New Approach to Facial Expression Recognition and Classification Using Typical Testors

Within the study of testors in the world of classification processes, the limitation for non-discrete data sets is presented. The usual technique is to binarize the data, but in this work, we seek to propose a method to discern between two objects from each other without the usual approach, with the aim that this method is helpful in the creation of the basic matrix and the algorithms for calculating testors. In this case, we take as a reference a set of reference points of faces for a simple classification process, and we will seek to find what in this project is known as a disparity matrix. The proposed method aims to get the testor according to a threshold of similarity, thus finding a subset of attributes that do not affect the property to be discernible between objects of different classes. The results show that it was possible to find a subset that reduces the number of attributes to 61.8% of the original set, while still maintaining the property to discern between classes, confirming that the proposed method is useful for this data set. The purpose is to expose this form of analyzing systems of information in order to have a more concrete understanding of what a testor demonstrates or details. Keeping track of this idea will help address certain limitations or experimental practices of this project, such as understanding the role of the similarity threshold within the obtained testors, and applying other algorithms with this method, like the LEX algorithm or the YYC algorithm.