Automatic script identification in the wild

Abstract

With the rapid increase of transnational communication and cooperation, people frequently encounter multilingual scenarios in various situations. In this paper, we are concerned with a relatively new problem: script identification at word or line levels in natural scenes. A large-scale dataset with a great quantity of natural images and 10 types of widely-used languages is constructed and released. In allusion to the challenges in script identification in real-world scenarios, a deep learning based algorithm is proposed. The experiments on the proposed dataset demonstrate that our algorithm achieves superior performance, compared with conventional image classification methods, such as the original CNN architecture and LLC. 1

A Fine-Grained Approach to Scene Text Script Identification

“…A much more recent approach to scene text script identification is provided by Shi et al [4] where the authors propose the Multi-stage Spatially-sensitive Pooling Network (MSPN). The MSPN network overcomes the limitation of having a fixed size input in traditional Convolutional Neural Networks by pooling along each row of the intermediate layers' outputs by taking the maximum (or average) value in each row.…”

“…provided by the user, or inferred from available meta-data. The unconstrained text understanding problem for large collections of images from unknown sources has not been considered up to very recently [4]. While there exists some research in script identification of text over complex backgrounds [5], [6], such methods have been so far limited to video overlaid-text, which presents in general different challenges than scene text.…”

A Simple and Effective Solution for Script Identification in the Wild

“…Most of these datasets are used for scene text localization and recognition in English. There are also few datasets [8,1] of multiple scripts e.g., east Asian languages or Indian language video text. In this work we introduce Indian Language Scene Text (ILST) dataset which is a comprehensive dataset for Indian language scene text containing six scripts commonly used in India, namely Telugu, Tamil, Malayalam, Kannada, Hindi and English.…”

“…There are many methods in the literature for script identification [1,2,5,6,7,8]. Texture based features such as Gabor filter [7], LBP [9] have been used for script identification.…”

“…We compare our method with the entries for this competition and show that our method is comparable to the top performing methods in this competition. There have been few contemporary methods based on deep learning [1,8] and RNN [3]. These methods achieve noticeable success on some of the selected benchmarks.…”