Endmember extraction is the key step in the mixed pixel decomposition of hyperspectral remote-sensing images. Traditional endmember extraction algorithms ignore the spatial correlation and nonlinear structure of hyperspectral images, which restricts their accuracy. To consider the spatial relationship and nonlinear structure of hyperspectral images, a nonlinear endmember extraction algorithm based on homogeneous region segmentation is proposed. A hyperspectral image was divided into several homogeneous regions using a superpixel segmentation method, and the manifold learning method was used to ensure the nonlinear structure of the hyperspectral images, extracting preferred endmembers within homogeneous regions. Simulation data and real hyperspectral image experiments showed that the algorithm proposed herein can guarantee the nonlinear structure of hyperspectral data, and the endmember extraction results were better than those of other traditional linear endmember extraction methods. Even in the case of a low signal-to-noise ratio, effective endmember extraction results were obtained.