A Multi-Purpose Shallow Convolutional Neural Network for Chart Images.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Visualization

Background:

• Graphical representation of data using charts is prevalent across numerous fields.
• Automated retrieval and processing of information from chart images require efficient algorithms.
• Convolutional Neural Networks (CNNs) show promise in image processing but require careful parameter tuning.

Purpose of the Study:

• To propose a novel Shallow Convolutional Neural Network (SCNN) architecture for chart-type classification and image generation.
• To evaluate the SCNN's performance in classification and generation tasks.
• To demonstrate the SCNN's efficiency compared to Deep Convolutional Neural Networks (DCNNs).

Main Methods:

• Development of a novel Shallow Convolutional Neural Network (SCNN) architecture.
• Validation through three use cases: a single SCNN classifier, two parallel SCNN models, and a generative adversarial network (GAN) setup.
• Extensive experimental analysis on classifying seven chart classes.

Main Results:

• The SCNN classifier achieved an average accuracy of 97.14% for chart-type classification.
• A parallel SCNN model configuration reached an average classification accuracy of 100%.
• The generative model produced plausible chart images, demonstrating effective image generation capabilities.

Conclusions:

• The proposed SCNN is a highly effective tool for both chart image classification and generation.
• SCNN offers comparable performance to DCNNs with superior computational efficiency, reduced time, and space complexity.
• This research facilitates more efficient automated analysis and creation of chart-based data representations.