Reasearch And Implementation Of Future-frame-based Temporal Anti-aliasing

With the efficiency advantage of temporal anti-aliasing,this algorithm has been one of the most widely used real-time anti-aliasing algorithms in recent years.To achieve real-time antialiasing,this algorithm assigns the sampling points to the multiple history frames and reuses historical data.When the sampling information in the time domain is sufficient and historical data is usable,this algorithm can achieve an effect similar to supersampling anti-aliasing.However,such a condition is not fully met in practical application.In the situation that historical data can't be used,problems such as geometric aliasing and ghosting will occur.Besides,for reducing the cost of graphics memory,this algorithm replaces multiple history frames with the accumulation frame,which will lead to error accumulation and image blurring.Based on the research of temporal anti-aliasing,this thesis analyzes the specific causes of geometric edge jagging,ghosting,image blurring,and subpixel detail missing.To solve the problems mentioned above,this thesis proposes the future-frame-based temporal anti-aliasing.The basic idea of this algorithm is that: in addition to the existing information in the time domain,this algorithm takes the next aliasing future frame into account,using samples in such frame to solve the problems of temporal anti-aliasing.In the process of implementing this algorithm,the following difficulties will be encountered.Firstly,in order to reduce the adverse influence of the future frame on ghosting,this algorithm should consider how to extract reusable shaded data from the future frame.Secondly,for the regions with geometric edge jagging and subpixel detail missing,using one more frame data does not always meet the demand of fully sampling geometric information in the time domain,which makes it difficult to solve the problem of geometric aliasing via the future frame without anti-aliasing.Thirdly,under the influence of the aliasing future frame,how to realize ghosting suppression and deblurring while maintaining the anti-aliasing effect is the core problem of the algorithm.In order to solve the difficulties mentioned above,this thesis proposes the following innovations:· For the problem of reusable future data extraction,this thesis puts forward two solutions,including G-Buffer-and-resampling-based reusable future data extraction method and sampling-based reusable future data reconstruction method.The former takes the motion vector and depth information provided by G-Buffer into the scope of future sample visibility evaluation and then extracts reusable future data by resampling the visible sample.The latter uses the sampling mechanism to obtain reusable sampling data and employs the sampling data fusion mechanism to solve the problem of future data reconstruction so that the purpose of extraction can be achieved.Moreover,the extraction effect of the latter is better than the former.· For the problem of how to use a future frame to improve the geometric anti-aliasing effect,this thesis proposes a future data reuse mechanism.The reuse mechanism includes a future data fusion scheme and an improved neighborhood clipping method based on the future frame.According to the aliasing of current shaded data,the fusion scheme changes the fusion weight of future data to improve the geometric edge anti-aliasing effect.In fact,the fusion scheme can also improve the situation of subpixel detail missing,which is related to the increase in the success probability of subpixel detail sampling.The improved neighborhood clipping method is only used to reduce the probability of subpixel detail missing.The basic idea of the improved method is that: the future frame is introduced to increase the accuracy of historical data limitation and reduce the error frequency of neighborhood clipping so that the goal of improvement will be achieved.· For the problems of ghosting and image blurring,this thesis proposes a future-framebased fitting scheme.Under the premise that the future data can be reused,this scheme will use the result processed by the future data reuse mechanism instead of the shaded result of the target pixel at the current moment to maintain the effect of anti-aliasing.To solve the problem of ghosting,the fitting scheme will use the ghosting suppression method to adjust the fusion coefficient of historical data and then reduce the degree of historical data being misused to achieve anti-ghosting.To solve the problem of image blurring,the fitting scheme will use the error control method to realize the dynamic adjustment of historical data reuse weight and then control the error to realize deblurring.Experimental results indicate that the G-Buffer-and-resampling-based reusable future data extraction method and the sampling-based reusable future data reconstruction method can solve the problem of reusable future data extraction,and the extraction effect of the latter is better.The proposed future data reuse mechanism can effectively deal with the problem of geometric edge jagging and improve the subpixel detail missing.The future-frame-based fitting scheme helps the algorithm proposed in this thesis to achieve better ghosting suppression and deblurring.Compared with the temporal anti-aliasing,the proposed algorithm can achieve a better effect.