Deep neural combinational model (DNCM): digital image descriptor for child’s independent learning

• Published in: Multimedia tools and applications 2022-09, Vol.81 (21), p.29955-29975
• Main Authors: Naqvi, Nuzhat, Islam, M. Shujah, Iqbal, Mansoor, Kanwal, Shamsa, Khan, Asad, Ye, ZhongFu
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Digital imaging; Learning; Multimedia Information Systems; Preschool children; Qualitative analysis; Special Purpose and Application-Based Systems
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-12291-7

This project is an endeavor to address preschool children’s independent learning. Currently, technology is invading our lives, and working parents are busily overpowering their social setup. As a result, the art of preschool education to young children has become rare or vanished. Automatic image descriptors (captioning models) have recently shown their effectiveness, motivating us to utilize such models for address purposes. Unfortunately, developed image descriptors produce only complex and generic visual descriptions irrelevant to children’s understanding. Therefore, it is important to have a suitable image descriptor as teaching material for young children at the initial educational stage. To fill this gap, we introduced a novel digital image descriptor and 3k-Flickr-SDD dataset using smart augmentation that originally extracted and labeled solitary dogs’ images from Flickr8k and Stanford Dogs Dataset (SDD) datasets. The newly developed 3k-Flickr-SDD dataset split further into two versions, making it meet the standard experimental requirements. The proposed method accumulates Convolutional Neural Networks (CNNs) for image contents extraction, whereas; Long Short-term Memory (LSTM) language model customizes to generate understandable and attractive text from Dogs images. We performed the quantitative and qualitative analysis; the finding reveals that the proposed model outperforms in contrast to existing models.