How to detect specific shaped objects with any different orientations in Image processing ?

Hello, I want to know that how to detect the objects with different orientations using Area ratio, in which the problem is the bounding box is not oriented with the same orientation as the object, due to that the area ratio is not accurate and object with different orientation is not detected. How to rotate the bounding box also with the same angle as the object orientation....?

[KSSV]

Attach image and tell your expectations.

[SACHIN PATEL]

I am not talking about specific image, but the problem is the bounding box is not oriented as the object orientation, due to which the area of oriented shape bounding box is larger. And the area ratio of Bounding box to Original object area is not the required for detection of object in image, So I thought if there is any means to create the bounding box that can be oriented also according to object orientation, then possibility of oriented object can be detected. So please help me that how can I create the bounding box with orientation of same as object orientation ?

Answer 1

[KSSV]

YOu can rotate the bounding box by using your desired angle as below:

% bounding box coordinates 
B = [ 0 0 ; 1 0 ; 1 1 ;0 1 ; 0 0]' ;
th = pi/4 ;
R = [cos(th) -sin(th) ; sin(th) cos(th)] ;
% Rotate by th 
B1 = R*(B-mean(B,2))+mean(B,2) ;
figure
hold on
patch(B(1,:),B(2,:),'r')
patch(B1(1,:),B1(2,:),'k')

[SACHIN PATEL]

Thank you sir, The rotation concept is working properly for above code. But when I use regionprops function of matlab for find bounding box, how I can find the bounding box inclined itself and surrounding the object in the image. Because, the bounding box should be of appropriate size surrounding the object. If the bounding box is detected first surrounding the object, and then rotated, then the size of bounding box will not be appropriate. But I got somewhat logic and idea from your answer to rotate Bounding Box. Thank you for your time and help. Also sir I want to know the logic behind subtracting mean from B Array. and after rotation adding mean to it.

Thank you.

Answer 2

[Image Analyst]

Take a look at Hu's moments which are rotationally invariant: Click here

[SACHIN PATEL]

Hello Sir, I have gone through the video link, but I have not complicated logoes to find out, but My task is to find out the standard shapes (like square, hexagon etc.), but the problem is if for example the square is inclined at 45 degree it can't detected using Area ratio (Ratio of bounding box area to that of original object area using regionprops); because the bounding box is not oriented as 45 degree. Isn't there any easy way other than a hu's moments ?; because we have not to detect complex logo or objects but they are of the simple standard shapes.

Thanking you.

[Image Analyst]

See my shape recognition demo, attached. It puts up simple polygons of various numbers of vertices, angles, and sizes, and then finds them.

[SACHIN PATEL]

ok sir, thank you for your support. I will try to implement that.