An Efficient Real-Time Text Spotter

Abstract

Scene text spotting is the method to perform detection and recognition task sequentially. The spotting technique required a costlier CPU and GPU to perform it sequentially. At present, development in vehicle focus on the self-decision vehicle which is known as an autonomous vehicle. Autonomous vehicles enable vehicles to take decisions regarding driving instruction. Last decades witnessed improvement in CPU and GPU power. Still, scene text requires more CPU and GPU power compared to the current developed system. To solve computational problem for scene text spotting, in this paper we proposed, An Efficient Real-Time Text Spotter which is a customized yolov3 network with a novel backbone network that identifies as a Light Block Network, which aims to perform spotting task in real-time. Efficient Text Spotter is an end-to-end light-weighted deep neural network. It is made up of two parts, the first part is Lbn which is working as a backbone to ERTTS network, and the second part is a detection network that detects text instances from natural scenes and predicts classes based on threshold value. Using An Efficient Real-Time Text Spotter any autonomous vehicle which has a low-cost CPU is perform spotting task in real-time. An ERTTS is enable to perform scene text spotting task in real-time.