PowerPoint 簡報

Size: px

Start display at page:

Download "PowerPoint 簡報"

雹单
4 years ago
Views:

1 數 Study on the Three PCA Methods for Feature Extraction and Classification of Digital Image 羅祿立 94 年 9

2 (Principal Component Analysis, PCA) 理料度更利 PCA 益. 行 MS. 不 PCA 略理論行 (Fe ature) 不 3. 利類 (Bayes classifiers) 不 PCA 略行數

3 理論理 () y One Pixel 300nm 400nm 500nm 600nm 700nm 800nm z x Reflectance Wavelength(nm) 度 (:nm) 3 數

4 理 () 濾濾 CCD camera 濾理 4 數

5 料理 (Principal Component Transform ation, PCT) 念數立數量量 (Princip al Components, PC) PCA 理論 ( 略 A)( 略 B) ( 略 C) 5 數

6 ( 略 A) (N N) pixels' 行量 (N )(Turk & Pentland, 99) Γ Γ i, ( i,..., N ;, k,..., N ). N, k N N 3 (N N) 列 N 量 6 數

7 M 列行量 (N ) Sample Sample Sample Sample Sample M Sample M N Γ i N Γ i N Original Image Column vectors 4 量串 7 數

8 M 量量 Ψ M Γ M M 量 i 量 Φ i Γ i Ψ i i (i, K, M) 異 (Covariance matrix, C) M T T C Φ nφ n AA, A M n [ Φ, Φ 度 N N,, Φ M ] 量 x i (Eigenvectors) 8 數

9 L 度 M M (Turk & Pentland, 99) L T A A 量 v k k M<N C 量 U l M (Eigenspace, U) 度 M N U l M v Φ l, k l, k k..., M M' 量 ( ) 9 數

10 兩類識 G G 量 T Ω ( Γ Ψ) T U U ( Γ Ψ) Ω G G 類參數 t U Ω U N N' t T ( Γ Ψ) t M' 量 0 數

11 ( 略 B) B(N N) pixels X(N ) 量 ( 林 00 Gonzalez & Wood 00) Band Band Band Band Band B Band B N Γ i N X i N Original Image Column vectors 5 量串列數

12 數數異 (Covariance matrix, C) ],,, [, B T B n T n n A AA B C Φ Φ Φ Φ Φ 度 B B 數 X 異數 C 量 E B BB B B B B B X X X e e e e e e e e e PC PC PC M L M M M L L M PCE T X X i 量 E( B ) (PC image)

13 ( 略 C) Band B N Band Band Γ i N Sample N Original Image Γ i Γ i Sample Column vectors Sample M (Quintiliano & Santa-Rosa, 003) 行量列 Γi, Γ, k, p ( i,..., BN ;, k,..., N; p,..., B). N N B' Band Band Band B Γ i Sample Sample Sample M 6 量串列 M' 量 3 數

14 類理論率 (Piror Probability) 類率 (Posterior Probability) 率類率數 d (x) d ( x) p( x / ω ) P( ω ), p ( x/ ω ) 類類率 P( ω) 類率 4 數

15 數數 5 兩類 G G 率 0.5 ) ( ) ( ω P ω P, ) ( ) ( - exp ) ( ) ( / / m x C m x C x d T n π N i i x N m n 類量度 N 練數異 N i i i m x m x ) )( ( C 數數, log ) ( ) ( - ) ( C m x C m x x d T 若 x 數 x x ) ( ) ( x d x d < ) ( ) ( x d x d >

16 實驗方法與設備多分光影像擷取系統紫外線 Japan, SONY,XC-EU50, 300~40nm CCD 規格可見光 Japan, SONY, XC-ES50, 400~770nm 紅外線() Japan, SONY, XC-EI50, 400~870nm 紅外線() Japan, Sakai, KC-700E, 850~050nm 濾鏡長波通濾鏡 LP400~700每隔0nm及850 nm共6種短波通濾鏡 SP400 ~700每隔0nm一種共5種紫外光源 USA, UVP, UVGL-58,366nm波段光源可見光源(短弧氙燈) Taiwan, Rocoes, 300~800nm 紅外LED光源() Japan, Saikai, 850nm及880nm 紅外線光源() Taiwan, CAMTEK, 0W,840nm 圖8 多分光影像擷取系統實體圖鏡頭 Japan, Sakai, LMV43H, 焦距4.8 mm, 光圈F.3 其它影像擷取卡 IMAQ PCI-4 (NI) 6 三種主軸成分法於數位影像特徵擷取與判別之研究

17 流 MS MS ~7 7 理理 PCA 不 PCA 略立類流 7 數

18 理 Step Step Step 3 Step 4 Step 5 Step 6 Step 7 切 8 數

19 論 ~7 MS 400~770nm Day Day 3 Day 4 Day 7 Day 6 Day 5 9 數

20 不 MS 400~700nm 400~770nm 480~560nm 560~60nm 300~400nm 850~870nm 850~050nm 0 數

21 不略 PCA() 略 A ( ) 3 PC PC3 PC6 PC50 略 B ( ) ( ) PC PC PC3 PC6 數

22 不略 PCA() 略 C ( nm nm) 3 PC PC3 PC PC4 PC34 PC50 數

23 略 A 率數 nm nm nm nm nm nm nm () ( 3.7) ( 3.8) ( 5.0) (.) (.3) (.0) () 77.0 ( 3.) 79.0 ( 3.3) 76.8 ( 3.6) 78.0 (.5) 77.5 ( 3.3) 76.8 (.7) 74. ( 5.6) 3 數

24 略 B 率類率數

25 略 C 率 B B B3 B4 B5 B6 B7 () (.5) ( 4.) (.7) (.3) ( 0.9) (.) () 80.7 (.8) 8.7 ( 3.8) 80.0 (.8) 8.0 (.6) 78.3 ( 5.) 8. (.) 80.5 ( 3.6) 5 數

26 略 C 不數率 B4 B 數不數率 ( 4 ) 6 數

27 論類行類識料度降不流不略利數略 7 數

28 度例數識行識識率量數數索識 8 數

29 National Pingtung University of Science & Technology LAB of Multi-Spectral & Image Processing 9 數

九十三年第三期檔案管理工作研習營學員建議事項答覆情形彙整表

九十三年第三期檔案管理工作研習營學員建議事項答覆情形彙整表 94 年 1 理 1 1. 2 18 3 理 16 4 行 26 2 不錄 2.96 年 7 1 理錄 1. 3 錄年年陸都年利錄年理年度列 94 年錄行錄數量錄 2. 錄年錄利錄理落行錄理錄更 1 2 臨省北 4 2 行不 ( ) 行理理理 3 隆 1. 理 ( 降 ) 更錄理見錄理見理數 3 數數量 2. 若來