Researchers from TCS Robotics Research Lab in India has launched an artificial intelligence (AI) system called ‘Chitrakar’ that converts an image of a human’s face into a recognizable non-self-intersecting loop, known as a Jordan curve. Finally, this Jordan curve can produce realistic portraits of the person’s original image using a robotic hand.
Aniruddha Singhal, one of the researchers at TCS Robotics Research Lab, states in a blog that this paper’s idea occurred to him when he came across the article by Fiona Ross and William T. Ross named “Jordan Curve Theorem Is Non-trivial’. He connected it with another book ‘In Pursuit of the Traveling Salesman’ by William J. Cook, discussing traveling salesman problem (TSP) art. He interpreted the article from a Vedantic (Hindu philosophical) perspective, which connecting the notion of TSP art and the Jordan curve, turning it into an illustration of the ultimate beauty.
What is a Jordan Curve
A Jordan curve is a single line that ends at the same point in space where it started, drawing an image without ever intersecting itself. While in TSP artform, a single, non-intersecting line is drawn to create images.
Aniruddha Singhal started investigating the possibility of automatically producing TSP drawings, which are very difficult for humans to create. With a TCS Robotics Research Lab team, he built Chitrakar, a system that can convert an image into a drawing formed from a single line, eventually transforming it into a Jordan curve. Chitrakar signifies one who creates a painting in Hindi/Sanskrit.
The system used SOTA deep-learning techniques to segment the human face from a photo and combined this with image enhancement techniques. The enhanced image is then stipped with the points being connected by a TSP solver (where every point is considered to be a destination of a traveling salesman). The intersection removal technique converts the final route of a traveling salesman into a Jordan curve.
Chitrakar can automatically convert any image of a human face into a Jordan curve that can be used to create artistic sketches on paper using a robotic gripper. The robotic gripper can hold various pens, producing distinct drawings with thinner/thicker lines, different colors, etc. It can create pictures with a satisfying result in less than 30 minutes, almost impossible for human artists to achieve. The system is patented in several countries and can become widely available.
The team aims to extend this for other mediums, which are challenging for humans to handle. They plan to use Style-GAN and other generative models to generate portraits from a provided photograph, converting them into pencil sketches next.