Abstract: The edge of infrared video fire images in large space is fuzzy, which takes difficulty to the image accurate segmentation, a segmentation method based on background subtraction and C-V model is researched. It uses the background subtraction to get the motion image; it obtains the complete motion area through the morphological processing, and gets its circum-rectangle; initial curve evolution of C-V model is the circum-rectangle to do the segmentation, the closed and complete motion target contour is obtained. This method can avoid the whole image segmentation and reduce the amount of computation. Compared the simulation experiment with other threshold segmentation algorithms, the algorithm is proved to be accurate and effective, it is conducive to the next step of fire feature extraction and recognition.

Key words: image segmentation, background subtraction, circum-rectangle, C-V model

摘要： 针对大空间中红外视频火灾图像边缘模糊，不易准确分割问题，研究了一种基于背景差分和C-V模型的分割方法。通过背景差分得到运动图像；利用形态学处理得到完整的运动区域，并获得其外接矩形；以外接矩形作为C-V模型的初始轮廓曲线进行分割，得到封闭、完整的运动目标轮廓。该算法避免了对整幅图像分割，减少了运算量。通过实验仿真并与阈值分割算法比较，证明了该算法的准确性和有效性，有利于下一步火灾特征提取与识别。

关键词: 图像分割, 背景差分, 外接矩形, C-V模型