Real-time Human Vision Simulation And Rendering

Vision-realistic rendering is expressing human vision's characteristics in computer graphics.The most significant effect to simulate human vision is blur.Human eye enables us to focus on different objects by accommodation,which changes optical power to maintain a clear image.Besides,objects lie in background or peripheral area are hard for us to distinguish color and shape.They are blurred due to Depth-of-Field(DoF)or peripheral vision.When focus point changes,persistence of vision causes motion blur.These characteristics are not express in ordinary computer rendering.In most real-time applications,people use a simple thin-lens model to simulate the principle of imaging of human eye.Result of thin-lens model shows DoF effect,but objects in peripheral vision are still sharp and clear.In this paper,we proposed a method based on blur distribution function that is used to render human vision-realistic image in real-time.The method ensures real-time efficiency as well as focusing on the rendering effect of human vision by computing pixel-wise blurriness in GPU.It is helpful in medical and vision simulation field.It enables users to view what human sees in reality more precisely.The main research and achievements of this thesis include:(1)Proposing a Blur Distribution Function(BDF)to describe the distribution of intersection points on retina by a cone of rays refracted in eye model.The BDF is fitted using a large number of sample points so that the Circle-of-Confusion(CoC)size on retina can be estimated efficiently.(2)Proposing a blur distribution simulation method based on forward ray tracing.Simulating refraction process of light passing human eye optical model by tracing rays through Navarro schematic eye model under certain parameters(focus distance,pupil size,spherical coordinate).Intersection points of a cone of light ray and retina are used to calculate CoC size.(3)Using a neural network to train and simulate BDF in real-time.It can express the special blur distribution of human eye and is easy to be added to common frameworks.The extra burden of it meets the need of real-time application.(4)Presenting a method that updates CoC size weighted by previous value to simulate the persistence of vision effect.In real-time application,motion of focus point causes the delayed change of blur distribution.The delay time of each pixel is related to its brightness.Implementing motion blur effect of human vision.(5)Implementing an interactive application that implements real-time DoF computation and post-processing blurring in GPU shader.Scene size,focus point,pupil size are all user defined.It enables user to simulate human vision effect in real-time.Comparing with thin-lens method under the same condition,our method shows more realistic depth-of-field blur,peripheral area blur and motion blur.