图像拼接matlab程序

INRIA%Visual%Recognition%and%Machine%Learning%Summer%School%

Instance9level%recognition:%Practical%session%

!

Stitching!photo!mosaics!

Cordelia%Schmid%and%Josef%Sivic%(adapted%from%Andrew%Zisserman)%

!

!

The!goal!of!the!practical!session!is!to!automatically!stitch!images!acquired!by!a!

1.!Choose!one!image!as!the!reference!frame.!

2.!Estimate!homography!between!each!of!the!remaining!images!and!the!

reference!image.!To!estimate!homography!between!two!images!use!the!following!

procedure:!

a.!Detect!local!features!in!each!image.!

b.!Extract!feature!descriptor!for!each!feature!point.!

c.!Match!feature!descriptors!between!two!images.!!

d.!Robustly!estimate!homography!using!RANSAC.!

3.!!!!!!Warp!each!image!into!the!reference!frame!and!composite!warped!images!

into!a!single!mosaic.!

!

2. Open!file!mosaic.m!in!the!Matlab!text!editor.!Follow!the!instructions!below!by!

executing!the!provided!code!(also!given!in!mosaic.m).!Note!that!the!code!is!

partial!and!you!will!have!to!complete!it.!

!

3. Read!in!the!three!images!of!Keble!college!and!look!at!them!!

imargb = double(imread('keble_a.jpg'))/255;

imbrgb = double(imread('keble_b.jpg'))/255;

imcrgb = double(imread('keble_c.jpg'))/255;

These!images!are!acquired!by!a!panning!camera!and,!as!you!will!recall!from!the!

lecture,!this!means!that!the!images!are!related!by!a!planar!homography.!!

“harris.m”.!!Is!the!choice!of!single!scale!features!(vs.!multi‐scale!features!such!

as!Harris‐Laplace)!reasonable!for!this!example?!

topn = ??; % how many Harris corners points?

[xa,ya,strengtha] = harris(ima,topn);

[xb,yb,strengthb] = harris(imb,topn);

[xc,yc,strengthc] = harris(imc,topn);

Use!the!provided!code!to!display!the!detected!features.!

5. Extract!feature!descriptor!for!each!feature!point.!Descriptors!are!simple!

21x21!grayscale!pixel!patches!extracted!from!an!image!blurred!by!a!(fairly!

large)!Gaussian!(sigma=3!pixels).!The!blurring!is!done!to!obtain!some!

tolerance!to!feature!localization!errors.!This!is!a!very!simple!version!of!the!

descriptor!used!in!the!paper!“!"#$%&'()*+,!)$-.%/*,"0%/*,!"#$%&1-)#+,

23%+/$+4,5)$-.+06,78,93:;/,+$,)#<=,>?5@ABC<,!

[Da,xa,ya] = ext_desc(ima,xa,ya);

[Db,xb,yb] = ext_desc(imb,xb,yb);

similar!(have!similar!descriptors)!and!are!thus!likely!to!be!in!

correspondence.!

Db = dist2(Da',Db'); % compute pair-wise distances between descriptors

[Y,I] = sort(Db,2); % sort distances

rr = Y(:,1)./Y(:,2); % compute D. Lowes' 1nn/2nn ratio test

inDa2 = find(rr<.8); % take only points with a 1nn/2nn ratio below 0.8

I = I(inDa2); % select matched points

xat = xa(inDa2);

yat = ya(inDa2);