Introduction/Background: 

Pen to Print is the first handwriting to text OCR app converting scanned handwritten notes into digital text available for edits, search and storage in any digital platform.  This app will take in live photos or photos from one’s camera roll.  The scanner will align segment sections and scan each line of the handwritten text individually in order to provide a given result.

The purpose of this lab is to test how the quality of photos affects the accuracy of Modern Day OCR’s, in this case pen to print, accuracy and to determine what factors are the most determinant to the quality of each image.

Each participant will be tested on how the photo angle, photo brightness, and stroke of the image affects accuracy. Any discrepancies will found compared to the original prompt will be recorded along with any following observations.

Materials:

• Pen to Print App
• Stand/Selfie Stick
• Phone/Camera
• Protractor
• Loose Leaf Paper

Factors:

• Neatness of Handwriting

• Photo Angle
  • 0 degree flat angle
  • 45 degrees from the right
  • 45 degrees from the bottom

• Photo Brightness
  • Standard Brightness
  • -100 Brightness
  • +100 Brightness

• Stroke of Handwriting
  • Pen vs. Pencil

Data Results:

Angle – Total Errors

• 0 degrees (flat): 4
• 45 degrees (left): 10
• 45 degrees (right): 12
• 45 degrees (from bottom): 5

Brightness – Total Errors

• 0 brightness: 4
• +100 brightness: 5
• -100 brightness: 8

Stroke – Total Errors

• Pen: 4
• Pencil: 6

Analysis:

Some causes for lower accuracy rates were low lighting, angling the photo to the left and right, and using pencil rather than pen. For the angle factor, providing an image that was tilted 45 degrees to the left or right produced more errors than the photos that were taken at a 0 degree angle above the page and a 45 degree angle from the bottom. For brightness, a decrease in brightness created the most errors. For stroke, the results from the document written in pencil were slightly less accurate than the results from the document written in pen. Overall, the program did not produce a large number of errors.

Conclusion:

Factors such as the image’s angle, brightness, and writing utensil used can impact the accuracy of the OCR results. OCR programs use processes to provide users with accurate results. They must refine the image and use binarization in order to allow the algorithms to accurately read the text that was provided. Therefore, although it is generally important to provide OCR software with high quality images, modern day OCR technology is capable of accurately deciphering a variety of image qualities.