EventGAN: An Unsupervised Low-Light Grayscale Image Enhancement Method Based on Event Camera

Event cameras, as a novel class of bioinspired vision sensors, are ideal for low-light enhancement thanks to their high dynamic range (HDR) characteristics. However, two critical challenges emerge when applying event camera data to low-light image enhancement (LIME) tasks: ineffective fusion between conventional images and event data streams, and the lack of paired training data. To address these problems, in this article, we propose the EventGAN, an eventassisted unsupervised low-light enhancement method. First, we propose a modified image-to-event simulation method that transforms low-light-enhanced images into event representations, enabling effective joint processing of images and event data. Second, we design an event similarity loss function that establishes mappings between input event images and simulated event images, eliminating the need for paired training data. Finally, we introduce a local discriminator designed to suppress regional overexposure or underexposure artifacts, thereby significantly improving the effectiveness of LIME. Experimental results conclusively demonstrate that our method outperforms current LIME techniques. As a sensor signal processing approach for extreme lighting, it not only solves key challenges in using event cameras for low-light enhancement but also produces clearer, more natural, and higher quality images in very dark conditions.