A physics-based vision enhancement algorithm
In a new paper published in eLight, a team of scientists led by Professor Bahram Jalali and graduate student Callen MacPhee from UCLA have developed a new algorithm for performing computational imaging tasks. The paper “VEViD: Enhancing Vision Through Virtual Diffraction and Coherence Detection” uses a physics-based algorithm to correct for poor brightness and low contrast in low-light images.
Under such conditions, digital images often suffer from undesirable image qualities such as low contrast, loss of features, and poor signal-to-noise ratio. Low-light image enhancement aims to improve these qualities for two purposes: increasing image quality Because human perception and increase the accuracy of computer vision algorithms. In the past, real-time processing could serve as a benefit for viewing convenience. It is then a requirement for emerging applications such as autonomous vehicles and security therein image processing must be completed with low latency.
The paper shows that physical diffraction and coherent detection can be used as a toolbox for digital image and video conversion. This approach leads to a new and surprisingly powerful algorithm for color enhancement and low light.
Unlike traditional algorithms which are mostly hand-generated empirical rules, the VEViD algorithm simulates physical processes. In contrast to deep learning-based approaches, this technique is unique in that it is derived from deterministic physics. The algorithm is interpretable and does not require labeled data for training. The authors explain that although mapping to physical processes imprecisely, in the future it is possible to implement a physical device that executes the algorithm in the same domain.
The paper demonstrates the high performance of VEViD in a number of imaging applications such as security cameras, night driving and space exploration. Also demonstrated is VEViD’s ability to perform color enhancement.
The algorithm’s exceptional computational speed is demonstrated by processing 4k video at more than 200 fps. Comparisons with leading deep learning algorithms show the same or better image quality but with processing speeds one to two levels faster.
Deep neural networks have proven to be powerful tools for object detection and tracking, and they are key to several emerging technologies that leverage automatons. The authors show that the utility of VEViD is a preprocessing tool that increases the accuracy of object detection using a popular neural network (YOLO).
First image processing using VEViD allows neural networks trained based on daylight images to recognize objects in night environments without retraining, making these networks robust. while saving huge amounts of time and energy.
Bahram Jalali et al., VEViD: Enhance vision through virtual diffraction and coherent detection, eLight (In 2022). DOI: 10.1186 / s43593-022-00034-y
Provided by
Chinese Academy of Sciences
Quote: VEViD: A Physically Based Vision Enhancement Algorithm (2022, Nov 8) retrieved Nov 8, 2022 from https://techxplore.com/news/2022-11-vevid-vision- algorithm-based-physics.html
This document is the subject for the collection of authors. Apart from any fair dealing for personal study or research purposes, no part may be reproduced without written permission. The content provided is for informational purposes only.