Research And Implementation Of Image-based Depth Of Field Rendering Algorithm

Depth of field is an important feature in photography. The effect of depth of field can be used to selectively emphasize some parts of an image, and attract the viewer’s attention to the focused region. Depth of field can also be used to create the illusion of depth, make the scene a three-dimensional appearance. It can also be used to create a special artistic conception and effect.The image base post processing depth of field rendering methods contains many artifacts. Isotropic filtering based methods incorrectly blend the colors from different depth, causing color leakage artifact. Methods that do not consider the different properties of the background blur and the foreground blur that deal different regions of the image with same filter will result in depth discontinuous artifact. Most post processing methods do not consider and simulate partial occlusion. So it also causes artifact.The proposed algorithm first classify the pixels of the image into three categories:background region, foreground region and in-focus region. Then it uses different anisotropic filtering methods to deal with different situations. It used a scatter based filter to avoid the color leakage artifact in the background region. It used a gather based filter to simulate partial occlusion in the foreground region. And in the in-focus region, the pixels near the foreground region are blended with some color of the foreground, in order to avoid the blur discontinuity artifact. Further this thesis makes an improvement so that the bokeh effect can also be simulated in the proposed algorithm. The implementation is carried in GPU by using the CUDA parallel computing architecture. And a user friendly real time interactive image-based depth of field rendering system is built based on the proposed algorithm.Compared the experimental results of the proposed image-based depth of field rendering algorithm with the results of other state-of-the-art methods, we can conclude that the algorithm proposed in this thesis can effectively avoids the common artifacts exist in the post processing depth of field rendering method, including color leakage artifact, blurdiscontinuity artifact and the artifact caused by partial occlusion and achieves a realistic depth of field effect.