基于超像素和密度峰值聚类的图像分割

图像分割是计算机视觉领域的一个重要组成部分。密度峰值聚类已应用于图像分割领域。但由于密度峰值聚类在聚类时只能考虑数据的全局空间信息，不能有效去除图像噪声，因此提出了一种超像素的图像预处理方法。该方法能充分考虑局部空间信息，具有较强的鲁棒性。通过改进的形态梯度重建和分水岭算法得到具有精确轮廓且去噪效果较好的超像素图像。在此基础上，加入密度峰值聚类完成后续分割。通过在光学图像数据集BSDS 500上进行实验，验证了超像素算法及图像分割算法的有效性。

Image Segmentation Based on Superpixel and Density Peak Clustering

Image segmentation is an important part of computer vision field.Density peak clustering has been applied to image segmentation.However, density peak clustering can only consider the global spatial information of data in clustering, it cannot effectively remove image noise.Therefore, a superpixel image preprocessing method is proposed, which can fully consider local spatial information and has strong robustness.Firstly, the improved morphological gradient reconstruction and watershed algorithm are used to obtain the superpixel image with accurate contour and good denoising effect.On this basis, density peak clustering is added to complete the subsequent segmentation.Experiments on optical image dataset BSDS500 verify the effectiveness of superpixel algorithm and image segmentation algorithm.