Human facial recognition is an integral part of computer vision. It has to do with the identification and analysis of facial images for forensic or medical reasons. A number of techniques for recognizing facial images exist but these techniques suffer high computational complexities. Subsequently, an improved technique is proposed to reduce the complexities associated with the recognition process. To achieve this aim, an optimized swarm kernel linear discriminant-based technique is proposed. Real life data consisting of 360 images of different persons were acquired; out of which 240 were used for training and 120 were used for testing. Images were pre-processed and normalized using histogram equalization method. The feature extraction process was optimized using particle swarm optimization and similarity indexes of the images and were measured with Euclidian distance function. The technique was implemented in Matrix Laboratory 9.0 (R2018a). Results showed that, the optimized technique outperformed the existing one based on false positive rate, sensitivity, specificity, accuracy and complexity.