6626070
2997924

PAPER-REVIEW-0001, Visualizing for the Non-Visual:Enabling the Visually Impaired to Use Visualization

Back to the previous page ｜ Download to file in pdf
List of posts to read before reading this article

Contents

1. Introduction




2.1. Assistive Technologies




2.2. Assistive Technologies for Visual Impairment




2.3. Visualization for the Non-Visual




2.4. Extracting Data from Charts

Approach extracting specific components of a chart and does not extract raw data

  • Extraction of simple graphical entities from SVG images in PDF for chart classification ｜ PDF
  • Web crawler that collects and classifies SVG images using image elements ｜ PDF
  • The Hough transform algorithm to extract information from bar charts, even hand-drawn ones ｜ PDF
  • Extraction of information from additional chart types, including pie charts and line charts ｜ PDF
  • Extraction of data tables from raster images for accessibility use and SIGHT generates summary of a simple bar chart for visually impaired users
  • A system that transforms gray scale charts into XML format
  • The performance of these systems can be improved with human guidance, as demonstrated by the ChartSense tool
  • Extraction of information from certain charts
  • Extraction of data from existing raster chart images and generates interactive animated visualization
  • ReVision, to classify images of charts into five categories and extract data from bar charts and pie charts
  • An automatic pipeline for extracting a visual encoding specification given a raster chart image
  • Recovering the color mappings of charts that include a color legend




3. Domain Characterization

3.1. Method




3.2. Findings




4. Extracting Data from Chart Images

4.1. Chart Classification

  • Convolutional Neural Networks (CNNs) as a classification model, which have shown impressive performance on image classification tasks
  • Residual networks that yield state-of-the-art performance in most computer vision tasks
  • Existing Resnet trained on the Imagenet dataset and appending a global average pooling layer before the last fully connected layer
  • Adam optimizer, learning rate as 0.0005
  • Resizing to 512×512 pixels




4.2. Text Extraction




4.2.1. Textual Region Detection

  • The PixelLink model that shows state-of-the-art performance in text detection tasks
  • VGG16 as a feature extractor, prediction of text and link, and performance of instance segmentation
  • The model takes an input image and predicts the bounding box, which informs coordinates in the image for each label.
  • Training it on the SynthText dataset for 400K iterations




4.2.2. Text Recognition

  • convolutional recurrent neural networks (CRNNs) ｜ PDF




4.2.3. Text Role Classification

  • For text role classification, Reverse‐Engineering Visualizations: Recovering Visual Encodings from Chart Images ｜ PDF




4.3. Data Extraction

4.3.1. Decoding Visual Encodings of Charts




4.3.2. Decoding Bar Charts

  • Detection of bars with the Yolo2 ｜ PDF




4.3.3. Decoding Pie Charts




4.3.4. Decoding Line Charts




4.4. Browser Plugin for Visually Impaired Users




5. Quantitative Evaluation

5.1. Evaluation Dataset




5.2. Recognition Accuracy

5.2.1. Chart Type Classification




5.2.2. Text Extraction




5.2.3. Data Extraction




6. Qualitative Evaluation

6.1. Method




6.2. Findings




7. Conclusion and Future Work




List of posts followed by this article

Reference

OUTPUT