
COMPUTER ENGINEERING
1999 25 8 Vol.25 No.8 1999



ɫͼ

ժҪ˻ɫʵͼɫռѡɫԼɫƲԽۡڴ˻ϣֵ͵Ļɫͼص㡢ؼʵַԼڵĸϸķͱȽϡ
ؼʣͼݣͼݿ⣻ɫֱͼ
Color-based Image Retrieval Techniques
Ling Ling
(Dept.of Eng.and Computer Graphics,Guangdong Univ.of Tech.,Guangzhou 510643)
AbstractThe color-based image retrieval methods are discussed in the paper,with emphasis on the choice of color space,the quantization of color and the similar measure of color.On the basis of this ,the featrues of the two typical color-based image retrieval methods,the realizable mechods of key techniques and the existing basic problems are analysed and compared in greater detail.
Key wordsImage retrieval;Content-based;Image database;Color histogram
Ŷý塢ͨš缼ķչͼӦԽԽ㷺ӴͼЧĹԵΪҪͼݵĹǴͼгȡϢڹϵݿаǵɸʽݣͼļڶͼԵַƥϣֻԵķҪûܾȷѯݷǳѡǵݵͼдͳı޷ԽԣˣоͿݲѯͼݿǵоȵ㡣 
ɫʶͼҪ֪֮һɫĲѯǻݵͼķΪֹݼͼݿĵʹҪУ IBM  QBICTM ױǴѧ Visual SEEK ͼѯϵͳִѧ MIT ýʵҿ Virage ϵͳȣӦ漰ýݿ⡢ͼݡ̱ҽͼϵͳͼ [1] ͻɫͼ漰ĹؼԼͱȽϡ 
1 ɫģ͵ѡ 
ɫͼɵĻҪأѡһӾԵɫģɫͼĹؼ 
ͳ RGB ɫģ͵ĿɷֱɫǷԵģûֱ , Բһõɫϵͳ HSV ɫģ˵ӾԱȽϽӽɫ H Ͷ S  V3 ɡģ;ԣɸ֪ɫɫӦֵϵŷ¾ɱ [2]  HSV ɫģͱ RGB ɫģ͸ֱۡ׽ܡ 
ͨԱ任ɽ RGB ɫռ rgbֵת HSV ռhsvֵ [2]  
 v'=max(r,g,b)  r',g',b' Ϊ 
1

