18 application example photo ocr
- 1. 18. Application Example – Photo OCR: The Photo OCR Problem Machine Learning Pipeline: A system with many stages/components, several of which may use machine learning.
- 2. Sliding Window: PEDESTRIAN DETECTION
- 3. To detect the pedestrians, small frames are allowed to scan the whole image. It is tried with many frames of different sizes and all the captured images are resized to a particular size and then that image is sent to Neural network to determine if there is a pedestrian or not SLIDING WINDOW DETECTION
- 4. For text detection example:
- 5. The white regions show where the text is detected The grey regions show where there is a probability of text. The algo has lower confidence in those parts ➢In expansion (on the right), we ask if a pixel is in within 5 pixels of a white pixel, then that pixel is also made white pixel Next we filter to only those white boxes, where the aspect ratio is likely to be suitable for text We now cut out these regions from the image and use them in later stages of detection
- 6. Artificial Data Synthesis: to amplify the training set: Synthetic data is prepared by using different fonts and putting letters on different backgrounds
- 8. Ceiling Analysis: What part of the pipeline to work on next?
- 10. Machine Learning by Stanford University on Coursera. Certificate earned at Friday, April 12, 2019 9:43 AM GMT coursera.org/verify/4VW5AT4B38TZ