Photoacoustic microscopy suffers from limited depth-of-field due to the strong focus of laser beam, which implies that only some part of the imaging result is clear. Such shortage limitsthe further application of this powerful imaging technique, which is the problem we hope to address. In this work, we consider to solve the target problem through information fusion method. By fusing multi-focus 3D photoacoustic images, a large volumetric and high resolution photoacoustic microscopy can be obtained. However, the task of fusing photoacoustic signal is different from general 2-D multi-focus image fusion problem. The core challenge for our work is the fusion of different 3D photoacoustic volumes. We simplify the task as 2D problem, which is achieved by slicing the 3D data to 2D and reconstruction of the 2D slices. We propose a 3D fusion method which involves proper data preprocessing, slicing of the 3D data, fusion of slices and 3D reconstruction. Experiment results verify that the fused data shows larger depth of field and contains more useful information, which supports our thought of extending depth-of-field through information fusion method. To further demonstrate the superior of deep-learning (DL) method, a few non-DL 2D algorithms are selected for a comparison study based on objective assessments to show the generalization ability of convolutional neural network based image fusion algorithm.
|