2
 
3
ʽУ r,g,b  [0  255]  h  [0  360]  s  [0  1]  v  [0  1] ɫռɫc1 =(h1,s1 ,v1)  c2 =(h2 ,s2 ,v2 ) ֮ľΪ [3]  
d(ci ,cj )=[(vi-vj) 2 +(sicos hj-sjcos hj )2 +(si sin hi -sj sin hj) 2 ] 1/2(4) 
ɫ֮ƶΪ (5) 
2 ɫ 
۶Ӿķֱһԣ˶ɫռʵǱҪģѡһԵɫЩӾϢӶʹͼļܽԼļʱ䡣 
ɽ HSV3˵ɫ֪зǵȼɫ H ռֳ 8 ݣͶ S  V ռֱֳ 3 ݣɫʵĲͬΧ 3 ɫϳΪһάʸ 
l=HQs Qv +SQv +V(6) 
УQs(  S  )=3  Qv(  V  )=3  (6) ʽɱʾΪ 
l=9H+3S=V(7) 
ʽ H  S  V3 ɻ 72  (bin) һάֱͼ [2]  
Ҳɲģʽʶеľ෽ɫռ [4] ɫռеɫҪΪ N ɫͳƷѰ N ŵĳʼģѰɫռ N дԵɫֵȻõķɫռеɫС׼ N ľ࣬ɵõĸ֪ڽ N  s i (1  i  N)-- ɫϣʹÿɫ s i  cen(s i ) γµ N ģ N ɫϵĴɫ 
(8) 
ʽУ x Ϊ si ɫһɫ count[x] Ϊ x ͼгֵĴ Ni Ϊ si ɫɫͼгֵܺ͡ 
3 ɫƲ
ֵЧʾͼɫϵƳ̶ȣɫƶ⣬ֱӰЧҪڡģʽʶУƶþ뷨ھɲþֵ롢ŷ¾롢ξ [5]
ֵʽ£ 
 (9) 
ŷ¾ʽ£ 
 (10) 
У Hq Ϊͼɫֱͼ Ht Ϊͼɫֱͼ hq [ci]  ht [ci] ֱʾͼͿͼ i ɫصƵ
ŷϾڼûпɫԣȻ˵ӾоĿǰ˷һȱݵķǲֱͼξ뷨ŷϾмӽɫƾʽ [3]  
D(Hq,Ht)=(Hq-Ht)T A(Hq-Ht)(11) 
ʽУ AnnΪɫƾ
(12)
Ԫ sij ʾɫci cj ֮ƶȣ (5) ʽáֱͨͼ֮趨統ǵľСڸֵʱ D(Hq ,Ht )  t ʱ Ht ӦͼΪҪͼ 
4 ڶɫչͼ
һͼɫܹӳͼĻòΪѯҪƥ䣬ǵ۶ɫ֪һΧھģԣѡɫʱҲһԺ͸˵ءˣɫչֲȱ©һЧķ 
ֲ̬ϷԶָɫչûָɫΪ l0 ģΪ  򾭹ֲ̬Ϻֱͼ f (l) £ 
13
֤ΧڵɫûʵҪɫأ˿԰ʽȷڼɫչֵ 
⣬һͼϵɫЩɫͼռıҪ̶ҲͬòͬģȺͱķһЧɫֵƥ䡣 
ʵ [2] ÷Դ߽緶ΧɫļЧԵĸơڴͼɫɫѡԼ˵ʹģ  ϴƱɼΧؽ˼Чʡˣ׼ȷٵѡͼɫҪһо⡣ͼֱָͼЧͼ׼ȷ[3] 
5 ͼֱָͼ 
ɫֱͼȻܽϺõطӳͼиɫƵʷֲûбɫĿռֲϢȻ˵ӾӦʽϸߡΪˣɲûͼֱָͼ[3]Խ 
һǶͼĶı߽ȡ ( ȡͼ )  Snake ߽ȡ㷨ֹϵķԼС任ķȡͼȻÿɫÿɨ㷨ֱͼͳƣþ뷨ԴͼͿͼƥ䡣ʵöָͼƼ 7% 15%  
ǽһͼ򵥵ػΪ n  n ͼ {P1 P2 Pn2 } Ȼֱÿͼֱͼͼֱͼ͵ڸͼֱͼ֮ͣ 

ֱͼɵÿռĶʽΪ 

ɼͼֱָͼļ㾫ҪȶͼֱͼľҪߣӶЧ˼ȡɴҪΪԾ븶ϸߵĴۣԽȷľ빫ʽҲԽߡθЧ׼ȷزѯûͼͼҪĿ꣬ö׶ε "  "  " ϸ " ϵƥʧΪһ߼ЧʵЧ 
6  
ݲѯͼݿͼͼʶݿ 3 ļɹһǰķչͼݿͼʹͳݿ⼼ӦãΪͼṩ֧֡Ŀǰɫͼо׶ΣлؼļδõܺõĽ磬ЧȡɫμɫֱͼƾļȡܺܺõؽЩ⣬ؽʹͼһԾ 
ߵλ㶫ҵѧͼѧң 510643
ο
1 Finn R. Query by Image Content. IBM Research,1996, 3:22  25 
2 򻪣ΰ . ڶɫͼ㷨о ʵ . о뷢չ 1999,36(1):96  100 
3 ³ƺ . ɫͼݿ  . о뷢չ 1999  36(3)  359  363 
4 ᣬ . ɫͼһ¾㷨 . 㶫еѧԺѧ  1996  14(2):58  62 
