Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Image edge enhancement for effective image classification

Tianhao Bu, Michalis Lazarou, Tania Stathaki

TL;DR

This work introduces Edge Enhancement (E^2), a data augmentation strategy that fuses high-frequency edge features extracted via Canny edge detection with RGB images to improve image classification. Grounded in high-boost filtering, E^2 trains networks on concatenated inputs consisting of original and edge-enhanced images, yielding faster convergence and higher accuracy on CIFAR-10 and CALTECH101 across LeNet-5, CNN-9, and ResNet-18. Empirical results show notable gains (e.g., CIFAR-10 LeNet-5 +5.7 percentage points; CALTECH101 +1.56 percentage points) and reduced training epochs, with additional improvements observed when combined with common geometric transforms. The approach offers a computationally efficient, principle-driven augmentation that enhances semantic information in training data and generalization for CNN-based image classification, with potential for scaling to larger datasets and semi-supervised settings.

Abstract

Image classification has been a popular task due to its feasibility in real-world applications. Training neural networks by feeding them RGB images has demonstrated success over it. Nevertheless, improving the classification accuracy and computational efficiency of this process continues to present challenges that researchers are actively addressing. A widely popular embraced method to improve the classification performance of neural networks is to incorporate data augmentations during the training process. Data augmentations are simple transformations that create slightly modified versions of the training data and can be very effective in training neural networks to mitigate overfitting and improve their accuracy performance. In this study, we draw inspiration from high-boost image filtering and propose an edge enhancement-based method as means to enhance both accuracy and training speed of neural networks. Specifically, our approach involves extracting high frequency features, such as edges, from images within the available dataset and fusing them with the original images, to generate new, enriched images. Our comprehensive experiments, conducted on two distinct datasets CIFAR10 and CALTECH101, and three different network architectures ResNet-18, LeNet-5 and CNN-9 demonstrates the effectiveness of our proposed method.

Image edge enhancement for effective image classification

TL;DR

This work introduces Edge Enhancement (E^2), a data augmentation strategy that fuses high-frequency edge features extracted via Canny edge detection with RGB images to improve image classification. Grounded in high-boost filtering, E^2 trains networks on concatenated inputs consisting of original and edge-enhanced images, yielding faster convergence and higher accuracy on CIFAR-10 and CALTECH101 across LeNet-5, CNN-9, and ResNet-18. Empirical results show notable gains (e.g., CIFAR-10 LeNet-5 +5.7 percentage points; CALTECH101 +1.56 percentage points) and reduced training epochs, with additional improvements observed when combined with common geometric transforms. The approach offers a computationally efficient, principle-driven augmentation that enhances semantic information in training data and generalization for CNN-based image classification, with potential for scaling to larger datasets and semi-supervised settings.

Abstract

Image classification has been a popular task due to its feasibility in real-world applications. Training neural networks by feeding them RGB images has demonstrated success over it. Nevertheless, improving the classification accuracy and computational efficiency of this process continues to present challenges that researchers are actively addressing. A widely popular embraced method to improve the classification performance of neural networks is to incorporate data augmentations during the training process. Data augmentations are simple transformations that create slightly modified versions of the training data and can be very effective in training neural networks to mitigate overfitting and improve their accuracy performance. In this study, we draw inspiration from high-boost image filtering and propose an edge enhancement-based method as means to enhance both accuracy and training speed of neural networks. Specifically, our approach involves extracting high frequency features, such as edges, from images within the available dataset and fusing them with the original images, to generate new, enriched images. Our comprehensive experiments, conducted on two distinct datasets CIFAR10 and CALTECH101, and three different network architectures ResNet-18, LeNet-5 and CNN-9 demonstrates the effectiveness of our proposed method.
Paper Structure (21 sections, 12 equations, 6 figures, 3 tables, 1 algorithm)

This paper contains 21 sections, 12 equations, 6 figures, 3 tables, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. BACKGROUND
  3. Data augmentation
  4. High boost filtering
  5. Canny edge detection
  6. METHODOLOGY
  7. Problem definition
  8. Edge enhancement
  9. Training phase
  10. Inference stage
  11. EXPERIMENTS
  12. Setup
  13. Datasets
  14. Implementation details
  15. Networks
  16. ...and 6 more sections

Figures (6)

  • Figure 1: Edge enhancement process. Leftmost: original image. Mid left: corresponding R,G,B channels of the original labelled data. Mid right:corresponding extracted edges Rightmost: edge enhanced image.
  • Figure 2: A batch of 4 images. Top row: original data. Mid row: corresponding extracted edges. Bottom row: corresponding edge enhanced data
  • Figure 3: The flowchart of the edge enhancement method for model training. i) Original data is edge enhanced to the transformed data. ii) The original and transformed data is concatenated then feed into the network. iii) Model training. iv) Classification results obtained through test dataset.
  • Figure 4: CIFAR10 through CNN-9
  • Figure 5: CIFAR10 through LeNet-5
  • ...and 1 more figures