Implementation of Face Recognition Using Discrete Cosine Transform on Convolutional Neural Networks

Abstract

Authentication in a system is one way to protect user data, several methods to authenticate are with passwords, fingerprints and face recognition. The use of deep learning in face recognition requires relatively large and large datasets. Convolutional neural network (CNN) is the method used in this experiment, but this method requires a large enough storage space so that it becomes one of the problems in terms of data storage. The application of discrete cosine transform (DCT) is one way in terms of image compression to reduce storage space on the platform. The use of the DCT image compression method in this study was carried out to compare the image datasets transported by DCT and those that were not transported to the speed of the learning process and the results of the learning process. From the results of experiments carried out that changes to the activation function and epoch it was found that the application of the sigmoid activation function was the most optimal compared to other activation functions. The application of DCT to all activation functions found that on average the learning process could be completed faster. the accuracy value on the sigmoid both from the dataset with DCT application or not, the accuracy value is above 95%, with the optimal value at the 60th epoch the accuracy value reaches 99.8% without -DCT and 99.6% with DCT.