基于高斯模糊核渲染离焦图像的研究与实现

efocus image refers to the image with defocus blur, which can be applied to depth estimation, hyperspectral reconstruction and other computer vision tasks. Defocus image can be rendered by all-in-focus image, i.e., convolution of the all-in-focus image with a point spread function. In this paper, by studying the rendering method of Gaussian blur kernel as point spread function,point rendering and layer renderingare implemented, and then the differences between the real defocus image and the defocus images obtained by these two rendering methods are compared. The proposed point and layer rendering methods using Gaussian blur kernel as point spread function are highly innovative and valuable, among which the layer rendering method is better than the point rendering method in terms of accuracy and running speed, and can be applied to simulate the generation of defocus images and provide defocus image datasets for downstream computer vision tasks, which has strong practicality.