基于显著性区域检测和水平集的图像快速分割算法

Image Fast Segmentation Algorithm Based on Saliency Region Detection and Level Set

为了实现含有复杂背景和弱边界图像的快速准确分割,传统的水平集常采用重新初始化的方法,但是这种方法存在计算量大、分割不准确等问题。因此,结合显著性区域,该文提出一种基于边缘信息与区域局部信息结合的变水平集图像快速分割方法。首先用元胞自动机模型检测出图像的显著性区域,得到图像的初始化边界曲线。然后,采用改进的距离正规化水平集演化(Distance Regularized Level Set Evolution,DRLSE)模型把图像的局部信息结合到变分能量方程中,用改进的能量方程去指导曲线的演化。实验结果表明,与DRLSE模型相比,提出的算法平均消耗的时间只需要前者的2.76%,且具有较高的分割准确性。

In order to achieve fast and accurate segmentation of images with complicated background and weak boundaries, the re-initialization method is often adopted in the traditional level set function. However, this method has many problems such as large computation and inaccurate segmentation. Thus, combined with the saliency detection algorithm, a new image segmentation method of variable level set based on the combination of edge information and regional local information is proposed. Firstly, the saliency region of the image is detected by the cellular automata model to obtain initial boundary curve of the image. Then, an improved distance normalized level set evolution（Distance Regularized Level Set Evolution, DRLSE） model is used to combine the local information of the image into the variational energy equation, and the evolution of the curve is guided by the improved energy equation. Compared with the DRLSE, the experimental results show that the average time of the proposed algorithm only needs 2.76% of the former with further improvements in the accuracy of image segmentation.