圖學導向材質感知模型與感知最佳化材質處理演算法的開發

Abstract

計算機圖學的主要目標之一是產生具現實感知的圖像。然而，感知或許是人體中最複 雜和最神奇的功能。我們需要在視覺感知上有更好的瞭解，才能有效調整圖學演算法 來滿足使用者在視覺感知上的要求。因此，將感知評估機制納入計算機圖學演算法中， 可對於現實化的成像，有很大的助益。這個計劃的其中一個目標，即是在探討視覺感 知模型和計算機圖學演算法可行的整合機制。 材質是真實世界中非常常見的視覺現象。因此，材質長期以來一直是視覺感知和計算 機圖學領域中熱門的研究主題。然而，現有的材質感知模型，大多數是探討材質偵測 或分割，和針對人工材質發展出來的，它們並不適合使用在圖學的應用上。我們確實 需要的是一個由真實世界中的自然材質所建立出材質感知模型。最近在材質合成和操 作演算法的成功，我們已能控制自然材質，這使得自然材質刺激可以被使用在視覺感 知研究實驗中。此外，由於在神經科學和心理生理學上的進步，我們現在能測量和分 析人腦中神經系統的活動。這對於我們瞭解人腦中的感知和認知過程有很大地幫助。 由於以上這些領域的快速發展，開發適合計算機圖學應用的材質感知模型的時機，已 然成熟。 在本計劃中，我們提出一個圖學感知互作用的閉迴圈架構。它可以交互式地改進視覺 感知模型和計算機圖學演算法。根據這個架構，我們將結合心理物理學的方法、心理 生理學的測量(腦電圖和眼睛運動跟蹤)和計算機圖學演算法，開發一個針對圖學導向 的材質感知模型和符合感知最佳化的材質處理演算法。本計劃是第一個嘗試建立針對 圖學導向的材質感知模型的研究，和第一個展示出真實世界的材質刺激是可能受控制 和使用在人類視覺研究中的計劃。本計劃如果成功的話，將對計算機圖學和人類視覺 研究領域產生重要的影響。

One of the primary goals of computer graphics is to render perceptually realistic images; however, perception is perhaps the most complicated and mysterious function of human beings. We need better understandings on visual perception so that we can orient a graphics algorithm to fulfill one』s perceptual demands. It would be of great advantage to achieve realistic rendering by integrating a perceptual evaluation mechanism into a computer graphics algorithm. An objective of this project is to explore a feasible integration of visual perception models and computer graphics algorithms. Texture has long been drawing research interests from both the visual perception and computer graphics communities since it is such a common visual phenomenon in the real world. Though there exists texture perception models, most of them are developed for texture detection/segregation and for artificial patterns. We indeed need a texture perception model built on real-world, complex textures. Recent success in texture synthesis and manipulation algorithms makes it possible to control texture stimuli as desired for visual perception study. Furthermore, because of the advance in neuroscience and psychophysiology, we are now able to measure and analyze the neural activities in human brains. This greatly helps us to understand the perception and cognition process in the human brains. Thus, it is possible to develop a texture perception model suitable for computer graphics applications. In this project, we proposed a graphics-perception-interaction close-loop framework to interactively improve visual perception models and computer graphics algorithms. Based on this framework, we will combine psychophysical methods, psychophysical measurements (electroencephalogram and eye movement tracking), and computer graphics algorithms to develop a graphics-oriented texture perception model and perception-optimized texture processing algorithms. This project, if succeeds, will have great impacts to the fields of computer graphics and human vision study as it is the first attempt to construct a graphics-oriented texture perception model and to demonstrate that real-world texture stimuli can be controlled and used in the human vision study.