FILM.LrTL: FusIng MuLtiFocus IMages using Low-rank Transform Learning

This work introduces a novel technique for fusing the multi-focus images using Low-rank Transform Learning. Multi focus images are images captured from different sources. The idea is to fuse such images to make a single image which captures all the details. For such details, capturing both local and global structures is important which is taken care of by Low rank Transform Learning. The foremost step is to segment the source image into many patches using the sliding window technique. Then HoG features are obtained to categorize these mini images. On these mini-images, Transform Learning is applied to learn sub-transforms w.r.t. each class. Then, these sub-transforms are merged into a global transform in the low rank form to obtain LRR coefficient vector corresponding to each patch. Finally, the image fusion strategy is used to obtain the final image. The convergence guarantees are established on well-sounded techniques called as the alternate minimization in proximal operations. The set of experiments show that the proposed method outperforms many classical techniques quantitatively.