Classification of Image Classes Based on the PCA Algorithm Optimized by the KNN Algorithm Improved by Genetic Algorithms

In this paper we propose an optimized version of the PCA algorithm by using genetic algorithms and the KNN technique with applications in the classification of image classes. The algorithm are developed by optimizing the images projection stage based on the eigenvectors that ensure the extraction of essential features from the data. In that stage, the optimal value regarding the size of the projection space of the initial images is determined by using a genetic algorithm and and the improved KNN algorithm regarding the key parameter represented by the number of nearest neighbors used in the testing stage. For the validation and testing of the developed algorithms, the classification of face image classes corresponding to several people is considered. The proposed algorithm includes several versions, analyzed in the experimental study, depending on the similarity measure used (the Euclidian distance, Manhattan distance, Mahalanobis distance, Minkowski distance and cosine) to determine the nearest neighbors to the KNN algorithm, as well as based on the crossover recombination method chosen at the level of binary genes with a crossover point or more many crossing points. The experimental study on the methods established in the work attests good performance through the percentages of correctness, resulting in applications for recognizing people through facial images.