Abstract: To solve the problems of dark light and difficult feature extraction of night-time images, a no-reference night-time image quality evaluation method based on global and local features is proposed. Firstly, the contour principle is used to divide the image into light region and dark region, and the proportion of the bright region is taken as feature1. Secondly, the global brightness information of the night-time image is extracted and used as feature2. Then, the differential operator method is adopted to obtain the edge of the image as feature3. Finally, the night-time image is converted from RGB to HSI, and the hue, saturation and brightness components are extracted as feature4, feature5 and feature6. Combining the above features, an evaluation model is established by BP neural network to evaluate the quality of night-time images. The test results on the public database show that the proposed method is more consistent with the subjective score and better than the existing image quality evaluation methods.

Key words: night-time image, quality evaluation, local bright region, edge detection