| Abstract: | 本研究探討非靜態 (Nonstationary) Gaussian空間變化autoregressive (SVAR)程序的傳輸率與失真度函數 (RDF,R(D)),並對此一隨空間變化,非靜態二維數位影像傳輸系統,作理論及實驗性的研究。本研究針對非靜態Gaussian SVAR程序,先推導其傳輸率與失真度函數。並藉由此傳輸率與失真度的計算,我們在此研究中發展建立起一完整的適應性餘弦轉換 (ADCT)影像傳輸系統。本計劃以變動式影像方塊SVAR非靜態Gauss-Markov影像模式,來更精確估算真實世界影像的非靜態統計性。所建立的適應性餘弦轉換(ADCT) 影像編碼系統,是以指定每一個不同尺寸之影像方塊,具有相同的平均失真度,並以變動每一影像方塊的編碼壓縮率,來達成此一固定失真度的要求。系統模擬結果,比較具有SVAR影像模式之適應性餘弦轉換影像編碼系統與傳統式具有固定編碼壓縮率的靜態餘弦轉換影像編碼系統。此適應性系統的執行效能,並與提出之傳輸率與失真度的理論值比較。結果明顯證明,應用精確的統計性影像模式及適應性影像編碼系統的設計,將大幅減低影像傳輸所需之資訊傳輸量,並能夠保有影像資料重組的品質。
Rate-distortion function (RDF, R(D)) for nonstationary Gaussian space-varying autoregressive (SVAR) process and two-dimensional space-varying nonstationary digital transform image transmission system are investigated theoretically and experimentally. We have derived the R(D) function for nonstationary SVAR Gaussian processes. The rate-distortion measure then serves as a guide to develop an adaptive transform image transmission system. An adaptive discrete cosine transform (ADCT) image transmission system is established in this study. To accurately estimate the nonstationary statistics of a real-world image, we proposed a variable block size SVAR Gauss-Markov image model. The proposed ADCT image coding system is implemented with the same average distortion designated for each variable size image block and the code rate for each image block allowed to vary in order to meet this fixed distortion criterion. Simulation results are given to compare the reconstructed images coded by the ADCT scheme, based on the SVAR image model with the fixed-rate, stationary cosine transform coding scheme. The theoretical rate-distortion measure serves as an ultimate bound on practically achievable system performance. Overall performance for the transform image transmission system based on an SVAR image model is studied and compared with the theoretical R-D bound. The results demonstrates the efficacy of employing an accurate statistical image model and adaptive image coders design to minimize the overall transmission rate requirements while preserving the quality of the reconstructed imagery data. |
