CNN Number Detection
A basic CNN model was generated using Keras/Tensorflow in Python, to detect house numbers 0 to 9. Training included samples of numbers and non-numbers.
This NumPy pipeline for house number detection involved blob detection, number region localization, shape correction, character size constraints, size conversion, number detection with trained CNN model and non-maximal suppression.
The outputs show the number region detected in green boxes and the number itself is provided in red below each green box. These outputs show that the CNN model works at different scales, orientation and illumination.