Research On Multi-focus Image Fusion

Restricted by the depth-of-field of the real lens used in optical imaging system,once the focal distance is set, objects at a certain distance from the lens can be focused,whereas objects out of the certain distances will be defocused. As a result, the recordedimage will not be full focused in case any object presents outside the range ofdepth-of-field. In practice, in order to get a full focus image, we can firstly obtain aseries of multi-focus images that focus on each object, and then employ the imagefusion techniques to fuse these multi-focus images to get the full focus image.In this paper, we fully research on the multi-focus image fusion techniques. Weelaborated the imaging principle of multi-focus images, given a comparative analysesabout kinds of multi-focus image fusion method, introduced the quality measuremethods and functions for fused images. Aimed at the problems existed in multi-focusimage fusion algorithms, we explored in depth and given corresponding solutions.①Aiming at the block-based fusion methods for multi-focus images, such keyproblems like finding a suitable block size and overcoming the blocking-artifact oftenoccurs. To address them, we present an intelligent fusion method using particle swarmoptimization (PSO) algorithm, which is guided by a structural similarity (SSIM)criterion. In this method, the performances of some kinds of focus measure arecompared and the ones who sensitive to defocus blur are employed as the focus measurein fusion algorithm. PSO is used to fast search for an appropriate block size, while theSSIM is employed as the fitness function in PSO to evaluate the blocking-artifact, thusenabling this method to have a significant improvement in fusion performance.Experimental results show that the proposed method performs better than some classicalmethods, such as a block-based method similar with ours in fusion rule andmultiresolution-based method. Our method is able to find the suitable block sizeadaptively; furthermore, it preserves more details than others in fused image.②For multi-focus image fusion, the focused regions are regarded as rectangle inblock-based algorithms. Nevertheless, the focused regions are irregular-shaped ingeneral. For further improved the performance of the fused images, we propose a newmultifocus image fusion algorithm using focused region detection. Unlike thefocus-region-based fusion methods common used, we utilize a converse way to get thefocused regions. The specific procedures of the proposed algorithm are as follows: firstly, using the Q-shift DT-CWT to gain an initial fusion image; secondly, comparingthe differences between the source images and fused image to get a label map, and thenremoving the defects in label map to obtain the region map; finally, extracting thefocused regions in source images to generate the final fusion image. Experimentalresults show that the presented algorithm can effectively detect the irregular-shapedfocus regions, and the fusion result performs well.