Text Detection Using Adjacency Relationship in Connected Components

This undertaking, present a new scene text detection algorithm established on two machine discovering classifiers: one permits us to produce candidate word spans and the other filters out non text ones. To be precise, they remove related constituents (CCs) in images by employing the maximally stable external span algorithm. These extracted CCs are partitioned into clusters so that they can produce candidate regions. Train an AdaBoost classifier that determines the adjacency connection and cluster CCs by using their pair wise relations. Then normalize candidate word spans and determine whether every single span encompasses text or not. As the scale, skew, and color of each candidate can be approximated from CCs, develop a text or non text classifier for normalized images. This classifier is based on multilayer perceptions and we can control recall and precision rates alongside a single free parameter.