一种基于显著性活动轮廓模型的图像分割方法

An Active Contour Models based on Saliency Maps for Image Segmentation

为了有效地提取图像中物体的轮廓,结合视觉注意机制,提出一种改进的距离正则化水平集活动轮廓模型的分析方法。首先提取图像的初级特征,构成图像显著图;然后采用最大类间方差法获得显著区域的初始轮廓,以此作为活动轮廓模型中曲线演化的初始位置;最后利用距离正则化水平集演化,获得目标物体的边界,完成图像分割。这种结合视觉注意机制与改进的距离正则化水平集演化方法能够显著降低水平函数演化次数,提高图像分割效率。仿真结果表明,它能有效检测单个及多目标物体的边界,且定位准确。

In order to effectively extract the outlines of image objects,a combination of visual attention mechanisms and distance regularized level set evolution was presented.Firstly,the primary characteristics of the image were extracted,which formed a saliency map.Secondly,the initial outline of the region was obtained using the Otsu’s method as initial position of curve evolution in the active contour model.Lastly,the object boundaries were acquired using distance regularized level to set evolution.The algorithm proposed reduce the iterations of curve evolution evidently,and improves the efficiency of image segmentation.The simulation experiment results on images of the different characteristics show that it can detect boundary of single object or edges of multi-objects and position accurately.