Architecture Proposal for Data Extraction of Chart Images Using Convolutional Neural Network

Chagas, Paulo; Freitas, Alexandre Abreu de; Akiyama, Rafael Daisuke; Miranda, Brunelli; Araújo, Tiago; Santos, Carlos Gustavo Resque dos; Meiguins, Bianchi Serique; Morais, Jefferson

doi:10.1109/iv.2017.37

Cited by 12 publications

(18 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, the CNN technique to classify chart images has attracted increasing attention from researchers [5][6][7][8][9]. Amara et al [5] presented a CNN architecture for classifying 11 different chart types and achieved 89.5% accuracy over 3377 images.…”

Section: Related Workmentioning

confidence: 99%

“…Bajic et al [6] used the VGG (Visual Geometry Group) model, which is one of the well-known CNN architectures, and achieved 81.67% accuracy for 10 chart categories on 541 test images. Another work [7] used the CNN technique to predict what type of chart a given image is representing (i.e. area, bar, line, pareto, pie or radar).…”

Section: Related Workmentioning

confidence: 99%

“…As shown in Table 1, our work differs from the previous works in many respects. First, while many studies [1][2][3][4][5][6][7][8][9] have been focused on the classification of chart images by type (i.e. area, bar, line, pie), our study was conducted to classify line charts according to their trend (increasing or decreasing) and functional (linear or exponential) properties.…”

Section: Related Workmentioning

confidence: 99%

“…area, bar, line, pie), our study was conducted to classify line charts according to their trend (increasing or decreasing) and functional (linear or exponential) properties. Second, in the literature, many studies [1][2][3][4][5][6][7][8][9] have been focused on multi-class classification, while a study [10] conducted the chart type classification as a multi-instance classification problem. Unlike these previous works, our paper presents an experimental study of multi-label classification (MLC) methods (LP and BR) and gives suggestions for MLC that are effective for automatic chart image interpretation applications.…”

Section: Related Workmentioning

confidence: 99%

“…In the literature, while a wide range of studies [1][2][3][4][5][6][7][8][9][10][11] on chart classification cover various types of charts (i.e. area, bar, map, line, pie), several studies only cover a single chart type such as only pie chart [12], line chart [13] or radar chart [14].…”

Section: Related Workmentioning

confidence: 99%

See 4 more Smart Citations

Multi-label classification of line chart images using convolutional neural networks

Kösemen

Birant

2020

SN Appl. Sci.

View full text Add to dashboard Cite

In this paper, we propose a new convolutional neural network (CNN) architecture to build a multi-label classifier that categorizes line chart images according to their characteristics. The class labels are organized in the form of trend property (increasing or decreasing) and functional property (linear or exponential). In the proposed method, the Canny edge detection technique is applied as a data preprocessing step to increase both the classification accuracy and training speed. In addition, two different multi-label solution approaches are compared: label powerset (LP) and binary relevance (BR) methods. The experimental studies show that the proposed LP-CNN model achieves 93.75% accuracy, while the BR-CNN model reaches 92.97% accuracy on the test set, which contains real-world line chart images. The aim of this study is to build an efficient classifier that can be used for many purposes, such as automatically captioning the chart images, providing recommendations, redesigning charts, organizing a collection of chart images and developing better search engines.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%