为了提高低照度图像的清晰度和避免颜色失真,提出了基于注意力机制和卷积神经网络(CNN)的低照度图像增强算法,以改善图像质量。首先根据Retinex模型合成训练数据,将原始图像从RGB (red-green-blue)颜色空间变换到HSI (hue-saturation-intensity)颜色空间,然后结合注意力机制和CNN构建A-Unet模型以增强亮度分量,最后将图像从HSI颜色空间变换到RGB颜色空间,得到增强图像。实验结果表明,所提算法可以有效改善图像质量,提高图像的清晰度,避免颜色失真,在合成低照度图像和真实低照度图像的实验中均能取得较好的效果,主观和客观评价指标均优于对比算法。

To improve the clarity of low-light images and avoid color distortion, a low-light image enhancement algorithm based on the attention mechanism and convolutional neural network (CNN) is proposed to improve image quality. First, the training data is synthesized based on the Retinex model, and the original image is transformed from RGB (red-green-blue) color space to HSI (hue-saturation-intensity) color space. Then, an A-Unet model is constructed to enhance the brightness component by combining the attention mechanism and CNN. Finally, the enhanced image is obtained by transforming images from the HSI color space to the RGB color space. Experimental results show that the proposed algorithm can effectively improve the image quality, improve the image clarity, and avoid the color distortion. Good results can be obtained in the experiments of synthesizing low-light images and real low-light images, and the subjective and objective evaluation indexes are better than that of the comparison algorithm.