Image understanding: Semantic Segmentation of Graphics and Text using Faster-RCNN

This paper presents a Faster Regional Convolution Neural Network (FRCNN) model, capable of detecting and segmenting text and graphics from road sign boards for automatic self-driving vehicles. This proposed model also detects and segments logos from natural scene images, newspaper images, ID cards for image understanding. The designed and implemented model can detect 12 classes of logos and text from a standard sign board having green colored background with white color foreground text. For better accuracy and optimized results focused images with good quality text data information along with sufficient size is considered. The model detects text from road sign boards and logos accurately. The Algorithm also detects and segments text in newspaper, in which there is a no clear differentiation between text and background. This network is able to achieve 96.5% accuracy on text classification and 92.7% accuracy on logos classification. The results obtained from our module are promising and impressive than the other methods.