A Graphical Computing-Based Image Registration Method for LED Chips

Image registration is a key problem and a technical difficulty in the field of machine vision and image processing research. For the problems that the detection accuracy of the machine learning method depends on a large number of training samples, and it is difficult to take samples in the LED chip production process, a graphical computing-based LED chip image registration method is proposed. The geometrical representation of the LED chip image features is used to realize the registration of the chip image in the form of geometric calculation. Image morphology processing and Hough transform are used to calculate the registration angle of chip images. With the variational thinking of tessellation grid, a simplified computational model of tessellation grid is proposed to replace the global information with key features of the image for fast registration. The step-by-step generalized image registration algorithm is designed for multiple types of LED chip image characteristics to achieve efficient registration of multiple LED chip images. According to the experimental results of multi-type LED chip image registration, it shows that the proposed graphical computational registration method has a high average registration accuracy and a good registration effect for a variety of LED chip images, which can lay a good foundation for large-scale online detection of LED chip defects.