Improved Learning of Combining LBP and Original Image Datasets for Feature Extraction and Classification Using CNN

Contenu principal de l'article

Pallavaram Venkateswar Lal, Uppalapati Srilakshmi, D. Venkateswarlu

Résumé

Face Recognition System is a computer application technology and it senses, tracks recognizes, or authenticates human appearances from any seized image or video via digital camera. Even though a ton of progress has been made in face detection, identification, and security recognition, but issues are preventing the advancement to reach or outperform human-level exactness in human facial appearance, such as noise in face images, divergent illumination circumstances, pose scale. LBP is used in this paper to perceive characteristics of facial texture to diminish the influence of lighting up and disposition. CNN and skip affiliation are used to lessen the training time for parallel convolution processing and progress the accurateness of classification. The proposed approach predominantly considers the FR reliant on parallel automatic extraction of features from LBP image dataset and original image dataset and classify the data using Deep Convolutional Neural Networks (DCNN). In the first part, LBP is utilized first to examine the texture of the information from the image dataset, later, the training dataset of combining the original image data set and LBP image dataset is applied to the CNN for automatically further extraction of features and lastly classification is performed. The proposed methodology achieved training accuracy 99.99%, validation accuracy 99.77%, and finally achieved test accuracy of 99.28% in 10 epochs. The investigational results proved that the projected methodology outperforms compared standing methodologies.

Renseignements sur l'article

Rubrique
Articles