Hi experts,
given a library setting and a picture of a bookshelf (e.g., something similar to this [picture](https://socalchurchlibraries.files.wordpress.com/2013/08/ecla-socal-2012-fall-13.jpg)), I want to detect the boundaries of the "library-labels" of the different books. All labels have a white background but their size might
differ (as the width of the book varies).
So far, I found some related papers (e.g., [Combining Image and Text Features: A Hybrid Approach to Mobile Book Spine Recognition](http://web.stanford.edu/~bgirod/pdfs/Tsai_ACM_Multimedia_11.pdf)). I've also tried various combinations of preprocessing steps and then line / contour detection. While I was able to detect multiple books in some pictures, I have the feeling that because of the fine-tuning of the parameters of the preprocessing operations I "overfitted" it and it might not work with different light settings etc. Currently, I'm experimenting with an approach that is similar to [square detection](http://stackoverflow.com/questions/10533233/opencv-c-obj-c-advanced-square-detection).
Thus, I would like to ask two questions:
- Based on your experience, would a cascade classifier might work in this setting?
- In more general terms, is there an approach which is preferable in this setting?
Thanks and best regards,
Christoph
↧