一种基于自适应超像素的改进谱聚类图像分割方法

An Improved Spectral Clustering Image Segmentation Method Based on Adaptive Superpixel

图像分割是一种基于图像中各个像素的相似性,把图像分成多个区域,以提取出所需目标的技术。当前图像分割研究过程中,由于图像分割需处理的数据量较大,导致传统的基于聚类算法的图像分割算法难以处理大规模的图像分割问题。针对该问题,本文根据对传统谱聚类算法进行改进研究,提出一种新的基于自适应超像素的改进谱聚类图像分割方法。首先通过运用贝叶斯自适应超像素算法,作为谱聚类算法的预处理,从而达到降低算法运算的复杂度;然后将超像素作为传统NJW算法的输入,利用NJW算法迭代运算的优点,达到提升算法准确率的目的。通过在BSDS500数据集上验证算法有效性,本文所提出算法在不同类型的图像上较传统的聚类算法分割准确率提高近10%,实验结果表明了本算法的有效性。

Image segmentation is based on the similarity of each pixel in the image,which divides the image into several regions to extract the re⁃quired objects.In the process of image segmentation research,due to the large amount of data to be processed in image segmentation,the traditional image segmentation algorithm based on spectral clustering algorithm is difficult to face the problem of large-scale image segmen⁃tation.To solve this problem,a new algorithm is proposed which is based on the improvement of traditional spectral clustering algorithm combines the Bayesian adaptive superpixel segmentation.Firstly,an adaptive superpixel algorithm is used as a preprocessing part of the spectral clustering algorithm to reduce the complexity.Secondly,the superpixels are using as input to the traditional NJW spectral cluster⁃ing algorithm,then improves the accuracy by advantages of the iterative operation.To verify the effectiveness of our proposed method,we use the BSDS500 as experimental data set.Compared with the traditional clustering algorithm,the segmentation accuracy of the proposed algorithm in different types of images is nearly 10%higher.The experimental results show the effectiveness of the proposed algorithm.