基于多进程的焦点堆栈图像融合方法

Method of Focus Stacks’ Fusion Based on Multi-Processing

摘要: 传统上为了拍摄前景与后景均清晰的图像，相机光圈应当尽可能的小。但是受到衍射极限的影响，过小的光圈会造成图像发生畸变。然而通过将多张相同构图但不同焦距的图像合成，可克服上述困难。本方法包含硬件设备和算法设计，重点在算法设计上，包括高斯滤波、拉普拉斯算子、均值滤波以及探测点的极大值映射，最终实现图像的合成。除此而外，算法中还加入了多进程的思想，使得算法在获取同样锐度图像的条件下，程序运行时间极大缩短，这通过仿真实验得以论证。本论文提出的算法适用于多核CPU的处理，具备普适性。

Abstract: Traditionally, the camera’s aperture should be as small as possible in order to make the foreground and the background clear, respectively. However, affected by the diffraction limit, tiny apertures can cause the image distortion. The above difficulties can be overcome by merging multiple images with the same compositions but with different focus. The thesis includes hardware and algorithm design, emphasizing on algorithm design. The image can be eventually fused including in Gaussian filtering, Laplacian operator, mean filtering and maximum mapping of detection points. Besides, in order to greatly shorten the running time of program, the multi-process is added to the algorithm without deprived of any sharpness, demonstrated by simulation. The algorithm is suitable for the multi-core CPU, universally.