Contour Sparse Representation with SDD Features for Object Recognition

TL;DR

This paper introduces SDD features derived from object contours, which form a sparse representation enabling highly accurate object and gesture recognition, outperforming existing methods.

Contribution

The paper proposes SDD features for contour-based sparse representation, enhancing object recognition accuracy and robustness over previous segmentation and recognition techniques.

Findings

Achieved 100% accuracy on NUS and near-infrared gesture datasets.

Achieved 100% accuracy on Kimia 99 object dataset.

Demonstrated robustness and effectiveness of SDD features in recognition tasks.

Abstract

Slope difference distribution (SDD) is computed for the one-dimensional curve. It is not only robust to calculate the partitioning point to separate the curve logically, but also robust to calculate the clustering center of each part of the separated curve. SDD has been proposed for image segmentation and it outperforms all existing image segmentation methods. For verification purpose, we have made the Matlab codes of comparing SDD method with existing image segmentation methods freely available at Matlab Central. The contour of the object is similar to the histogram of the image. Thus, feature detection by SDD from the contour of the object is also feasible. In this letter, SDD features are defined and they form the sparse representation of the object contour. The reference model of each object is built based on the SDD features and then model matching is used for on line object recognition. The experimental results are very encouraging. For the gesture recognition, SDD achieved 100% accuracy for two public datasets: the NUS dataset and the near-infrared dataset. For the object recognition, SDD achieved 100% accuracy for the Kimia 99 dataset.