Զѧ
ACTA AUTOMATICA SINICA
1997 23 1 Vol.23 No.1 1997



һֶ߶ȷĿⷽ
Сκաɣũ
ժҪȻµԶĿĿһ.ݳ߶ȱ仯ʱȻĿķ仯һص㣬һַβֵԶĿⷽ.ʵβֵԶĿⷽܽϺõȻСĿԶ.
ؼʡ߶ȣβĿ.
A METHOD FOR TARGET DETECTION 
USING MULTISCALE FRACTAL
WANG GUOYOUZHANG TIANXUWEI LUOGANGSANG NONG
(Institute of Pattern Recognition & Artificial Intelligence
Huazhong University of Science & Technology,Wuhan 430074)
AbstractAutomatic object detection from natural scene clutters is fundamental to object detection.When the scale is changed,the fractal features of man-made objects in a natural scene change much more violently.In terms of the characterization,this paper presents a new method for automatic target detection using fractal maximum.Experimental results show that the method based on fractal parameter maximum map(FPMM)performs very well for small man-made object detection from a natural scene clutter.
Key wordMultiscale,fractal feature,target detection.
1
ĿǰԶĿ·ΪࣺһǴĿԶ⣬ڶСĿԶ.ڵһԶĿУĿ걾ϴ󣬰лҶȡͽṹϢԶƥʶ12.ڵڶԶĿУĿ걾СҶȡͽṹϢܲʶĿԶⳣԼⷽ3.ȻĿԶʶ˺ܴ.ҪУһ֡ͼвصܱ֤֡ͼԣڶӵҲȷĿ״ҲȷƥʶѣʵʱͼеĿҶȺƣģʵʱͼиӹ⡢⼰ɵ壬ʵʱͼĿСȷĿĻҶȻпַܳת.ЩصӰ˺ĿԶɿʶ.ڴʵϣһֻڷβֵĿԶʶ㷨Ϻõؽ˺СĿԶʶ.
2ԶĿʶԭ
2.1θ
Ϊһ˵Ŀѧ֧ΪоʹֲṩĹ.ף4Keller÷νͼķָף5PentlandָȻķάڳ߶Բ仯Ŀķά߶ȱ仯仯άȻеĿ.ȻǲúӼ͵̺Ƿ6ݼάʱȻĿķά仯ԣˣҪѰµķμģ.
 M()Ϊ߶  ¼ F ĳֶd Ϊά1DΪάУM()=K.d-DôǳƼFM()¾зD Ϊ Fķά.
2.2µķβ
άͼźI(X) X  X Gx(x)Ϊ
(1)
1ٶ Gx(x)Ƕƽ̣
(A)Gx(x)ľֵи̬.
(B)I(X)ľֵȾз.
(C)I(X)ľֵȷβĿ괦ڼֵ.
֤ƽ֪̿
EGx(x)=m (mΪ)(2)
ɶ֪̿DۣGx(x)=0,
Gx(x)=E(Gx(x)).(3)
 Gx(x) ľֵи̬.
(1)(3)ʽ֪ I(x+x)-I(x)ľֵҲи̬ԣ
(4)
(4)ʽ֪
(5)
(5)ʽ֪I(X)ľֵȾз
(6)
ʽУNX+XNXֱΪX+XXԪصĸٶȻֵԣĿܱԾȵ򣬸ʶСĿԭĿ򱳾Ų.ͼ1ʾûYԲZֱʾX+XXüYZڵľֵƼX+XX֮ľֵ


ͼ1 X+XXλùϵʾͼ
(7)
β H(X)Ϊ
(8)
1) ĳ߶ X Z øĿʱ Y Ϊ.ĩǰļ֪ڼֵβ H(X)ڼֵ
2)  YZĿʱβ H(X)=0
3)  YZǱʱβ H(X)=0
ٶĿСĿ꣬ Z ĴСʹüZøĿ꣬YǱʱβ H(X)һڼֵ.
2.3βļ㷨
Ϊ Z ΪٶĿ괰ؼY ΪٶĻαؼ(ͼ2ʾ)Ii Ϊ i ĻҶȣNY,NZ ֱΪ YZ ظֵֵȵķβ H(X)ļ㷽£


ͼ2ʵʱĿ걳ʾͼ
1)ͼĵ(x0,y0),ΧƸõһ 11*11 Ĵ.
2)ڵ (x0,y0)Χ 1*13*39*9Ϸֱ Hr(X).
3) H(X)=max(H1(X)H3(X)H9(X))βֲֵͼ(Fractal Parameter Maximum Map FPMM).
4) FPMMľֵ mͷ.
5) t=0.8m+0.2.
6) FPMMͼУĳ FPMMֵtжΪĿ.
7)ڳָУӦСƽֵֵжΪĿλ.
3㷨ʵ
ʵȡһ߶ȷΧڷβֵΪĳڵķֵ.ͼ3һͼ.ͼ4˷βֵͼάͼ.ʵڷβֵĺĿԶʶ㷨ͳƵʶʴ 0.9Ϻõؽ˺СĿԶʶ⣬ΪѰСĿԶⷽṩһ˼·.


(a)۹ⱳͼ󡡡(b)βֵͼ󡡡(c)Ŀ
ͼ3۹ⱳͼĿ
ʽĿȻѧ͹ƹίĿ.
ߵλѧͼ人430074
ο
1Daniel I B,Harvey F S.A class of algorithms for fast digital image registration.IEEE Trans.on computers,1972,C-21(2):179-186.
2Peleg S.Multiple resolution texture analysis and classification.IEEE Trans.on Pattern. Anal.Machine Intell.,1984,PAMI-6(4):518-523.
3Blostein S D,Huang T S.Detection of small moving objects in image sequences using multistage hypothesis testing.Proc.of IEEE Int.Conf.on ASSP,1988.
4Keller J M,Chen S.Texture description and segmentation through fractal geometry.CVGIP,1989,45:150-166.
5Pentland A P.Fractal-based description of natural scenes.IEEE Trans.on Pattern Anal.Machine Intell.,1984,PAMI-6(6):661-674.
6Pell T.Multiscale fractal theory and object characterization.J.Opt.Soc.Am.A,1990,7(6):1402-1412.
ոڡ1994-10-13
