現今視訊串流的品質取決於傳輸網路之頻寬，目前以區塊為基礎的離散餘弦轉換方法，廣泛地應用在靜態影像及動態視訊的壓縮上，當視訊透由網路進行傳輸時，於有限頻寬的狀態下往往採用較低位元率之壓縮方式，導致影像重建後產生嚴重的區塊效應。近年學術界提出以稀疏表示法進行影像重建，惟習知作法之運算複雜度較高或需額外參照大量自然影像以進行影像字典學習，為了提升運算速度以符合目前大量視訊之處理需求，本文提出一種具自我調適性(adaptive)，可動態使用鄰近畫幀群組作為影像特徵字典之學習機制。本文以PSNR及SSIM作為視訊品質衡量指標，實驗顯示本方法與相關文獻比較，於處理後可達近似畫質條件下，相較現有技術可節省運算時間達30倍以上。

The quality of video streaming depends on the transmission bandwidth of network. Block-based discrete cosine transform method is widely used in static images and dynamic video compression. When the video is transmitted through the Internet under the limited bandwidth, lower bit rate compression method yields severe blocking effects after image reconstruction. In recent years, the sparse representation method is proposed and adapted for image reconstruction. The high computation complexity and additional reference to a large number of natural images for dictionary learning are still challenging problems. To enhance the computing speed to meet the demand of processing a large number of video s, this paper presents a self-adaptive and dynamic neighboring group learning mechanism for dictionary training. In this paper, PSNR and SSIM are used as video quality metrics. Compared with prior work, experimental results show that this method can significantly reduce computing time more than 30 times under similar image quality.