A Deep Learning Framework on Generation of Image Descriptions with Bidirectional Recurrent Neural Networks

Thomas J. J., Pillai N.

1st International Conference on Intelligent Computing and Optimization (ICO), Pattaya, Tayland, 4 - 05 Ekim 2018, cilt.866, ss.219-230 identifier identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 866
  • Doi Numarası: 10.1007/978-3-030-00979-3_22
  • Basıldığı Şehir: Pattaya
  • Basıldığı Ülke: Tayland
  • Sayfa Sayıları: ss.219-230
  • Anahtar Kelimeler: Deep learning, Convolutional Neural Network, Data augmentation, Malaysian food chain
  • Orta Doğu Teknik Üniversitesi Adresli: Hayır

Özet

The aim of the paper is to develop a deep learning framework for a model that generates natural descriptions of pictures (data) and their sections so as to search out a lot of insights. Image recognition is one in all the promising applications of visual objects. In this study, a small-scale food image data set consisting of 5115 pictures of fourteen classes and an eight-layer CNN was made to acknowledge these pictures. CNN performed far better with associate degree overall accuracy 54%. The approach influences information sets of images and their patterns bi directional recurrent neural network (BRNN) will concerning the intern-model correspondences between prediction and visual information for calorie estimation. Data expansion techniques were applied to extend the dimensions of trained images, that achieved a considerably improved accuracy of 74% stop the over fitting issue that occurred to the CNN while not misclassified.