Separate bounding boxes for parts same character

[uvindub]

I got separate bounding boxes for parts of same character

how can i overcome this issue

[TomMakesThings]

Optical character recognition is a challenging problem and the code was written with the aim of playing around with simple techniques rather than being a top-of-the-range classifier. At a guess I'd say the lower characters are being separated into different boxes because of the slight white space between them after thresholding. Several ways to overcome this could include:

• Increasing the amount of blur in the x-axis and lowering the threshold
  • This way the characters would appear thicker in the 'Otsu Adaptive Thresholding' step
• Merging bounding boxes if they are very close together
• Adding padding to the edge of the bounding boxes

I hope this answers your question and that with some experimentation you are able to improve the performance of your classifier!