Efficient Multi-focus Image Fusion Using Parameter Adaptive Pulse Coupled Neural Network Based Consistency Verification

Multi-focus image fusion technique is an important approach to generate a composite image with all objects in focus. Accurate focused pixel detection from multiple source images is crucial for multi-focus image fusion. However, false detection of focused pixels is inevitably due to the low-level image features being usually used to achieve focus pixel classification in most fusion methods. Consistency verification operation is frequently used to revise the falsely detected focused pixels in many fusion schemes. However, most consistency verification strategies cannot achieve the desired results. In this paper, we modify the parameter adaptive pulse coupled neural network (PA-PCNN) by introducing a new strategy to measure the linking strength of neurons. Thus, the PA-PCNN can greatly improve the accuracy of identifying focused pixels. The proposed method contains four steps. Firstly, the residual between an image and its filtered version by efficient mean filter is used to calculate the sharpness of a source image. Then, a new consistency verification method based on adaptive pulse coupled neural network (PA-PCNN) is adopted to improve the accuracy of the initial sharpness. Next, the focus detection maps are constructed by comparing the refined sharpness of two source images. Finally, the fused image is constructed according to the focus detection map. Experimental results show that the proposed method has achieved comparable or even better results compared with the state-of-the-art approaches in both visual quality and objective evaluation.