Deep Learning of Color Constancy Based on Object Recognition

Color constancy is the ability of human beings to recognize the colors of objects independently of the characteristics of the light source. Computational color constancy aims to estimate the illuminant and subsequently use this information to correct the image and display how it would appear under a canonical illuminant. The deep learning method is among the most successful illumination estimation methods to date and typically relies on a training set of images labeled with the respective scene illuminant. Although the human visual system is often compared to a machine learning algorithm, during evolution it was never presented with ground truth illuminants. Instead, it is hypothesized that the ability of color constancy arose because it helped other crucial tasks, such as recognizing fruits, objects, and animals independently of the scene illuminant. With the development of science and the improvement of people's quality of life, the field of artificial intelligence has developed rapidly, and the progress in image recognition has been even more rapid in recent years. This paper studies object detection in low lighting environments and uses deep learning algorithms to detect and analyze images. In low lighting environments, the detected objects are compared with the big data to obtain the objects with the closest similarity for identification and confirmation and compare the accuracy of the learning model for object recognition in complex lighting environments.