What is it about?
Oscillatory neural networks (ONNs) are artificial neural networks inspired by the collective dynamics of neural oscillations in the brain. ONNs can be trained for signal processing tasks such as pattern recognition and image edge detection. The principal advantages are the robustness to noise and the possibility of implementing them on analog electronic chips. These specialized chips aim to be faster and more energy efficient than GPU implementations for the same task. This article shows the possibility to take advantage of the fast ONN computation by integrating it into a complex image processing algorithm. In particular, we propose to accelerate the SIFT feature detection algorithm, performing the image contour detection with ONN.
Featured Image
Photo by Hal Gatewood on Unsplash
Why is it important?
This work is part of the European project NEURONN which aims to develop and find applications for ONNs. Advances in these technologies can be the key to energy-efficient embedded objects. Chips for specialized tasks could help push back the power limitations of embedded systems.
Perspectives
The advances of ONNs may have a significant impact on fields where energy and power are critical points, such as mobile robotics.
Sylvain Gauthier
Read the Original
This page is a summary of: SIFT-ONN: SIFT Feature Detection Algorithm Employing ONNs for Edge Detection, April 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3584954.3584999.
You can read the full text:
Resources
NeurONN
Link to the NeurONN website
Video
SIFT-ONN: SIFT feature detection algorithm employing ONNs for edge detection This work is a collaborative work of CNRS and A.I.Mergence in the context of the European H2020 NEURONN project. SIFT is a feature detection algorithm with high precision but slow computation latency, not adapted to real-time embedded applications. In this work, we propose to use fast and efficient ONN performing image edge detection as a first step of the SIFT algorithm, creating the SIFT-ONN.
Contributors
The following have contributed to this page







