Docstoc

行動傳訊代理人之建置 在網際網路傳訊機制之應用

Document Sample
行動傳訊代理人之建置- 在網際網路傳訊機制之應用 Powered By Docstoc
					Color Image Retrieval based on Primitives of Color
                   Moments

J.-L. Shih, L.-H. Chen, IEE Proceeding-Vision, Image and Signal Processing, Vol. 149
No. 6, Dec. 2002, pp. 370 -376.



           Advisor:Prof. Chang, Chin-Chen
           Student:Chen, Yan-Ren
           Date:2003/03/25




                                                                                       1
Outlines


   Introduction
   Proposed Method
   Extraction of Primitives of Color Moments
   Color Image Retrieval
   Relevance Feedback Algorithm
   Experimental Results
   Conclusions




                                                2
Introduction

                         Image Retrieval
                            Methods




            Text-based                         Content-based



                      Text
       Keyword     Description
                                             Color             Shape     ......


                                   Color              Color
                                 Histogram           Moments    ......




                                                                                  3
Proposed Method Flowchart




                 Extract Features
   Query Image   (Primitives)
                                     Similarity               Matched
                                     Measure                  Results


     Image                                        Relevance
    Database                                      Feedback
                                                  Algorithm
                 Features Database




                                                                        4
Extraction of Primitives of Color Moments



                                   Y, I, Q
                     Divide
       Image                       Color
                     Image
                                   Space



                     Cluster       Extract
       Extract
                      Color         Color
      Primitives
                    Moments       Moments




                                             5
Color Moments

                          Y component                                  N
                                                            1                                 20  30  30  40
                         P1 P2 … Pj                     M 
                                                            N
                                                             1
                                                             i         P
                                                                       j 1
                                                                              i, j   M 11 
                                                                                                      4
                                                                                                                 30
                          I component

                          Q component
                                                                                              1
                                                         1 N                          h
                                                                                     
                                                                                                  h
                                                   M i    Pi , j  M i1
                                                      h
                                                         N
                                                                                        
                                                                                        
                                                          j 1                                  1
                                    20  302  (30  30) 2  (30  30) 2  (40  30) 2  2
                              M1  
                                2
                                   
                                                                                            50  7.07
                                                                                          
                                                             4                           
  M: moment                                 1
                                  h=1, M i is mean of i component
  N: total pixels
                                  h=2, M i2 is standard deviation of i component
  P: color value
  i: ith component
  j: jth pixel in i
  h: total M in i
                         CT=[ct1,ct2,..,ct6]= [1M 1 , 1M 1 ,  2 M 2 ,  2 M 2 ,  3 M 3 ,  3 M 3 ]
                                                         1         2            1             2        1        2
  : weights for Y,I,Q
  CT: feature vector     =[Y(1×30,1×7.07), I(2×10,2×2), Q(1.5×20,1.5×4)]


                                                                                                                6
Primitives of the Image (1)

 CBa  [ cba ,1 , cba , 2 ,, cba ,Z ]
      [1M a ,1 , 1M a ,1 , ,1M aH,1 ,  2 M a , 2 ,  2 M a , 2 , ,  2 M aH, 2 ,  3 M a ,3 ,  3 M a ,3 , ,  3 M aH,3 ].
            1          2                         1             2                              1            2



                Y(1×30,1×7)                            I(2×10,2×2)                         Q(1.5×20,1.5×4)




                                                          a
M: moment
i: ith component
H: total M in i
z: H×3
: weights for Y,I,Q
a: ath block of the image
CB: feature vector


                                                                                                                                 7
  Primitives of the Image (2)


    CB1       CB2
                                         Clusters CB2 PC1 CB                                                      CB4
                                                             3                                CB1 PC2

    CB3       CB4
                                             Extract Central Vector Examples                                          Y
                                                                                      Block a              M1                 M2
                                                         (pc1,1, pc1,2)                   a=2              20                 4
                                            PC1
                                                          =(21.5, 4.5)                    a=3              23                 5
M: moment
H: total M in i                             PC2          (pc2,1, pc2,2)                   a=4              26                 6
z: H×3                                                     =(28, 6.5)                     a=1              30                 7
k: kth cluster
                                           Weight=1, Threshold=5
n: size of kth cluster
J:1,2,...,nk                                                                  nk                  nk                  nk                       nk
a: ath block of the image
CB: feature vector
                                                                               CB
                                                                              j 1
                                                                                          k
                                                                                          j        cb  cbk
                                                                                                           j ,1
                                                                                                                                  k
                                                                                                                                  j,2          cb         k
                                                                                                                                                           j ,Z

PC: primitive (central vector)   PCk  [ pck ,1 , pck , 2 , ..., pck ,Z ]                     [ j 1            ,   j 1
                                                                                                                                        , ..., j 1               ]
                                                                                     nk               nk                     nk                       nk

                                                                                                                                                                      8
Color Image Retrieval – Similarity Measure


            Query
            Image
           Features
 PCkq  [ pck ,1, pck ,2 , ..., pck , Z ]
            q       q             q
                                            Distance                 Minimum
                                            calculate                Distance
          Features in                                                                      Matched
           Database                                                                        Results
                                                              Z

  PC  [ pc , pc , ..., pc ]
      s
      
                s
                ,1
                        s
                        , 2
                                    s
                                    , Z
                                             D _ PCq, s
                                                   k ,       ( pc
                                                              i 1
                                                                     q
                                                                     k ,i    pcs,i ) 2




                                                                                                     9
Relevance Feedback Algorithm


             Proposed method




                                                    User Interface
             Color moments
             Color set                 Relevance
             Color correlograms        Feedback
  Features
             Dominant color            Algorithm
  Database   Color layout
             Color structure
             Color histogram...



   Image
                 1.System give query results by combined features.
  Database       2.User choices r similar images.
                 3.According user response, R.F.A choices query method.




                                                                          10
Retrieval Results from D2 Database




                                     11
Precision Comparison on D1 Database (1)

                                                 N=50×0.8=40




                  N   N: number of relevant images retrieved
           Pr        K: total number of retrieved images
                  K


                                                               12
Precision Comparison on D1 Database (2)




                                  K=50, T=100




                                                13
Precision Comparison on D2 Database




                                      14
Conclusions


   Proposed a image retrieval method based on primitives of
    color moments.

   The color moments of all blocks are extracted and clustered.

   Central vectors are considered as primitives (Feature vectors).

   Similarity measure is used to perform color image retrieval.

   Relevance feedback algorithm determines the most
    appropriate feature.



                                                                   15

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:4/18/2013
language:English
pages:15