Novel CNN with investigation on accuracy by modifying stride, padding, kernel size and filter numbers

• Published in: Multimedia tools and applications 2023-06, Vol.82 (15), p.23673-23691
• Main Authors: Naseri, HamidReza, Mehrdad, Vahid
• Format: Article
• Language: English
• Subjects: Accuracy; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Deep learning; Face recognition; Investigations; Kernels; Microprocessors; Multimedia; Multimedia Information Systems; Neural networks; Special Purpose and Application-Based Systems; Unmanned aerial vehicles
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-14603-x

Face recognition is most important knowledge whom many researchers are working on. The challenges with the existing approaches can allude to upscale input data from 48 × 48 to 64 × 64 which can increase load of processor. Another challenges could be using PCA and LDA to find most important data. It can be elapsed more time. Therefore, we proposed a network to solve this challenges. Proposed network constitutes 4 convolutional Layers with 1 and 3 Max-pooling and Average-pooling respectively. Second, we apply ADAM optimizer to enhance network accuracy. After designing our network, we investigate impact of the dimension of kernel size and number of filters with different padding and stride on accuracy which was led to find optimal kernel size and number of filter. Eventually we discuss about convergences of each kernel size and number of filter. Dataset used for training and testing is going to be Faces96 with 150 classes.