ANPR detection using opencv c++

edit

I look at the tutorial in chp 5 (Licence Plate Detection) provided in the book of Packtpub Mastering opencv with practical computer vision projects , i want to know some points on it :

1 - Why we need to focus on aspect ratio much if we already doing SVM ?

2 - Is it compulsory that we should know/provide the size of licence plate , while we are using SVM ?

3 - Why our focus is on the width and height of numbers and gap between them ?

Do the small difference between different pictures , like the sizes of number plates and difference between alphanumeric words and gap between them matters much ?