利用边界运动显著性的红外运动目标分割方法

针对传统算法无法对短暂静止的红外运动目标进行准确有效地分割，该文提出了一种利用边界运动特征的红外运动目标分割方法。首先，定义了一种新指标边界运动显著性，该指标利用边界点时空域特性，可以准确反映图像中边界点的运动特征，显著性越高，则该边界点属于运动目标的可能性越大。然后，通过Otsu阈值法提取出显著性高的边界点，并利用历史数据对其进行修正，修正之后的运动边界点作为运动目标种子。最后，通过一种逐层生长的区域生长方法，在运动目标种子上分割出完整的运动目标掩膜。该方法在多组红外图像序列中进行测试与对比，结果证明该方法运动目标分割效果良好，目标背景的错分率低，可以准确检测并分割出短暂静止的运动目标。

A Method for Segmentation of Moving Object in Infrared Videos Based on Motion Saliency of Edge

A novel method is proposed for segmentation of moving object using motion feature of edge in infrared videos. At first, a new index Motion Saliency of Edge (MSE) is defined. MSE can reflect the motion feature of edge points of an image based on the spatial-temporal characteristic. The higher the MSE of an edge point is, the more likely it belongs to a moving object. The edge point with a high MSE value is extracted by using Otsus thresholding. The results obtained by Otsu are updated by using historical data. The edge points which are extracted and updated can be considered to be the seed of moving objects. At last, the segmented masks of moving objects are grown from the seeds by using a region growing method of layer-by-layer. The proposed method is successfully tested over three infrared image sequences and compared with the other two methods. The experiment results demonstrate that the proposed method has better performance of moving object segmentation with less effect of object-background misclassification in infrared videos.