Open Conference Systems, MISEIC 2019

Font Size: 
Deep Learning for Face Identification
Kartika Fithriasari, Ulfa Siti Nuraini

Last modified: 2020-02-02


Face identification or recognition has already entered aspects of our lives. Face identification uses a computer algorithm that can generate catching image face using camera, detect the pattern of it, also compare it with faces data that had been trained before, so computer can identify individual. Because of faces data have many feature and use very complex computer algorithm, we can use deep learning. One method of deep learning used in this paper is Deep Feedforward Network. In this paper, identify classification of individual identity is done using primary datasets. The images used to train is not only front look but also 45 degrees and 90 degrees to right and left. Each image also uses a different colour clothes and different accessories. First pre-process them from RGB to grayscale and resize it. For adding image, augmentation is done to multiply the dataset. After augmentation has been carried out, we classified them with deep feedforward network method which the number of hidden layer are 2,3,4 and 5, and the number of neurons within each hidden layer are 25,50,75, and 100. Based on result and discussion, the conclusion performances of classification are all above 70% which indicated that the result is excellent for face classification. From the result, we can conclude that more number of hidden layer and more number of neurons within hidden layer not always make better performance of classification. But they can make time longer for training. In this paper, best model is 2 hidden layers and 75 neurons within hidden layer.


Deep Learning; Face Identification; Multilayer Perceptron

Full Text: XML