PROJECT TITLE :

Text-Attentional Convolutional Neural Network for Scene Text Detection

ABSTRACT:

Recent Deep Learning models have demonstrated strong capabilities for classifying text and non-text parts in natural images. They extract a high-level feature globally computed from a full image element (patch), where the cluttered background information may dominate true text features in the deep representation. This leads to less discriminative power and poorer robustness. During this paper, we gift a new system for scene text detection by proposing a unique text-attentional convolutional neural network (Text-CNN) that significantly focuses on extracting text-related regions and options from the image elements. We have a tendency to develop a replacement learning mechanism to train the Text-CNN with multi-level and rich supervised information, together with text region mask, character label, and binary text/non-text data. The made supervision information enables the Text-CNN with a strong capability for discriminating ambiguous texts, and also will increase its robustness against difficult background parts. The coaching process is formulated as a Multi-Task Learning problem, where low-level supervised information greatly facilitates the most task of text/non-text classification. Additionally, a strong low-level detector known as contrast-enhancement maximally stable extremal regions (MSERs) is developed, that extends the widely used MSERs by enhancing intensity distinction between text patterns and background. This permits it to detect highly difficult text patterns, resulting in an exceedingly higher recall. Our approach achieved promising results on the ICDAR 2013 knowledge set, with an F-live of 0.82, substantially improving the state-of-the-art results.