基於分散式視訊編碼理論的雙導程心電圖壓縮系統

Abstract

因應高齡化社會的未來趨勢，居家健康照護已成為先進國家重點發展的新興服務產業。本論文研究旨在針對低耗能的無線心電感測配備，探討低複雜度的雙導程心電訊號壓縮機制。主要是採用分散式視訊編碼架構，其優點是利用不同導程間的相關特性，將系統的運算負荷由傳統的編碼端移至解碼端。第二肢導程的心電圖像視為Key frame，並採用H.264標準技術進行畫面內編解碼。第三肢導程的心電圖像則視為WZ frame，其離散小波轉換係數的高頻部分採用SPIHT編解碼技術；而低頻部份則執行通道編碼並只傳送校驗子，在解碼端再結合Key frame提供的邊訊息重建其原畫面。有別於前人研究，本系統的兩種畫面源自於不同導程的心電圖像，其時間同步且不存在相對移動的關係，因此無法沿用傳統基於移動估測的邊訊息生成機制。針對此項議題，我們提出以類神經網路為基礎的邊訊息生成機制。針對心電圖資料庫進行的實驗結果顯示，類神經網路能具體描述畫面間的相關性，其生成的邊訊息也大幅降低編碼失真率。

Wireless telecardiology is an active research with the goal of ubiquitous health-care services. This work focuses on the distributed video coding(DVC) of two-lead ECG images for data compression. The key frames composed of lead II ECG images are compressed by H.264 intra coding, and Wyner-Ziv(WZ) frames composed of lead III ECG images employ the WZ and SPIHT coding to compress its low-frequency and high-frequency wavelet sub-bands, respectively. In WZ coding, the encoder acts as a channel encoder and only transmits the syndromes, and inter-frame decoding is performed by combing the received syndromes and side information generated from the reconstructed key frames. Recognizing that two-lead ECG images are time synchronized and lack the movement, we propose to apply the neural network to model their inter-correlation for the side information refinement. Under the DVC framework, the heavy computing burden was shifted from the encoder to decoder, which makes it suitable for wireless ECG monitors. Experiments on St. Petersburg INCART 12-Lead Arrhythmia Database demonstrate that our proposed method achieves better rate distortion performance than motion estimation based side information generation.