EDBV project for indexing books with unique labels in library shelves.
The program was tested on Matlab 9.3 R2017b and should work with all newer versions.
Additionally a dataset or image of choice must be provided. The following criteria must be met for the program to work properly:
- The image must display a section of a bookshelf
- The bookshelf must contain books, which are labeled in the same matter as the TU Wien library books.
- The shelf must not be tilted more than +-30° in the image
- The books must not be tilted more than +-5° (vertically) in the image
- The resolution of the image must be high
To use the program, change the image path in "main.m" to match the image you want to be processed and run the program. You can use two strategies for the OCR, nncor2 and ssd.
The program will output a .json file containing the results of the optical character recognition applied on each label. The labels are indexed with 1 to n, whereas n is the number of labels/books displayed in the image.
Anand Eichner, Laurenz Fiala, Aleksandar Vucenovic