Consider if we have taken an image from a side view but want to make a perspective of that image.
If we captured a photograph of an inclined number plate, but for character recognition purpose i want to front view of that number plate, in this situation I want to perspective of that number plate.
Why dont you try a recoginition approach that is orientation/perspetive independent. Frequency domain recognition and analysis may help a lot in this.
Image registration may be your answer. but in the case of plate recognition the best choice is using affine invariant moments, Zernike moments, scaling invariant moments,...
first of all you have to separate the object that you want to process blindly, for example a character "2" from the image and then you can use affine invariant moments to recognize it whether it is a side view image of "2" or not.
for blindly separating the objects from the image you can use morphology (well explained in MATLAB) and then you may send the separated object to a affine moment calculator and then decide if the object is for example "2" or not.
@Shrikant
The use of "perspective" is a bit vague in your message. From the computer vision point of view, "perspective" transformation is just the projection mapping scene to the image plane (whatever it is inclined, frontal of whatever the viewing direction) as long as the camera is close to a pinhole model.
For your problem, the simplest solution is to estimate an homography mapping your side view to your frontal view. For that, I assume you have a clear view of your plate edge, let's say the its four corners c_i, with i \in {1,2,3,4}.
Now specify arbitrarily the frontal position of the four corner (let's note them C1,C2,C3,C4). apply a DLT to estimate H such that C_i= Hc_i.
note: H is a 3x3 matrix and c_i,C_i should be the 2d positions expressed in homogenous coordinate.
Have fun.
B. RADI, A. EL HAMI, Mathématique numérique pour l’ingénieur, Utilisation de MATLAB: Cours, exercices et problèmes de synthèse corrigés. Editions-Ellipses, Numéro ISBN 978-2-7298-5279-5, Novembre 2010 (240 pages, niveau second cycle)
Book: Advanced Numerical Methods with Matlab® 2: Resolution of Nonlinear, Differential and Partial Differential Equations
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing