基于标记的多尺度分水岭视频目标分割算法

Video Object Segmentation Algorithm Based on Marker Multi-measure Watershed

针对视频目标提取的问题,提出了基于标记的多尺度分水岭视频目标分割算法。该算法以帧间变化检测为基础,通过改进的最小Tsallis交叉熵进行去噪、滤波,经形态学处理后得到运动目标初始二值掩模,并利用初始二值掩模得到用于分水岭算法的前景与背景标记,用该标记修正当前帧的多尺度形态学梯度图像,最后进行分水岭分割,得到具有精确边界的视频对象。实验结果表明,该算法能有效地分割和提取视频序列中的单个、多个以及快速运动的目标,继承了变化检测和分水岭算法速度快的优点,克服了分水岭容易产生过分割的缺点,具有较强的适用性。

Arming at the video object-extracted issue, video object segmentation algorithm is proposed based on marker multi-measure watershed. First, based on the frame difference detection, the initial motion object binary model is obtained by improved minimum Tsallis-cross entropy and morphologic managed. Second, markers of foreground and background for watershed are derived from temporal information, which is used to modify morphological gradient image. Third, watershed segmentation is used to acquire video object with precise boundary. Experimental results show that the proposed algorithm can segment and extract the single, multiple and fast motion object of the video sequence effectively. It has the virtue of low complexity inherited from change detection and watershed and overcomes the defect that it is easy to engender the excessive segmentations of the watershed algorithm, so it has a high adaptability.