利用影像時空隱含資訊的疊代式通道解碼

Abstract

在影像壓縮的技術中常利用禎內預測壓縮的來避免錯誤波及到後續的畫面。 尤其是在分散視訊壓縮系統中，禎內預測的畫面更被使用於預測其他禎的畫面。 因此會有更多的畫面只使用到禎內預測。但是禎內預測的壓縮率遠低於使用禎間 預測的壓縮率，於是很多能量浪費在傳輸可以從另一張畫面預測出的資料上。我 們發現只使用禎內預測壓縮出來的資料有很高的關聯性，而這些關連性可以用來 減少所需的傳輸能量。但是這些關連性在傳統的壓縮技術中會非常的難以利用。 於是我們稍微改變了傳統的禎內壓縮，使得這些關連性變得比較好利用。另外， 由於60GH微波系統的興起，這個觀念更可以延伸到未壓縮的視訊壓縮上。我們利 用了雜訊模型以及馬可赫夫隨機場分別來描述禎與禎之間的關聯性和禎內之間像 素的關係。利用這些關連性可以使得經過通道解碼的資料錯誤率降低。因此我們 所提出的系統可以使得每個禎所需傳輸的能量相似於利用禎內壓縮所的畫面所需 的傳輸能量。這表示這個系統可以操作在非常低的信噪比下。由於60GHz這的頻 段的訊號能量遞減效應會非常大，這個系統非常適合這個頻段的視訊傳輸。此 外，由於我們提出的系統的紀錄器複雜度非常的低，非常適合監視系統以及傳輸 連續拍照這類需要低成本和低能量紀錄器的應用。

Intra-coded frames are frequently used to prevent error propagation, and moreover, in the distributed video coding system, the intra-coded frames is used to predict the adjacent frames. However, the bit-rate of the intra-coded frames is much more than that of the intercoded frames, which results in power waste. We find that the high correlation between the coded bitstreams of intra-coded frames can be used for channel decoding to reduce the transmission power. However the correlation is difficult to be exploited since the pixel values are encoded as logical bits. Therefore, the source encoder is modified for the channel decoder to utilize the identical data between the intra-coded frames. This concept can be extended to the uncompressed video transmissions, mainly due to the emergence of the 57-66 GHz mmWave systems. To transmit the uncompressed video power efficiently, we propose a channel decoder that utilizes both temporal and spatial redundancy among the video frames. The temporal redundancy between the frames is exploited via a correlation noise model, and the spatial redundancy inside each frame is exploited with the aid of Markov random field (MRF) model. By combining the above models and an error control code, the proposed system is more error robust. The recording devices of the proposed system is less complex than the system transmitting intra-coded frames while the PSNR performance is maintained under the same transmission power per frame comparing to the transmission of intra-coded frames. That is, the system can operate in an extremely low SNR condition which is common in mmWave frequency bands due to the signals suffer greater propagation loss. Besides, the low-cost and low power characteristics make it a perfect match for surveillance systems and the applications transmitting serially shoot images.