基於視覺注意力的視訊品質評估方法

Abstract

視訊品質評估在視訊處理系統中扮演重要的角色。由於以受測者來評估視訊品質過於昂貴及不便，所以人們希望能設計出由電腦自動評估視訊品質的方法。考慮到現有的影像品質評估方法大多沒有符合人眼通常只專注在視訊畫面中的重要部份而非整個畫面的特性，本篇論文利用影像中的顯著度和影片中的移動強度來更精準地抓出人眼視覺較專注的區域並加以處理。視訊品質的分數由每一幅幀算出來的分數分別依照時間順序套用不同的權重結合而成。最後我們使用包含多種失真視訊的LIVE視訊品質資料庫來驗證我們的方法，而實驗數據顯示出我們已成功地改善現有的影像品質評估方法來套用在視訊品質評估上，並且擁有計算複雜度低且運行時間短等優點。

Video quality assessment algorithms play an important roles in video processing systems. Since the evaluation by human is much expensive and inconvenient, it is necessary to design an automatic methods that evaluate the quality of videos by computer. Since the existing image/video quality assessment is not conform to the fact that human beings pay more attention to the important parts of video frames than whole image, we proposed a video quality assessment with utilization of saliency and motion strength in each frame to accurately detect the region that attract human attention. After evaluating with LIVE video quality database which includes different video distortion types, the experimental results show that the proposed algorithm can successfully improve the performance of image quality metrics and video quality assessment at low computation complexity.