Geometric Representations in Computer Vision

The field of computer vision is moving towards a greater emphasis on geometric representations, with a focus on interpretable and hand-engineered features. Recent research has demonstrated the effectiveness of curvature-based representations for handwritten character recognition, achieving high accuracy rates without the need for convolutional neural networks. Additionally, the role of Gaussian curvature in 3D surface modeling has been investigated, showing its potential as a sparse and compact description of 3D surfaces. Noteworthy papers include:

  • A study on using planar curvature and gradient orientation for handwritten character recognition, achieving 97 percent accuracy on MNIST digits and 89 percent on EMNIST letters.
  • A paper on the role of Gaussian curvature in 3D surface modeling, demonstrating its potential as a geometric prior that can inform and improve 3D surface reconstruction.

Sources

An MLP Baseline for Handwriting Recognition Using Planar Curvature and Gradient Orientation

Towards Understanding 3D Vision: the Role of Gaussian Curvature

PCA- and SVM-Grad-CAM for Convolutional Neural Networks: Closed-form Jacobian Expression

A Sobel-Gradient MLP Baseline for Handwritten Character Recognition

Mesh Processing Non-Meshes via Neural Displacement Fields

Exploration of Deep Learning Based Recognition for Urdu Text

Built with on top of