應用於視訊解碼系統之內嵌式壓縮器/解壓縮器使用向量量化/離散餘弦轉換混合編碼

Abstract

本論文提出一種應用於視訊解碼系統之內嵌式壓縮器與解壓縮器。該內嵌式壓縮器使用兩級向量量化器來編碼去方塊濾波器的輸出影像資料，向量量化器所產生的誤差影像再以離散餘弦轉換執行去相關的動作，接著使用修正式位元平面區域編碼來處理頻率域係數，最後將兩個向量碼索引值及誤差影像的頻率域係數資料位元封裝後存放到外部記憶體。壓縮後的資料採用固定長度封裝以方便讀取記憶體的資料。基於這樣的壓縮方式，解壓縮程序相當簡單。由讀取的資料封包中得到的頻率域係數後經逆向離散餘弦轉換後得到誤差影像，再加上向量碼索引值所對應的資料碼即可完成影像還原。經過12種不同特性的4:2:0 CIF格式影片序列測試，總共測試了3,183張影像圖片，我們所提出的編碼方式在平均33dB的PSNR為前提下達到2.0的壓縮比。外部記憶體大小的需求可降低為原來的一半。再者視訊解碼系統中的移動補償所需記憶體頻寛及資科存取次數可以有效地減少，進而降低記憶體介面之輸入與輸出接腳的功率消耗。

In this thesis, we propose a coding approach of the embedded compressor/de-compressor for video decoder. The data samples output from de-blocking filter are encoded by two-stage vector quantizer (VQ). Then de-correlations are performed by discrete cosine transform (DCT) for the differential values between source image and the quantized data. DCT coefficients of differential data are coded by the modified bit plane zonal sampling. Finally, two VQ-code indices and sampled data bits of DCT coefficients are packed and stored into the external memory. Fixed length compressed data of each image block is designed for good memory accessibility. Based on this coding approach, image reconstruction is quite simple. DCT coefficients bits of the read block are unpacked. Differential values of images are recovered by inverse DCT (IDCT) and added to the corresponding codes of these two unpacked indices. Image reconstruction is completed. Totally twelve 4:2:0 CIF-format video sequences with different characteristics which contain 3,183 frames are tested for performance measurement. Compression rate 2.0 with average PSNR 33dB can be achieved. The requirement of external memory is reduced to 50% compared with no embedded compressor. Moreover, both memory bandwidth requirement of motion compensation of video decoder and power consumption on I/O pins of memory interface can be reduced efficiently.