應用馬可夫鏈於瞳位凝視移轉分析

Application of Markov Chain to Fixation Transition Analysis

觀看是動態的過程，多數研究只探討瞳位追蹤之最終結果，而忽略中間的轉移。馬可夫鏈為狀態間的移轉過程，本研究將馬可夫鏈的概念運用在瞳位追蹤的分析上。研究主要分析影響凝視間移動之因素，預測凝視點在感興趣區域（AOI）間之移轉機率，再應用馬可夫鏈建立凝視移轉模型。研究選取80張景觀照片，依山景、水景、平坦景觀及森林景觀2×2排 列組合於畫面中，共有20組試驗。瞳位追蹤實驗請受測者自由觀看每組試驗10秒，並記錄凝視停留位置。研究分析並探討色彩、複雜度、偏好、位 置等因子對凝視移轉機率之影響。結果顯示，位置為影響凝視移轉之主要 因素，且跳視多為水平移動。在本研究之凝視移轉機率模型上，總預測誤差僅7．09%，顯示模型有極高的預測能力。

Eye movement is a dynamic process. The majority of eye tracking studies have focused on the accumulation of eye-tracking metrics in a series of eye movements to examine the eye movement processes within. The Markov chain is a process of the transition of states. The current study applied the Markov chain on the analysis of eye tracking. This study aimed to analyze the factors influencing fixation transitions to predict fixation transition probabilities between areas of interests (AOIs), further establishing a fixation transition model. We selected 80 landscape photographs and arranged them into a 2 by 2 display according to four landscape categories (mountain, aquatic, open, and forest), resulting in a total of 20 trials. In the eye tracking experiment, participants were told to observe freely for 10 seconds in each trial, during which the equipment recorded fixation positions. The effects of color attributes, complexity, preference, and position on fixation transition probabilities were examined. Results indicated that position was the main factor influencing transition probability, and saccadic direction was mainly horizontal. On our transition fixation model, the total prediction error was 7.09%, which indicated that our model was successful.