基于超像素随机游走的自然场景图像分割方法

Natural Scene Image Segmentation Method Based on Super-pixel Random Walk

随着成像技术的发展,人们使用普通成像设备采集的图像分辨率越来越高,细节越来越清晰,能更加准确地呈现真实自然场景中事物之间的关系。然而传统的图像处理方法在处理分辨率相对较高的自然场景图像时,效果和效率并不理想。针对现有的随机游走图像分割算法在处理背景复杂、分辨率大的自然场景图像时,目标边界难以贴合以及效率较低的问题,提出了一种基于超像素随机游走的自然场景图像分割方法。将超像素的思想引入到随机游走过程当中,先对图像进行超像素分割处理,然后以超像素为节点,对每个超像素区域提取颜色特征及LBP纹理特征构建无向加权图,图中的节点数量大幅度降低,最后进行随机游走实现超像素的分类并得到图像的分割结果。实验证明,由于超像素分割速度快以及对复杂纹理图像的边界描述准确等优点,基于超像素随机游走的图像分割算法对于自然场景下颜色、纹理信息复杂的图像和大分辨率图像在一定程度上有效地提升了分割效果及分割效率。

With the development of imaging technology,people use ordinary imaging equipment to collect images with higher resolution and clearer details,which can more accurately present the relationship between things in real natural scenes.However,the effect and efficiency of traditional image processing methods are not ideal when dealing with natural scene images with relatively high resolution.Aiming at the problem that the existing random walk image segmentation algorithm is difficult to fit the target boundary and has low efficiency when dealing with natural scene images with complex background and high resolution,a natural scene image segmentation method based on super-pixel random walk is proposed.The idea of super-pixel is introduced into the random walk process.Firstly,the image is segmented by super-pixel.Then,with super-pixel as the node,the color feature and LBP texture feature of each super-pixel region are extracted to construct an undirected weighted graph.The number of nodes in the graph is greatly reduced.Finally,the random walk is performed to realize the classification of super-pixel and the segmentation results of the image are obtained.Experiment shows that due to the advantages of fast super-pixel segmentation speed and accurate boundary description of complex texture images,the image segmentation algorithm based on super-pixel random walk effectively improves the segmentation effect and segmentation efficiency for images with complex color and texture information and large resolution images in natural scenes.