Test Cases & Results
The intensive and Exhaustive tests have helped us to improve the OCR system supported on the android G1 phone. The test done were as under
| IMAGE | BINARIZED IMAGE | ROTATED IMAGE | OCR RESULTS | AIM | Notes | |
|---|---|---|---|---|---|---|
 |
 |
 |
OCR Recognition Error | Effect of Perception Distortion | Partially solved by detectin tilt on the mobile phones using the Orientation Sensors at the front end . |
|
 |
 |
 |
OCR Recognition Error | Effect of Blurness arising due to lack of focus while taking images | Solved by using the autofocus support provided by the GI camera and also tuning some of the parameters in the binarization algorithm used at the back end |
|
 |
Output | Effect of rotation by approx 10 degrees on a purely binarized image | OCR produced extremely poor results |
|||
| |
 |
Output | To show improvement in OCR results induced by auto rotatio | Solved problem of skew by using branch-and-bound text line finding algorithm.
|
||
 |
Output | To show degradtion in OCR output on images with more than 35 degree rotation | Limitation of the Auto rotation algorithm | |||
 |
 |
 |
Output | Rotation by 90 degrees | Algorithm handles rotations less than 40 degrees but fails in such scenerios and hence it depends on how the user actually clicks the photo. |
|
 |
 |
 |
Output | Rotated image | One more example for rotated image and poor lightening |
|
 |
Output | Aim to show result of poor lightening without binarization on the OCR output | Problem can be solved by using good binarization algorithm | |||
 |
Output | Aim to show result of Sauvola Binarization on the OCR output | Almost 100% accuracy by using sauvola binarization algorithm | |||
|
 |
As Image is not visible, the algorithm gives error | OCR Recognition Error | Aim to show that even Sauvola Algorithm fails in case of extremely Poor Lightening | Solution : Change some parameters used in Sauvola algorithm |
|
 |
 |
 |
Output | Uneven Lightening | This factor was reduced at the backend while performing the binarization. By providing proper parameters this problem was reduced drastically. |
|
 |
 |
 |
Output | To show result of pre processing steps on OCR output in case of poor lightening and rotated images | OCR result is quite good, |
|
 |
 |
 |
Output | Poor Lightening and slight blurness | Solved by tuning the autofocus parameter of the camera and also tuning the binarization algorithm at the server end. |
|
 |
 |
 |
Output | More uneven Lightening | ||
 |
 |
 |
Output | Different Patches of Lightening with Bluring | ||
| Some Tests Performed on the Real Receipts | ||||||
 |
 |
 |
Output | |||
 |
 |
 |
Output | |||
 |
 |
 |
Output | |||
 |
 |
 |
Output | |||
 |
 |
 |
Output | |||
The Rest of the test cases with different lightening conditions is availabe for download. Please click HERE to download
