Segmentation Algorithm for Unmanned Aerial Vehicle Imagery Based on Superpixel and Ultrametric Contour Map

An image segmentation algorithm based on superpixel and ultrametric contour map is proposed to accurately segment different images captured in the high-resolution unmanned aerial vehicle(UAV).Firstly,the image was segmented into superpixels by linear spectral clustering.Secondly,the dissimilarity between superpixels was calculated according to the histogram features in the Hue,Saturation and Value(HSV)color space.The ultrametric contour map which can represent the strength of the corresponding contour was obtained with the idea of hierarchical segmentation and then normalized.Finally,using the appropriate threshold value,the contours with lower weights than the threshold were deleted and the regions with higher similarity were merged to obtain the segmented result.Compared with segmentation algorithms including Iterative self-organizing data analysis algorithm(ISODATA),fuzzy c-means(FCM)and globalized probability of boundary,oriented watershed transform and ultrametric contour map(gPb-OWT-UCM),the experimental results show that the presented algorithm has higher accuracy and lower computational complexity and less dependence on the initialized values.