Resource-Aware Scene Text Recognition Using Learned Features, Quantization, and Contour-Based Character Extraction.