I am working in License plate detection and I want to compare my work to other license detection software and I will use this comparison as a reference in my paper.
This might be too tangential, but I saw this article yesterday on Google uses training an 11 level deep NN to recognize house numbers up to five digits all at once without breaking the image into different digits. I assume it could be used for license plates as well and the linked paper might have useful references. http://www.technologyreview.com/view/523326/how-google-cracked-house-number-identification-in-street-view/