To detect an object in an image we first study its general characteristics and how it is different from other objects within the image, general characteristics of a plate number include and not limited to:

1- plates have high contrast foreground/back ground, this is designed for humans to be able to read  easily, which is a bless for a computer vision problem.

2- the shape of the plate is horizontal rectangle, the aspect ration and the size of this rectangle is standard within one country.

3- the location is normally in the bottom and middle of the car.

4- some other local characteristics, i.e. country dependent.

it is an easy task now to develop your filters and image pre-process to detect the plate.

P.S. the car plate in the second image is almost impossible to be recognized because of the camera angle.

The following posts at stackoverflow may be useful for you.

http://stackoverflow.com/questions/17987866/how-to-extract-and-recognize-the-vehicle-plate-number-with-matlab

http://stackoverflow.com/questions/24647074/how-to-remove-lights-deflections

http://stackoverflow.com/questions/4707607/what-are-good-algorithms-for-vehicle-license-plate-detection

To detect an object in an image we first study its general characteristics and how it is different from other objects within the image, general characteristics of a plate number include and not limited to:

1- plates have high contrast foreground/back ground, this is designed for humans to be able to read  easily, which is a bless for a computer vision problem.

2- the shape of the plate is horizontal rectangle, the aspect ration and the size of this rectangle is standard within one country.

3- the location is normally in the bottom and middle of the car.

4- some other local characteristics, i.e. country dependent.

it is an easy task now to develop your filters and image pre-process to detect the plate.

P.S. the car plate in the second image is almost impossible to be recognized because of the camera angle.

We have did a project on numberplate recognition on Android: http://dl.acm.org/citation.cfm?id=2558037

If you are interesting in, feel free to checkout the code: https://github.com/donnguyen/UIT-ANPR 

Cheers

Dears

First you must know that you are interesting in static or dynamic CLPN. Secondly you must fixed your camera on an  adequate distance.  After you cached the photo to the front of the car, then you remove back ground, then localize, separate and segmented the  characters and lastly comparing these characters with the ideal ones.  

Hello

I found this paper to be quite helpful when starting with my work on ALPR as this summarizes the various techniques which are available for doing LPR. 

http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6213519

First we can separate the region of interest by image croping.

then some image enhancement techniques may be applied. If the image (number) is not clear to see , then do some edge enhancement.

Use the suitable image segmentation method, then use the classification and AI method for recognizing numbers

The following posts hopefully be useful for you.

https://blog.devcenter.co/developing-a-license-plate-recognition-system-with-machine-learning-in-python-787833569ccd