数位影像中熵的计算与应用-义守大学.PPT

數位影像中熵的計算與應用 義守大學 資訊工程學系 黃健興 Outline Entropy Definition Entropy of images Applications Visual Surveillance System Background Extraction Conclusions Concept of Entropy Rudolf Julius Emanuel Clausius , 1864 化學及熱力學 測量在動力學方面不能做功的能量總數 計算一個系統中的失序現象 描述系統狀態的函數 經常用熵的參考值和變化量進行分析比較 Information Theory Claude Elwood Shannon , 1948 運用機率論與數理統計的方法研究資訊 編碼學 密碼學與密碼分析學 數據傳輸 數據壓縮 檢測理論 估計理論 數據加密 Definition E is the expected value, I is the information content of X. p denotes the probability mass function of X Advantage Whole Image M×N Matrix Histogram N×1 Vector Entropy Single value Entropy of Image Pixel Color Pixel Distribution Horizontal Vertical Texture The Statistic of gray-level Position Information Normalize the size of image Edge Detection Sobel Canny Horizontal Projection Vertical Projection Sobel Edge Detection Sobel Filter Sobel Edge Detection(cont.) Sobel Edge Detection(cont.) Horizontal Projection Horizontal Projection(cont.) Vertical Projection Vertical Projection(cont.) Local Binary Pattern Pattern Texture Pattern Center Pixel gc Surrounding Pixel gi(i=0, 1,…,p-1) Label Local Binary Pattern(cont.) Definition E is the expected value, I is the information content of X. p denotes the probability mass function of X Applications Visual Surveillance System variance of video information Background Extraction Block for pixel Visual Surveillance System Visual Surveillance System Gray Prediction – GM(1,1) Gray Prediction – GM(1,1) (cont.) Step 1: Step 2: Step 3: Gray Prediction – GM(1,1) (cont.) Step 4: Step 5: Gray Prediction – GM(1,1) (cont.) Step 6: Step 7: Visual Surveillance System Visual Surveillance System Background Extraction Non-recursive approaches Selective update using temporal averaging Selective update using temporal median Selective update using non-foreground pixels Non-parametric model Time Interval (It-L,It-L+1,It-1) Probability Density Function Background Extraction Recursive approaches Kalman filter