Detecting arbitrarily oriented text in natural images is still a challenging and unsolved problem in multimedia. In this work, we propose an efficient and accurate scene text detector. The detector first detects key points which are carefully designed and semantically meaningful. Then the key points are learnt to be associated together to form a hexagon for each text instance. Starting from a key point, the detector then predicts curves alongside the border of the text region. Simple heuristic post-processing followed by the predicted curve resulting in more accurate text region prediction. The predicted key points are used as anchors to correct errors from the search process. The proposed method is efficient since it is a single stage key point detection method with simple post-processing. It is also effective and shows state-of-the-art or comparable performance on several benchmark datasets.

Community-based question answering (CQA) websites represent an important source of information. As a result, the problem of matching the most valuable answers to their corresponding questions has become an increasingly popular research topic. We frame this task as a binary (relevant/irrelevant) classification problem, and present an adversarial training framework to alleviate label imbalance issue. We employ a generative model to iteratively sample a subset of challenging negative samples to fool our classification model. Both models are alternatively optimized using REINFORCE algorithm. The proposed method is completely different from previous ones, where negative samples in training set are directly used or uniformly down-sampled. Further, we propose using Multi-scale Matching which explicitly inspects the correlation between words and ngrams of different levels of granularity. We evaluate the proposed method on SemEval 2016 and SemEval 2017 datasets and achieves state-of-the-art or similar performance.

We study the problem of extracting text instance contour information from images and use it to assist scene text detection. We propose a novel and effective framework for this and experimentally demonstrate that: (1) A CNN that can be effectively used to extract instance-level text contour from natural images. (2) The extracted contour information can be used for better scene text detection. We propose two ways for learning the contour task together with the scene text detection: (1) as an auxiliary task and (2) as multi-task cascade. Extensive experiments with different benchmark datasets demonstrate that both designs improve the performance of a state-of-the-art scene text detector and that a multi-task cascade design achieves the best performance.

Email continues to be one of the most important means of online communication. People spend a significant amount of time sending, reading, searching and responding to email in order to manage tasks, exchange information, etc. In this paper, we focus on information exchange over enterprise email in the form of questions and answers. We study a large scale publicly available email dataset to characterize information exchange via questions and answers in enterprise email. We augment our analysis with a survey to gain insights on the types of questions exchanged, when and how do people get back to them and whether this behavior is adequately supported by existing email management and search functionality. We leverage this understanding to define the task of extracting question/answer pairs from threaded email conversations. We propose a neural network based approach that matches the question to the answer considering comparisons at different levels of granularity. We also show that we can improve the performance by leveraging external data of question and answer pairs. We test our approach using a manually labeled email data collected using a crowd-sourcing annotation study. Our findings have implications for designing email clients and intelligent agents that support question answering and information lookup in email.

Both politicians and citizens are increasingly embracing social media as a means to disseminate information and comment on various topics, particularly during significant political events, such as elections. Such commentary during elections is also of interest to social scientists and pollsters. To facilitate the study of social media during elections, there is a need to automatically identify posts that are topically related to those elections. However, current studies have focused on elections within English-speaking regions, and hence the resultant election content classifiers are only applicable for elections in countries where the predominant language is English. On the other hand, as social media is becoming more prevalent worldwide, there is an increasing need for election classifiers that can be generalised across different languages, without building a training dataset for each election. In this paper, based upon transfer learning, we study the development of effective and reusable election classifiers for use on social media across multiple languages. We combine transfer learning with different classifiers such as Support Vector Machines (SVM) and state-of-the-art Convolutional Neural Networks (CNN), which make use of word embedding representations for each social media post. We generalise the learned classifier models for cross-language classification by using a linear translation approach to map the word embedding vectors from one language into another. Experiments conducted over two election datasets in different languages show that without using any training data from the target language, linear translations outperform a classical transfer learning approach, namely Transfer Component Analysis (TCA), by 80% in recall and 25% in F1 measure.

