A Saliency-Based Convolutional Neural Network for Table and Chart Detection in Digitized Documents

Abstract

Deep Convolutional Neural Networks (DCNNs) have recently been applied successfully to a variety of vision and multimedia tasks, thus driving development of novel solutions in several application domains. Document analysis is a particularly promising area for DCNNs: indeed, the number of available digital documents has reached unprecedented levels, and humans are no longer able to discover and retrieve all the information contained in these documents without the help of automation. Under this scenario, DCNNs offers a viable solution to automate the information extraction process from digital documents. Within the realm of information extraction from documents, detection of tables and charts is particularly needed as they contain a visual summary of the most valuable information contained in a document. For a complete automation of visual information extraction process from tables and charts, it is necessary to develop techniques that localize them and identify precisely their boundaries. In this paper we aim at solving the table/chart detection task through an approach that combines deep convolutional neural networks, graphical models and saliency concepts. In particular, we propose a saliency-based fully-convolutional neural network performing multi-scale reasoning on visual cues followed by a fully-connected conditional random field (CRF) for localizing tables and charts in digital/digitized documents. Performance analysis carried out on an extended version of ICDAR 2013 (with annotated charts as well as tables) shows that our approach yields promising results, outperforming existing models.

Deep LSTM with Dynamic Time Warping Processing Framework: A Novel Advanced Algorithm with Biosensor System for an Efficient Car-Driver Recognition

“…Future developments of the proposed method should focus on the integration of vision algorithms always based on the use of deep learning and video saliency [27,28] for the characterization not only of the driver's identity but also of the correlated anxiety and stress level in order to improve driving safety and assistance systems. It is interesting to highlight that the author has already patented and published a study in which, through an innovative low frame-rate motion magnification technique applied to the driver's face (captured by a vision camera with which the latest generation cars are being equipped), it is possible to reconstruct some features of the corresponding car-driver PPG signal in order to address the problem that could occur if the driver does not put a hand on any of the sensors located on the steering.…”

TableNet: Deep Learning Model for End-to-end Table Detection and Tabular Data Extraction from Scanned Document Images

“…This work achieves state-of-the-art performance on the ICDAR 2013 table competition dataset. After this, [9] combined deep convolutional neural networks, graphical models and saliency concepts for localizing tables and charts in documents. This technique was applied on an extended version of ICDAR 2013 table competition dataset and outperforms existing models.…”

Weakly Supervised Learning-based Table Detection

“…This proves the quantity of annotated data, availability of pretrained models as well as test data play a prime role in prediction. In [24], author proposes a saliency-based approach for table and chart detection. In this work, feature extraction layer of VGG-16 is used; the first two layers of architecture have been modified to take facilitate detection of tables.…”