Զѧ
ACTA AUTOMATICA SINICA
1999 25 6 Vol.25 No.6 1999



뷴ϵķֲSFS㷨1)

ժҪһϢԭ״³㷨.㷨Чعϵ춥ԼͼϢԭ״ڿӰӰ£·ͼͳǺͱ淴ϵʹؽǿݶȽӽͼݶʵƽԼ.÷Ϊȶɿͬʱ±б߶ͼСɻڵʣ.SFS(Shape from Shading)㷨ķֲʵ.
ؼʡƣϵֲSFS㷨˹ģͣʣ.
A HIERARCHICAL SFS ALGORITHM BASED ON
ILLUMINANT PARAMETERS AND SURFACE ALBEDO
YANG Jingan
(Institute of AI,Hefei University of Technology, Hefei230009)
AbstractThis paper proposes a robust algorithm to recover shape from shading information. This algorithm can estimate effectively illuminant direction, surface albedo, the azimuth of the illuminant and recover shape using shading information along image contours. The elevation of the illuminant and surface albedo are estimated from image statistics, taking into consideration the effect of self-shading. With the estimated reflectance map parameters, we compute the surface shape using a new procedure that implements the smoothness constraint by making the gradients of reconstructed intensity close to the gradients of the input image. The new algorithm is data driven, stable, and updates surface slope and height maps simultaneously, and significantly reduces the residual errors in irradiance and integral terms. Finaly a hierarchical implementation of the SFS algorithm is presented.
Key wordsEstimation of illuminant parameters, surface albedo, hierarchical SFS algorithm, Lambertian surface model, residual errors
1
ڼӾУ״(SFSShape from Shading)ͨǰͼȽģɱļκٸ֪ģؽ棬ͼͬͼͬͼ.
SFSMITHorn19751.÷ɹҪأٽ˵״ͼȼϵĳģͣڲ֪ͼԭ״Чֵ㷨.
ڳģͣӦص⣺״ͼȼĹϵģͣϾȷعƳģ͵Ĳ.SFSоУģ÷ͼI=R(p,q)ʾʽIΪ﷢ȣRǷͼǸ߶zͼƫ.Ҫоͼɵ˹ģͣ12.ٶzͶӰƽĹᣬzָ˹ģͿд
(1)
ʽЦǺϳϵǿȺͱķϵ.ԾԾȣΪiǣ淨ߺͳԴļнǣλ(x,y,z(x,y))ı淨ߣ34ݣl=(cossinãsinsinãcos)ĵλЦǣlx-zƽнǣҲ춥ǣбǣlzнǣ0ΪƫȣڱתУ׼.зڵĽ϶⣬ϸѧµ㷨ҪؽǿȵݶȽӽͼݶʵƽԼ.
2ʽĵ
ùƵͱϵܹSFS.ͳSFSҪ÷ʽ
R(p,q)=I(x,y)(2)
ؽͼӦӽͼ56ݣҲ˵ʽ(2)ʹͼϢ.ڸֵSFS㷨УؽӦÿشֱʱ.ÿأʽ(2)ұֵߵ(p,q)ɱ.ʽ
(3)
еĶƽʩؽƽʽ.źǣǿα仯ƽȵƱ״仯ͼǿȵı仯״ı仯йأ78.ͼǿȼ仯ӦıƬܲƽģ9ݣڸϵƽͼ.˷ֶƽԼȱһֵַʺھֲǿȵƽ仯.SFSǿӦƽͨؽǿȵݶȵͼݶʵ֣ɱʾΪ
Rx(p,q)=Ix(x,y),(4a)
Ry(p,q)=Iy(x,y),(4b)
R(p,q)xy˶ʽ(4)ߵĵͨϲʽ(2)(4)SFSļɹʽΪʹ
(5)
СԼŻ⣬ʽ
(6)
ʽ(6)һԻԼǼȨϵ.ֵעǣѺԶƽ.ñļ㣬ʽ(5)ļСЧEularʽ
(7a)
(7b)
(7c).
ʽ(6)У̩ռչʽ(p,q)ΧķͼȡֵFpд
(8)
Ʋ()ıʾֵƲ()ıǰֵô
(9)
(p,q,Z)ƫӦΪ
px=px-pqx=qx-q,Zx=Zx-Z(10a)
py=py-pqy=qy-q,Zy=Zy-Z(10b)
pxx=pxx-2pqxx=qxx-2q,Zxx=Zxx-2Z(10c)
pyy=pyy-2pqyy=qyy-2q,Zyy=Zyy-2Z.(10d)
ʽ(9)(10)(8)չͼRɻ
(11)
߽׵Rpp,Rpq,RqqR(p,q)ΧԽƶΪ0.Ƶأʽ(7)еɵ£
(12)
(13)
(14)
(15)
(16)
Fz=0(17)
(18)
(19)
ʽ(12)(19)(7)дɾʽõ

Чд
(20)
ʽ

ʽ(20)õ޸ķ



ʽЦϵʽΪ


ɴ֤ȶ.

3㷨ʵ
㷨÷ֲṹ͵ʵֱ֣.
3.1ֲṹ
ΪͼĸӼ㣬ֲʵЧķ.ǵķֲSFS㷨УֱͼֱѹϵΪ2ͷֱʵͼߴ3264.ֲʵֵҪǰѲĽڲͨţұ״ڲֱͬʲӦһ.()ıϸ߷ֱʲ״û()ıʾͷֱʲת.
1.ϵͬ

ʵһƵͱ淴ϵֱܷʱ仯Ӱ.
2.ϸ߷ֱʲıڵͷֱʲڲ.ǽϸ߷ֱʵͼߴ磬ڣϸ߷ֱʲ״
ijΪż
iΪjΪż
iΪżjΪ(21)

ijΪ.

