H.264的主要貢獻之一是，它採用了在空間域的畫面內(Intra)預測方式，從而顯著地提高了I畫面壓縮比。針對每一個4×4的區塊，Intra 4×4模式提供了九種方向性的預測模式。然而，對於每一個4×4區塊，在編碼器需要1或4個位元用來標明Intra所使用的模式以順利解碼。因此，每一個4×4的Marcoblock(MB)共需要使用16到64個位元。為減少這方面的負擔，金大延(Dae-Yeon Kim)於2008年提出了『Adaptive Intra Mode Bit Skip (AIMBS)』演算法。假如鄰近區塊中所有的參考像素值非常相近，則可以使用DC模式做為預測模式，而用來標示模式的位元則利用跳位方式來省略。但是此法只針對DC模式跳位，其演算法仍有改進的空間。故我們提出另一種方法，利用鄰近參考像素與待編碼區塊往往有相同方向性之特性，增加垂直(Vertical)與水平(Horizontal)模式的跳位。實驗結果與AIMBS相比，在1080P的BD-PSNR平均可以增加0.027 dB，而BD-Rate平均可以減少0.66%。顯示本研究所提之增強型跳位機制可以有效地進一步降低編碼位元率，並達到更佳的視訊品質，同時也減少了些許編碼時間。One of the main contributions of H.264 is the intra prediction in spatial domain, which increases the coding performance of I-frame significantly. Intra4×4 mode provides nine directional prediction modes for every 4×4 block. However, for each 4×4 block, 1 or 4 bits are required for indicating intra mode in the encoded bit stream for successfully decoding. Therefore, 16 ~ 64 bits are required to represent intra mode bits for each intra4×4 macroblock. To reduce this burden, Dae-Yeon Kim proposed an algorithm called: “adaptive intra mode bit skip (AIMBS)” in 2008. If all reference pixels of current block are quite similar, the DC mode can be selected as the prediction mode without significant distortion so that the prediction mode bit can be skipped in the proposed method. But this method can be improved by considering the bit skip for more modes. The reference pixels usually exist the same direction with current block. In this work, the correlation between vertical and horizontal differences of reference pixels and encoded modes are investigated. The mode skip of vertical and horizontal prediction modes is integrated with AIMBS to achieve better coding efficiency. The simulation results show that the proposed mechanism can improve 0.027dB BD-PSNR and reduce BD-Rate by 0.66% at 1080P compared with AIMBS on average. We observe that this method can be further improved by including more modes for bit skipping. In this work, the correlation between vertical and horizontal differences of reference pixels and encoded modes are investigated.
