Image fusion has been an import research field in image processing. It can be applied to many tasks, such as surveillance, photography, medical diagnosis, etc. As a significant part of image fusion, many scholars have done a lot of research on multi-focus image fusion in this field in recent years. Multi-focus image fusion usually uses fusion rules to fuse two or more images with the same scene information in different focusing situations. Meaningful images are called full-focus images because full-focus images have more information and are more suitable for human visual perception system. In this paper, we present a novel method for multi-focus image fusion based on Laplacian eigenmaps dimension reduction in non-subsampled contour transform (NSCT) Domain. Firstly, we decompose the source images by using NSCT, and then its high frequency and low frequency subbands in different directions can be obtained. Futhermore, we design different fusion strategies for high frequency subbands by using the laplacian eigenmaps (LE) based on sliding window to generate the weight map. For low frequency subbands, we employ gray level co-occurrence matrix techniques to fusing them. In the end, the inverse NSCT transformation is used to get the final fusion result. Experimental results show that this method is satisfactory when compared with other popular fusion algorithms both subjectively and objectively. After processing, the fused image has clear edges, good visual effect and sharpness.