3.Ȼı߽岹߽أi=1j=1ı߽
(22)
ĽͼȡͲ岹Ľʵֲ֣ʵֵĸӷһо.
3.2
ϷͼơֲʵֺµSFSʽ÷ֲSFS³㷨䲽£
Step1.Ʒͼ(ӣãǣ0)
Step2.ͼı׼ͼߴͷֱʲĳߴ磬p0,q0z0ֵΪ0
Step3.޸е״ؽÿأƫµ

ʽжͼֵƹ3Ȼ߽ʽ޸״ؽ

ʽ

If(Solution is stable) OR (Iteration has reached Nmaxof current layer)continue to Step 4,otherwise repeat;
Step 4. IfCurrent image is in the highest resolutionstop;
Otherwise
ͼߴ磬ʽ(21)(22)չ״ؽڽϸ߷ֱʲ㣻
ѱ׼ͼǰķֱʣ
Go to Step 3.

4ʵ
4.1ϵ
öʵͼ㷨аɫͼ񡢷ֱʲͬͼͼ..r.g͡.bֱʾ졢̺.wʾֲɫļȨͲĺڣͼ.ֵעǣʵͼͼǿȳƫƣڽƬعʹ.ʵõĽ1.
1ʵͼϵ

ͼӦæǦ0
ͼ-1.b10.7954.08127.7341.00
ͼ-1.g7.7459.52192.013.00
ͼ-1.r9.2942.89190.4754.00
ͼ-1.w11.7352.46162.2737.00
-1.b26.0952.77137.460.00
-1.g32.7445.25221.850.00
-1.r22.566.50201.310.00
-1.w20.4024.06173.240.00


4.2SFS㷨ʵ
öͼSFS㷨ʵУ=1,߷ֱʲNmaxΪ500(p,q,Z)=(0,0,0)ʼԷͼƫRpRqֵ.ʵʹеͼΪ256256߶ͼ3-DͼСΪ6464.ͼ1(a)ͼ񣬹ƵķͼΪ=7.74㣬=59.52㣬=192.01ͦ0=3ͼ1(b)ΪSFS㷨õĸ߶(Z)ͼͼ1(c)ͼ1(b)3-Dͼͼ1(d)SFSɵpqͼͼ1(e)ʾ춥Ǳ仯=73.11SFSõ(p,q)ͼͼ.ͼ2SFS㷨ͼ(a)Ϊͼ񣬹ƵķͼΪ=32.74㣬=45.25㣬=221.85ͦ0=0(b)SFS㷨õĸ߶ͼ(c)(b)3-Dͼ(d)ΪùƵķͼSFSõ(p,q)ͼͼ(e)ʾ춥Ǳ仯=122.74SFSõ(p,q)ͼͼ.


ͼ1SFS㷨ͼ
4.3ᡡ
ʮ״Ѿо漰Ҫۡ.ʵϣӦ򡢱淴ϵԼùƵķͼԭ״.ʾ˹ģ͵ķͼؽ״³㷨ͨԸֲͬͼв飬ѾʮĽ.ڱ㷨УҪؽǿȵݶȽӽͼݶʵƽԼ£ܴͼ񵼳Լб߶ͼͬʱʹ㷨ȶ.⣬ӳɱȥСͿɻڵʣؽ߶ͼĶ̬ΧɹؽSFS㷨ֲʵֱָʲͨЧʣʹٶȴӿ죬ȷҲкܴߣΪSFSʵʱӦõ춨˻.


ͼ2SFS㷨ͼ
1) Ȼѧ(6958500269785003)ʿ(96035902)Ŀ.
߼飺1969ҵںϷʹҵѧߵϵΪУϢѧԺڣʿʦIEEE߼ԱŦԼѧԺFellow.ڴ¼Ӿͼ⡢ģʽʶ˹˵Ľѧо.ѧ140ƪר2ɹȻѧ𡢲ʿר𼰹ʺĿ10ʡƼȽڶѧ֯ڼְ.
ߵλϷʹҵѧ˹оϷʡ230009
ο
1Horn B K P. Obtaining shape from shading information. The Psychology of Machine Vision(P H Winston Ed.). New York:McGram-Hill,1975115155.
2Horn B K P. Height and gradient from shading. International Journal of Computer Vision,1990,1(5)584595
3Horn B K P. Robot Vision. Cambridge,MA:MIT Press,1986
4.ں϶ͼȷάȵо.Զѧ199622(1)5867
5Peleg S, Ron G. Nolinear multiresolution:A shape from shading example.IEEE Trans. Pattern Anal. & Mach. Intell,199012(12)12061210
6Pentland A P. Finding the illuminant direction. J. Opt. Soc. Amer.A,198372(4)448455
7Frankot R T, Chellappa R. Estimation of surface height in synthetic aperture images using shape from shading techniques.Artificial Intelligence,199043
8ŵ.Ӿƥ㷨о.ѧ199119(1)3643
9.˫Ŀͼ޶Ӧع3-D˶.ѧ199518(11)849857
ոڡ1997-11-07޸ĸڡ1998-04-21
