Double-flow bokeh rendering method and system based on physical optical model and neural network

Abstract

The invention discloses a double-flow bokeh rendering method and system based on a physical optical model and a neural network, and belongs to the technical field of computer vision. The method comprises the following steps: converting a full-focus image into an RAW format, and obtaining a defocus image with a symbol through a disparity map, a fuzzy degree parameter and a focusing disparity parameter; enabling a physical renderer to perform multi-style rendering on the depth smooth area in the image according to the aperture shape parameters; enabling a neural network renderer to render a depth discontinuous region in the image, and predicting a probability graph of the region; fusing the results of the physical renderer and the neural network renderer; and restoring the fused bokeh image from the RAW format to the initial format. The method and the system not only can render a bokeh effect close to that of a single lens reflex camera, but also can freely adjust the focusing plane, the fuzzy degree and the rendering style, and are high in authenticity and high in customizability.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and more specifically, relates to a dual-stream bokeh rendering method and system based on a physical optics model and a neural network. Background technique [0002] In photography, bokeh refers to the loose, soft blurring of out-of-focus areas of a scene. Depending on the lens design and aperture shape, the bokeh effect will also change. In order to meet the needs of different users, photography manufacturers usually provide a variety of different styles of lenses. Although the use of SLR cameras with different large aperture lenses can create real, natural and varied bokeh effects, the purchase cost and the high demand for photography skills are not suitable for ordinary users who only use mobile phones to take pictures. Most mobile phone manufacturers try to use the post-processing method, combined with the scene depth information obtained by the sensor, to convert the all-focus ima...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a two-stream bokeh rendering method and system based on a physical optics model and a neural network, aiming to solve the problem that the bokeh effect obtained by the prior art is difficult to have both authenticity and Technical issues with customizability

Images