區塊轉換編碼影像與視訊之後級還原

Abstract

常因傳輸率或儲存容量不足以維持完整的影像品質，影像藉由捨棄某些較 不重要的資料被壓縮，因此影像品質變差。還原系統便是嘗試在解碼端降 低這種失真。區塊編碼是一基本的的影像壓縮技術，因為許多二維轉換邊 碼的實現已是相當的簡潔且有效率。二維離散餘弦轉換（2D-DCT）被普遍 的用在許多流行的影像標準如 JPEG、 MPEG 等。一種稱為誤差樣本補償 （EPC）法被提出，利用在轉換領域上的殘留相關程度來估計這種資料損 失（或稱量化誤差）。它聲稱能還原影像品質無論在主觀或客觀上，特別 是對於高度失真的影像。這篇論文主要目的在還原動態視訊的品質。我們 研究誤差樣本補償法並發展使之成為可用在動態影像。 所有的模擬以 JPEG 及 MPEG 測試。 動態影像裡只有固定區塊（I-blocks）被補償。當 固定影像（I-frames）品質被提昇時， 觀察 P 及 B 型影像（P- and B- frames）發現它們的品質不能被完全保證亦有正面提昇，這是因固定影像 的品質還原被用最小均方（MMSE）測量以及動態預估在重建影像上執行之 故。 Usually because the channel rateor storage capacity is not sufficient to preserve complete quality ofimages, the data are compressed by discarding some less important information. As a result, the image quality will be degraded.A restoration system attempts to reduce the degradation at the decoder end.Block- based coding is an essential technique in image compression due tothe properties of efficiency and simplicity of implementation of many 2-dimensional (2D) transform coding methods. 2D discrete cosine transform is the one widely adopted in many popular image codec standardsincluding JPEG and MPEG.An approach called {\em Error Pattern Compensation} (EPC) has been proposed to estimate the quantization error (information loss) in thetransform domain by taking advantage of the residual correlation. It claimsto be able to restore the quality both subjectively and objectively, especially in highly distorted images. The main goal of this thesis is to restore the quality of motion video. We study this EPC scheme and develop a way to apply it in motionvideo. Simulation is tested JPEG and MPEG.For motion images, only I- frames are restored. We observe how the promotion of I-pictures quality affects the performance of P- and B-pictures. In particular, the quality of P- and B-pictures can not be guaranteedas a result that the restoration of I-pictures is measured based on the MMSE criterion, and motion estimation (ME) is performed on reconstructed images.