,计算出选择互相关系数矩阵;结合选择互相关系数矩阵,用规格化互相关算法在场景图中搜索目标区域。当场景图存在较严重的噪声时,可对选择互相关系数矩阵进行修正以克服噪声的影响。实验结果表明,基于选择互相关系数的目标识别算法对局部遮挡和高亮度变化情况有较强的鲁棒性。

The algorithm of normalized cross correlation is a universal method for target identification. But when the object is partially occluded, this method usually cannot correctly locate the target. A new algorithm based on selective cross correlation coefficients is proposed in order to search images under partial occlusion. The algorithm includes two steps： First using increment cross correlation algorithm to calculate increment images of template and scene images, comparing consistency of two increment images and getting matrix of selective cross correlation coefficient. Then using normalized cross correlation algorithm combined the matrix of selective cross correlation coefficients to search target in scene image. When noise is serious, selective cross correlation coefficients should be improved. The experimental results indicate that the algorithm of selective cross correlation is robust for partial occlusion and highlight.