A multifocus image fusion method by using hidden Markov model

An image with all objects in focus is desired for human visual perception and computer vision tasks. The multifocus image fusion technique is a solution to fuse images acquired with different focus levels of a scene to obtain a clear fused image. Past approaches such as based on multiscale transformation (MST) are unavoidably lose information due to processing in transform domain and are sensitive to mis-registration. To solve these problems, we propose a hidden Markov model (HMM) based method, which not only uses patches rather than pixels as processing units but also handles each patch in the spatial domain instead of the transform domain. In the proposed method, firstly, the source images are divided into overlapping patches. Then, the fused image is modeled with HMM relying not only on the fidelity, clarity of patches but also on the relationship of adjacent patches. Subsequently, a belief propagation (BP) algorithm is adopted to select the fused patches from the source patches. Finally, these selected patches are integrated into the fused image. The experimental results prove that the proposed method achieves better performance, and is more robust to mis-registration than the MST-based methods.