Region merging and localization in Binary Image

Take a look at the attached image. it has been processed up till this stage. But its a little difficult and complex algorithm trying to extract the relevant area (number plate region) using the following steps. 1. label connected components 2. Find centroid of each region 2. if centroid of region along x-direction (horizontal direction)are equalto or less than 10 pixels along the x-direction (horizontal direction), MERGE. 3. Boundbox of region larger than threshold T

Any other approach of localizing the number region of the image above would be appreciated.

[Walter Roberson]

http://www.mathworks.co.uk/matlabcentral/answers/43073-a-guide-to-tags

Answer 1

[Image Analyst]

See https://www.mathworks.com/help/vision/examples/automatically-detect-and-recognize-text-in-natural-images.html

[akpojotor princewill]

MSER is quite complex and resource intensive... would love a simple but yet effective technique that extracts character blobs in binary image.

[akpojotor princewill]

Thank you Image analyst for the link. Though it took a while i was able to extract information I needed from it to tune my algorithm. However, am facing a challenge achieving my intended result.

The summary of the algorithm is to input a binary image with license plate text characters.

 1. Identify and label connected components (as expected, this includes other invalid connected components)

2. compute aspect ratio from bounding box of connected components

3. filter components that does not meet criteria based on text character aspect ratio.

4. fill bounding box with valid aspect with white pixels

5. morphological dilate the image in order to get text character region connected, forming a larger component.

6. identify and label the image with new connected component

7. compute aspect ratio of bounding boxes and filter components that do not meet license plate aspect ratio criteria.

Attached is the input binary image and a snippet of what i have done and where am finding difficult. Your contributions and suggestions are highly appreciated. thank you

[Image Analyst]

You can ask regionprops to get the bounding box, though it won't be tilted. You can use propfilt() to filter out blobs not meeting criteria, or do it "manually" like I do in my Image Segmentation Tutorial: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862&sort=downloads_desc