Image registration based on mutual information which shown to be effective is a computationally intensive process, and thus difficult to implement in real-time. A real-time hardware platform for visible and infrared image fusion is presented. The architecture takes field programmable gate array (FPGA) as its main processor, and integrates both kernel algorithm and peripheral management on its. Real-time performance can be achieved by optimizing hardware resource and algorithm, alone with memory access parallelization and processing unit execute pipelining. The analysis result shows hardware architecture can evaluate image fusion for 160x120 arrays image less than 36ms, and fusion image improves the target detection and recognition probability.