Image fusion based on non-negative matrix factorization and infrared feature extraction

A new effective fusion method based on non-negative matrix factorization (NMF) and infrared target extraction is presented for infrared and visible images fusion. The two source images are taken as original data in NMF analysis, from which the feature base containing the global features of the source images can be extracted. The feature base image is replaced by the adjusted visible image, which is histogram matched with the raw feature base. As non-negative matrix factorization image fusion algorithm lacks details, we introduce a local gradient as an active measure, and combine weighting and selection methods to fuse the visible image with the feature base image. Then the target regions from infrared image are segmented through edge detection, region growing and morphological processing methods. The segmented target regions are fused with the background regions of the feature base image. Experiment indicates that the proposed method is simple in calculation. And it can retain texture details of visible image, highlight the thermal target of infrared image and enhance the readability of source images.