In this paper, the use of local and general features of spectral signatures separately for the fusion of hyperspectral and multispectral images is proposed. Nonnegative Matrix Factorization algorithm is used for fusion process. High resolution hyperspectral image is obtained by merging fusion images that produced for general and local spectral features. The proposed method is tested on sentetic and real hyperspectral images. Experimental studies show that carrying out fusion process individually for general and local features and then merging the fusion results improves the fusion performance.