Abstract: Given no prior knowledge, the process of converting from a grayscale image to a colorful image is an “ill-posed” problem. Most of the previous methods are based on convolutional neural network (CNN), sparse dictionary and user intervention, making colorization either at a huge cost or an arduous work. This paper aims at solving some of the deficiency in previous work, such as methods based on user-intervention require too many human resources, and methods based on machine learning cost too much computational expense. Motivated by this, a novel automatic face image colorization method based on contextual information is proposed by this paper. Our facial image colorization method is based on machine learning. Utilizing the strong correlation between grayscale lightness, texture and color, we first train a joint distribution from our training set, and then solve the color of a targeted grayscale image under multiple constraints including first-order LBP, Second-order LBP, and lightness. Several experiments were performed to show that the proposed method outperforms the previous approaches by offering better authenticity and naturalness. Aiming specifically at facial image colorization, our method manages to achieve convincing results under a relatively small amount of data resources. As a result, this paper achieve desired effect by applying the local binary pattern (LBP) in the field of colorization, and hopefully could be applied in the field of image processing.Abstract: Given no prior knowledge, the process of converting from a grayscale image to a colorful image is an “ill-posed” problem. Most of the previous methods are based on convolutional neural network (CNN), sparse dictionary and user intervention, making colorization either at a huge cost or an arduous work. This paper aims at solving some of the deficien...Show More
