符合感官知覺之壓縮失真平衡與模式選擇演算法

Abstract

為了減少資料的傳輸量或是資料的儲存空間，使用者透過編碼器(Encoder)盡可能將資料壓縮到最小。為了保持影片品質的統一，編碼器利用相同的QP值來壓縮影片資料。但是對於觀察者來說，均方根差(PSNR)的判別結果不一定與觀察者所觀察的結果一樣，因為相同的錯誤發生在圖片的不同地方，觀察者觀察的結果也不盡然的相同。因此，為了達到維持相同的感知上的品質，我們提供一個perceptual distortion equalization的方法調整每塊macro-block的QP值，藉由維持相同的感知上的品質，達到改善壓縮效益的目的。

In traditional video coding, maintaining constant quality is essential for reconstructed videos. To achieve this goal, video coding usually involves using constant quantization step size to encode video data. However, in human perception, each location in a frame may not have the same feeling for observers; depending on the distribution of reconstruction errors. Therefore, the same error existing in different locations might create a different feeling for observers. To achieve the same perceived video quality experienced by observers for each location, we propose using perceptual distortion equalization (PDE) to decrease the difference between original video and processed video, based on human perception of each location. PDE involves using a scaling factor, which is derived analytically, to adjust the quantization step size of each macro-block to approach this matter. Simulation results show that the variation of perceptual distortion is decreased. The improved video quality of the proposed method is also confirmed by performing objective quality evaluations.