Investigation of the Use of Artificial Intelligence Techniques for Automatic Detection of Geometric Patterns in Binary Images

Abstract

In this scientific initiation project, the objective is to use Cartesian genetic programming for pattern recognition in binary images, evolving operators that form morphological filters, in addition to performing such recognition also through convolutional neural networks. This will use several concepts covered in the Computer Engineering course, such as machine learning, evolutionary programming and digital signal processing, including the construction of filters.The solution based on Cartesian genetic programming will be implemented based on mathematical concepts of morphology to solve the pattern recognition problem applied to certain images, and this solution can be implemented in hardware. Then, a solution based on the YOLO convolutional neural network will also be developed to extract patterns from the same images, in order to obtain results through a state-of-the-art technology.Statistics will be generated for the two solutions, in order to compare with the results obtained in an original work in the literature, which used classical genetic programming for the same purpose.Both Cartesian genetic programming and the YOLO convolutional neural network and the concepts behind such techniques are not covered in graduation, which makes the study of these tools a differential for academic training. In addition, manipulation of image databases, statistical analysis of results and construction of morphological filters are important skills for the career of a computer engineer.