在實作H.264/AVC視訊壓縮標準的硬體中，編解碼器常需頻繁的從視訊框記憶體存取參考視訊框資料，導致記憶體容量及頻寬限制了多媒體硬體的效能。視訊框記憶體壓縮(FMC)是一項利用視訊框資料內容的關聯性進行壓縮的技術，此技術旨在減少處理器與記憶體間所需傳輸的資料量，改善記憶體頻寬與容量的使用效率，降低硬體成本。本論文提出一個新的FMC技術，所設計的FMC具有下列特性：(1)特殊設計的像素壓縮順序與預測結構，達到更精確的像素值預測，(2)使用多個Golomb-Rice編碼機率分佈，根據編碼像素的位置，分別採用最佳的編碼機率分佈，(3)使用溢位切割、品質補償技術達到固定位元率，(4)使用群組區塊的概念達到壓縮區塊的位元率分享及控制。實驗結果顯示，所提之FMC技術對於越高畫質之視訊達到越顯著的壓縮效果，在低位元率的條件下，本技術也能達到無失真或近似無失真的壓縮效果。

We address the problem about implementation of a H.264/AVC codec that the performance of a video processor would be limited by memory bandwidth and frame memory size. In this article, a novel Frame Memory Compression (FMC) technique is proposed to reduce memory bandwidth by compressing the reference frame data to be stored in frame memory. The proposed FMC technique compresses a 4×4 block by a modified quantizer and a predictor followed by a Golomb-Rice encoder. The prediction order of pixels and structure are specially designed such that correlation among pixels in the block could be sufficiently exploited. The residuals are encoded using four coding distributions according to their positions in the block. A packing unit is applied on the variable-length codewords to guarantee fixed-length bit rate. This article also proposes a rate-control mechanism to effectively share bit rates among nearby blocks. The experimental results show that the proposed FMC technique achieve better performance on higher quality video sequences (e.g., 1080p or 720p). Furthermore, the FMC technique could be lossless or near-lossless in low bit rate conditions.