The United States has developed an image segmentation algorithm based on thermal imaging

According to a report by ScienceDaily on June 21, researchers at Purdue University in the United States have developed a computer image segmentation algorithm based on thermal imaging, which allows computers to quickly identify the shape of an object, even if it is distorted or slightly deformed. This technology will make machine vision closer to human vision and can be widely used in image search, medical imaging, drone manufacturing and other fields. The detailed research results will be divided into two papers and will be announced at the IEEE (Institute of Electrical and Electronics Engineers) Computer Vision and Pattern Recognition Conference held from June 21 to 23.

Humans can easily distinguish a three-dimensional object from its background, and can also easily identify its parts and the whole. But this is more complicated for computers: according to current computer vision recognition technology, the computer must be told the initial information of the target object's appearance and how many segments it can be divided into in advance, and then a certain algorithm is used to compare the known image with the unknown image, and finally the target is identified through screening. The new algorithm is closer to humans and is an unsupervised machine learning technology (computers or robots can have perception and learning capabilities without any prior training). The computer can estimate the number of segments that can be divided without providing it in advance.

Karthik Ramani, a professor of mechanical engineering at Purdue University in the United States who led the research, said the new algorithm uses two new technologies, called heat maps and heat distribution. Because heat diffuses along the surface of an object and accurately displays the shape of the object, through this "intelligent heat", the algorithm can simulate the flow of heat from one point to another along the surface of an object, thereby describing the shape of the object. Its working principle is that in order to facilitate description and identification, the image is first divided into many triangular grids, and then the shape of the object is identified by calculating the heat flowing between the grids. This method can not only be used to track actual heat, but also simulate the flow of heat according to the corresponding principles. Ramani said that the technology was created by standing on the shoulders of giants. Einstein's contribution to the theory of heat diffusion and French physicist Fourier's research on the propagation of heat in solids have provided great help to the algorithm.

In the experiment, the researchers used the algorithm to test objects of various complex shapes, including hands, humanoid models and centaurs. The results showed that the computer could accurately identify these objects even if they were distorted or slightly deformed. In addition, the algorithm can ignore the "noise" generated by incomplete laser scans or other erroneous data.

Researchers say the technology has great potential and can be widely used in many fields, including image search, robot vision and navigation, 3D medical imaging, drone manufacturing, multimedia game consoles, and animation film character creation.