This study suggests a novel soft computing approach for enhancing face recognition performance using the feature learning methodology. The suggested system of Quasi-linear Partial Differential Equations (QPDE) based face recognition model is used to increase the accuracy of facial identification. The webcam first catches the user's image and saves it to the user's computer. The inference model receives the captured image. The input image is pre-processed to enhance the image, and then the image is split into numerous integral images at the inference model. Using speed-up robust features, an intriguing point is constructed for each integral image. The computed interest point is converted into a QPDE based weight matrix by representing the strength of features at all the regions of the face. With the generated weighted matrix of features extracted from the facial image, the method generates a Fast Artificial intelligence (AI) network with many layers. Each layer of the neural network has many neurons with dedicated functions to compute the similarity and feature weight using QPDE. The proposed approach has produced efficient results compared to earlier methods for the Flickr-Faces-HQ database (FFHQ) datasets.