Plate number localization or segmentation

Sample image need to go through segmentation process:

This is the progress of the project right now. In order to start segmentize the plate number, we have to do some deep study and analysis on how the plate number is different from a non-plate number. From the knowledge about the plate number characteristic, it make us easier to start writing the rules or I call it as "Rule Based Function". Before I explain further, need to be cleared by any readers that I am using Matlab proggramming that obviously uses Image Processing Tools in it.

All the program is written in the ".m" file with their suitable names according to their fucntion. Lets now jump to the topic as I don't want to drag further. First of all, we need to convert the color image to grayscale, and from gray scale image to black and white image and eventually start calling rule based function to do the job. There is a few function which I have split them out. It does not means anything to the expert programmer, but for the sake of simplicity and troubleshooting easiness it is suggested so.

The first stage of segmentation:

After the image has been converted to be a black and white image using a global thresholding technique, it is then being processed and the detected object will be boxed as shown in the figure above. On the left hand side, there is a text showing the background process. Don't worry, I am purposely print it out for me to see the process behind it by putting command "disp('your message')" if you are interested to know. Next, it will go through a few rule based function that I called compact check, angle check and HV check.

Compact Check:

This is a function to check the object compactness based on some formula inside it. I has a great posibility to removed unwanted object that is really obvious not a plate number. However this function alone does not enough to eliminate all the non-plate number object. Thus, the next function named "angle check" will be used to help him.

Angle Check:

What angle check does?. It simply measure the angle of every object as well as calculate the distance between them. If it is too far each other, there is a high posibility that the object not belong to a plate number. Bare in mind that angle check is also not good enough to removed all because the angle and the distance is very similar to the plate number's characteristic. Therefore we have to really focus on the character's characteristic to filter that out.

HV Check:

This is the final function so far, in it there is a few more verification being done. The way its work is that, it plot horizontal (H) and vertical (V) of the cropped object and do some analysis on it. H and V implies to the function's name HV check. After this check, we can get the result as shown below:

The green boxed with a dotted line is the detected plate number. Next process would be in the next post. bare with me..hehehe bye.