The current Process() implementation is not doing any edge detection and detecting text is only based on black/white level with a threshold. This means, areas like for example the center of the book, where there is a shadow from the other side, appear as "black" and contribute during the contouring step when we detect text.
Edge detection algorighms do a better job detecting the text because the shadow appear as a "gradient" and it not picked up.
We should try edge detection as an additional step to improve text detection.
The current Process() implementation is not doing any edge detection and detecting text is only based on black/white level with a threshold. This means, areas like for example the center of the book, where there is a shadow from the other side, appear as "black" and contribute during the contouring step when we detect text.
Edge detection algorighms do a better job detecting the text because the shadow appear as a "gradient" and it not picked up.
We should try edge detection as an additional step to improve text detection.