An artificial intelligence can easily turn simple line sketches into realistic photographs.
Lin Gao at the Chinese Academy of Sciences in Beijing and his colleagues have developed an algorithm that instantly turns a of a person鈥檚 face into a photo portrait.
The doesn鈥檛 require artistic expertise or coding skills. It could help rapidly generate images of suspects for criminal investigations or simplify the design process in making films and games, says Gao.
Advertisement
To train the algorithm, the team used a publicly available data set of 17,000 . For each photograph, they used image-processing software to simplify the photo until it resembled a pencil drawing.
The researchers then trained the algorithm on the sketch鈥損hoto pairs. For any given sketch of a , the AI learned to recognise five separate features: the left eye, right eye, nose, mouth and the rest of the face.
The AI then generates more for each of these components and stitches them together in a photorealistic representation. The process is automatic, with no manual input for features such as eye or skin colour.
Currently, the algorithm doesn鈥檛 produce photos of people of different skin colours, says Hongbo Fu at the City University of Hong Kong, a collaborator on the research. This is because the celebrity image data set used to train the AI consisted largely of people with white skin, which influences the AI-generated image.
鈥淢ost of the results are related to white skin colour 鈥� we don鈥檛 have any control about this,鈥� says Fu. In the future, the team would like to add a flexible control to manually select the complexion in a portrait.
The researchers also plan to expand the algorithm to generate photographs of non-human objects from sketches.
Reference:
