Artificial Neural Networks in Art - Face Colorization and 3D Visualization

For many years, visual artists have been trying to represent their perspectives and present ideas through their artistic languages and specific media such as painting, photography, sculpture, etc. One of the significant developments in visual and digital art in recent years is the reveal of visual information and reconstruction of image by manipulating image data through Artificial Neural Networks (ANNs). Recent researches proved that the neural networks and algorithms can leverage the ability to image output in self-training approaches under limited data provided. To examine the ability of such neural networks in image generation, this paper proposes the use of Convolutional Neural Network (CNN) for the integration of two essential visual elements - color and form in facial visualization in art. The proposed approach requires training of the CNN with image dataset, by providing a grayscale 2D image, the CNN can colorize the image automatically, recognize and volumize the face in the image to generate a digital 3D model. By doing this, it is believed that the neural network is able to extend the possibility of both 2D and 3D visualization in art, and to a certain extent overcomes the limits of conventional art form and technique. A simple evaluation of the accuracy of color and 3D geometry can help reveal the automatic visualization ability of CNN and the next area of adjustment and development. Relevant issues related to the meaning and value of art, and the cognitive level of knowledge and understanding based on the generated visual information through ANNs will also be explored in this paper.