OCR using Tesseract
- 1. Real time OCR using Tesseract 12BCE094 SHOBHIT CHITTORA
- 2. Brief History Of Tesseract Open Source OCR engine sponsored by Google since 2006. One of the most accurate open source OCR engines currently available. Originally developed by HP between 1985-1994. Lot of it is written in C and C++.
- 4. Adaptive Thresholding is Essential
- 5. Baselines are rarely perfectly straight
- 6. Spaces between words are tricky too Italics, digits, punctuation all create special-case font-dependent spacing. Fully justified text in narrow columns can have vastly varying spacing on different lines.
- 8. Outline Approximation Polygonal approximation is a double-edged sword. Noise and some pertinent information are both lost.
- 9. Why it’s called Tesseract? Elements of the polygonal approximation, clustered within a character/font combination. x, y position, direction, and length (as a multiple of feature length)
- 10. Character Classifier (Features and Matching) Static classifier uses outline fragments as features. Broken characters are easily recognizable by a small->large matching process in classifier. (This is slow.) Adaptive classifier uses the same technique!
- 11. Classifier as Histogram of Gradients Quantize character area. Compute gradients within. Histograms of gradients map to fixed dimension feature vector.
- 12. Character Segmentation Segmentation Graphs
- 14. Rating and Certainty Rating = Distance * Outline length ○ Total rating over a word (or line if you prefer) is normalized ○ Different length transcriptions are fairly comparable Certainty = -20 * Distance ○ Measures the absolute classification confidence ○ Surrogate for log probability and is used to decide what needs more work.
- 16. Implementation using Tess-two( Tess port for Android) The Tess-two library is an open source port of Tesseract engine for Android. Only the most basic and popular functionalities are ported. Things such as deep neutral nets are not ported. A lot of tweaking is required to produce desired results.
- 17. DEMO
- 18. Implementing Real Time OCR and challenges Image processing on memory limited devices is difficult. Limited clock speeds to process huge matrices. Running the Camera Surface Holder in MainUI and preprocessing and OCR on user threads. Maintaining huge Bitmaps for preprocessing and sending to multiple threads. Avoiding Garbage Collection of important preprocessed data.
- 19. Thank You