Scene text detection has attracted great attention these years. Text potentially exist in a wide variety of images or videos and play an important role in understanding the scene. In this paper, we present a novel text detection algorithm which is composed of two cascaded steps: (1) a multi-scale fully convolutional neural network (FCN) is proposed to extract text block regions, (2) a novel instance (word or line) aware segmentation is designed to further remove false positives and obtain word instances. The proposed algorithm can accurately localize word or text line in arbitrary orientations, including curved text lines which cannot be handled in a lot of other frameworks. Our algorithm achieved state-of-the-art performance in ICDAR 2013 (IC13), ICDAR 2015 (IC15) and CUTE80 and Street View Text (SVT) benchmark datasets.

We present an end-to-end, multimodal, fully convolutional network for extracting semantic structures from document images. We consider document semantic structure extraction as a pixel-wise segmentation task, and propose a unified model that classifies pixels based not only on their visual appearance, as in the traditional page segmentation task, but also on the content of underlying text. Moreover, we propose an efficient synthetic document generation process that we use to generate pretraining data for our network. Once the network is trained on a large set of synthetic documents, we fine-tune the network on unlabeled real documents using a semi-supervised approach. We systematically study the optimum network architecture and show that both our multimodal approach and the synthetic data pretraining significantly boost the performance.

We present an iterative refinement module that can be applied to the output feature maps of any existing convolutional neural networks in order to further improve classification accuracy. The proposed module, implemented by an attention-based recurrent neural network, can iteratively use its previous predictions to update attention and thereafter refine current predictions. In this way, the model is able to focus on a sub-region of input images to distinguish visually similar characters (see Figure 1 for an example). We evaluate its effectiveness on handwritten Chinese character recognition (HCCR) task and observe significant performance gain. HCCR task is challenging due to large number of classes and small differences between certain characters. To overcome these difficulties, we further propose a novel convolutional architecture that utilizes both low-level visual cues and high-level structural information. Together with the proposed iterative refinement module, our approach achieves an accuracy of 97.37%, outperforming previous methods that use raw images as input on ICDAR-2013 dataset [1].

We present a robust end-to-end neural-based model to attentively recognize text in natural images. Particularly, we focus on accurately identifying irregular (perspectively distorted or curved) text, which has not been well addressed in the previous literature. Previous research on text reading often works with regular (horizontal and frontal) text and does not adequately generalize to processing text with perspective distortion or curving effects. Our work proposes to overcome this difficulty by introducing two learning components: (1) an auxiliary dense character detection task that helps to learn text specific visual patterns, (2) an alignment loss that provides guidance to the training of an attention model. We show with experiments that these two components are crucial for achieving fast convergence and high classification accuracy for irregular text recognition. Our model outperforms previous work on two irregular-text datasets: SVT-Perspective and CUTE80, and is also highly-competitive on several regular-text datasets containing primarily horizontal and frontal text.

Distractor generation is a crucial step for fill-in-the-blank question generation. We propose a generative model learned from training generative adversarial nets (GANs) to create useful distractors. Our method utilizes only context information and does not use the correct answer, which is completely different from previous Ontology-based or similarity-based approaches. Trained on the Wikipedia corpus, the proposed model is able to predict Wiki entities as distractors. Our method is evaluated on two biology question datasets collected from Wikipedia and actual college-level exams. Experimental results show that our context-based method achieves comparable performance to a frequently used word2vec-based method for the Wiki dataset. In addition, we propose a second-stage learner to combine the strengths of the two methods, which further improves the performance on both datasets, with 51.7% and 48.4% of generated distractors being acceptable.

Text embedded in images provides important semantic information about a scene and its content. Detecting text in an unconstrained environment is a challenging task because of the many fonts, sizes, backgrounds, and alignments of the characters. We present a novel attention model for detecting arbitrary oriented and curved scene text. Inspired by the attention mechanisms in the human visual system, our model utilizes a spatial glimpse network to processes the attended area and deploys a recurrent neural network that aggregates the information over time to determine the attention movement. Combining this with an off-the-shelf region proposal method, the model achieves the state-of-the-art performance on the highly cited ICDAR2013 dataset, and the MSRA-TD500 dataset which contains arbitrary oriented text.