A Genetic Algorithm-based Image Enhancement Approach for Autonomous Driving at Night

Image enhancement increases the perceived quality and improves the experience of viewers by processing images that are difficult to see, such as due to low light or overexposure. This is specifically important for night monitoring cameras or for the performance of visual-based night autonomous driving algorithms. This paper proposes a multi-adaptive fusion image enhancement algorithm (MAAF) to adaptively select and fuse Histogram Equalization (HE), Multi-scale Retinex (MSR), and Gamma Correction (GC) in the image frequency domain through Discrete Cosine Transform (DCT). Based on a genetic algorithm, the proposed MAAF combines the advantages of the different methods in terms of lighting enhancement (HE), edge enhancement (MSR), and overexposed image enhancement (GC) to achieve an overall performance optimization. A comprehensive evaluation score (CES) is also proposed in this paper as an overall assessment metric. MAAF was evaluated in terms of multiple metrics, including entropy, average gradient, contrast, and PSNR. Experimental results showed that MAAF obtains the highest CES compared with other algorithms.