Research on UAV Multi-scale Superpixel Segmentation Algorithm

The traditional superpixel segmentation algorithm mainly uses K -means clustering method to generate superpixels. By selecting the initial seed point, the five-dimensional vector formed by three-dimensional color information and two-dimensional spatial coordinates is represented as a pixel point. The distance between pixels is calculated near the proximity area of the seed point and clustering is carried out until the cluster center no longer changes, and the final size is similar. Compact arrangement of superpixels. For different types of natural images. If superpixel segmentation is performed on a single scale, due to the limited reflection of the structural information of the image, the subsequent fusion processing of the algorithm will be affected, and the segmentation accuracy of the image will also be affected. In this paper, a multi-scale image segmentation algorithm is proposed, which can combine the precision of fine scale with the integrity of coarse scale, so that it can be better applied to bottom-up image processing algorithms. The improved superpixel presegmentation algorithm is an improvement on the traditional single-scale superpixel presegmentation algorithm, and improves the efficiency and accuracy of image segmentation.