Aiming at the problem that the traditional SLIC algorithm has poor quality in segmenting high resolution remote sensing images, this paper proposes an improved SLIC based on dimensionality reduction and region merging to segment the buildings. Firstly,it simplifies the dimensionality of the traditional SLIC, and the color information is replaced by the gray feature information to reduce the redundancy of the five-dimensional feature vector of the LAB color space. Secondly, the over-segmentation images are combined by using the region adjacency graph. Finally, the main parameters of the improved SLIC are analyzed and compared, namely, the number of super-pixels ‘k’, the compactness ‘m’ and the number of iterations ‘p’. The experiments show that this method can not only separate most of the building information, but also improve the operation efficiency and space efficiency. The running time efficiency is 25.5% higher than the traditional SLIC, and the segmentation precision of the building can achieve 97.6%.