新一代的視訊壓縮標準HEVC（High Efficiency Video Coding）畫面內預測之角度模式提升至33個不同的方向，使得畫面內預測更加精確，進而提升壓縮效率。本文提及半色調技術包含有序抖色法與錯誤擴散法，我們運用此技術於HEVC畫面內預測，可改善複雜紋理區塊之預測畫面。33種角度模式在加入半色調技術後，畫面內預測種類激增為原始預測角度種類的兩倍，此作法雖可提升編碼效率，但相對的需要增加處理時間。 為了減少編碼時間，將針對預測單元及編碼單元兩部分分別運用快速模式決策演算法與支持向量機（Support vector machine, SVM)將成本較高的預測模式與部分深度省略以減少找尋最佳模式與深度的複雜度。在畫面內預測此步驟，將RDO（Rate-Distortion Optimization）程序中各候選模式的SATD成本參數與該預測單元最小SATD成本參數比較，若大於門檻值，即刪除該候選模式，藉此減少候選模式個數以節省時間；在編碼單元上，根據支持向量機的預測結果，分類編碼單元為複雜或是平滑紋理，以判斷當前的編碼單元要提前略過或提早終止，藉此加速編碼單元深度的判定，以達節省時間的效果。 ;The new video compression standard is the High Efficiency Video Coding (HEVC). In order to predict more accurately, the intra prediction has increased 33 different angle modes to promote the efficiency of compression. Half-toning techniques mentioned in this article including order dithering and error diffusion. We utilize those techniques in intra prediction dealing with complex textures. After adopting Half-toning techniques with 33 intra angle modes, it augments the prediction modes tremendously to two times of the original angle modes. Though the methods achieve better performance, the time consumption increases. For the purpose to diminish time consumption, we aim at prediction units and coding units by employing fast-mode decision algorism and support vector machine (SVM) separately to skip modes with high candidate costs and some depths for reducing the complexities. In the procedure of Rate Distortion Optimization (RDO) of intra prediction, we will compare the SATD costs of all candidate modes with the smallest one. If the ratio is higher than the threshold, then we delete the modes to reduce the number of candidates for time saving. When dealing with coding units, we skip the current depth or terminate deeper depths according to the results of SVM. By the results of SVM, we can know the texture is complex or smooth to accelerate choosing depth of coding units for the expectation of time reducing.
