以視覺顯著性為基礎之視訊品質評估

Abstract

視訊品質評估在視訊處理系統中扮演重要的角色。由於以人類來評估視訊品質過於昂貴及不便，所以人們希望設計出由電腦自動評估視訊品質的方法。PSNR是最著名的品質評估準則，因為它擁有低計算複雜度及明確的物理意義。然而，許多研究結果顯示PSNR未能如同人眼般的評估視覺品質。因此，有一些影像品質評估方法(UQI及SSIM)將人類視覺系統列入考量，並且可以被輕易地延伸來執行視訊品質的評估。然而，這些方法都沒有考慮到人眼通常只專注在視訊畫面中的重要部份而非整個畫面。因此，我們提出了一系列的步驟來延伸PSNR、UQI及SSIM以執行視訊品質評估，並利用上述的特性來提升它們的效能。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達成目標。值得一提的是我們提出的方法也可以被套用在其他現有的影像及視訊品質評估演算法。

Video quality assessment algorithms play important roles in video processing systems. It is desired to design automatic methods to evaluate the quality of videos by computer, since the evaluation by human beings is usually too expensive and not convenient. The most famous quality evaluation criterion is PSNR, which has low computational complexity and clearly physical meanings. However, many research results show that PSNR do not guarantee to always perform visual quality assessment similarly to human eyes. Therefore, there are also image quality assessment algorithms such as UQI and SSIM which take human visual system into account, and they can be easily extended to perform video quality assessment. However, these methods do not consider the fact that human beings only pay attention to important parts of video frames instead of the whole frames. Therefore, we proposed a series of procedures to extend PSNR, UQI and SSIM to perform video quality assessment with utilization of this characteristic to further improve their performance. Experimental results show that the proposed method has successfully achieved our goal in the LIVE video quality database with different video distortion types. It is worth mentioning that the proposed method can also be applied to other existing image/video quality assessment algorithms.