Work

Application Research of Multi-scale Feature Attention Improvement Algorithm Based on Retinex Algorithm and Deep Learning in Low-quality Image Enhancement

Year: 2025
Type: article
Abstract: In recent years, intelligent driving technology has drawn attention to improving lowquality images in complex traffic environments.This study proposes a method combining multiscale spatial attention mechanisms and the Retinex algorithm to enhance image detail and contrast.By integrating the Retinex algorithm with convolutional neural networks, the method optimizes brightness and sharpness.Additionally, a non-local attention module is used to capture remote dependencies and improve image semantics.Experimental results show that the proposed method achieves an SSIM of 0.96, a PSNR of 24.24, and reduces the running time by 83.12%, from 21.45 ms to 3.62 ms.The parameter count is also reduced from 25.06M to 20.45M, demonstrating significant computational efficiency.This method effectively enhances low-quality images, making it suitable for intelligent driving and real-time image processing applications.
Language: English
Open Access status: bronze
OpenAlex ID: W4417117730
FWCI:
Cites: 21
Cited by:
Related to: 10