Araştırma Makalesi

Two-Stage Sequential Losses based Automatic Hash Code Generation using Siamese Network

5 Ekim 2020
PDF İndir
TR EN

Two-Stage Sequential Losses based Automatic Hash Code Generation using Siamese Network

Abstract

Today, large-scale image retrieval methods are used for fast access to increasing visual information. Hashing methods are an image retrieval approach that provides computationally effective and high-speed access. Hash codes created with hand-crafted features extracted from images in the past are now built with deep learning architectures. Convolutional neural network (CNN)'s an impressive performance using optimized features strengthens this trend day by day. However, it contradicts the hash codes consisting of '0' and '1' values and the procedure of updating the parameters and output of CNN architectures. To solve this problem, solutions such as bringing the bits in the CNN output to '0' and '1' values with a threshold value, and converting the output bits to binary values with the help of loss functions in the CNN output are presented in the literature. It is very useful for studies involving loss function solutions for end-to-end training. However, in these studies, loss weights are generally empirical. In order to solve this situation, a framework consisting of two steps is suggested in this study. In the first stage, Euclidean distance is used in the output of a CNN architecture that is trained in a pairwise manner. After reducing the distance between the two images below the specified value, in the second step, the network parameters are transferred, and the output is drawn to binary values with binarization loss. Thus, binary hash codes are obtained automatically for each image without using any additional weight.

Keywords

Destekleyen Kurum

TÜBİTAK

Proje Numarası

120E018

Teşekkür

This research is funded by the Scientific and Technological Research Council of Turkey (TÜBİTAK) under grant number 120E018.

Kaynakça

  1. Ashraf, R., Ahmed, M., Ahmad, U., Habib, M. A., Jabbar, S., & Naseer, K. (2020). MDCBIR-MF: multimedia data for content-based image retrieval by using multiple features. Multimedia tools and applications, 79(13), 8553-8579.
  2. Öztürk, Ş. (2020). Stacked auto-encoder based tagging with deep features for content-based medical image retrieval. Expert Systems with Applications, 161, 113693.
  3. Demir, B., & Bruzzone, L. (2015). Hashing-based scalable remote sensing image search and retrieval in large archives. IEEE transactions on geoscience and remote sensing, 54(2), 892-904.
  4. Alsmadi, M. K. (2020). Content-Based Image Retrieval Using Color, Shape and Texture Descriptors and Features. Arabian Journal for Science and Engineering, 1-14.
  5. Mukherjee, A., & Gaurav, K. (2016). Content based image retrieval using GLCM. International Journal of Innovative Research in Computer and Communication Engineering, 4(11), 20142-20149.
  6. Liu, X., He, J., & Lang, B. (2014). Multiple feature kernel hashing for large-scale visual search. Pattern Recognition, 47(2), 748-757.
  7. Ng, W. W., Li, J., Tian, X., Wang, H., Kwong, S., & Wallace, J. (2020). Multi-level supervised hashing with deep features for efficient image retrieval. Neurocomputing.
  8. Li, Y., Zhang, R., Miao, Z., & Wang, J. (2019, October). CapsHash: Deep Supervised Hashing with Capsule Network. In 2019 11th International Conference on Wireless Communications and Signal Processing (WCSP) (pp. 1-5). IEEE.

Ayrıntılar

Birincil Dil

İngilizce

Konular

Mühendislik

Bölüm

Araştırma Makalesi

Yayımlanma Tarihi

5 Ekim 2020

Gönderilme Tarihi

29 Eylül 2020

Kabul Tarihi

30 Eylül 2020

Yayımlandığı Sayı

Yıl 2020

Kaynak Göster

APA
Öztürk, Ş. (2020). Two-Stage Sequential Losses based Automatic Hash Code Generation using Siamese Network. Avrupa Bilim ve Teknoloji Dergisi, 39-46. https://doi.org/10.31590/ejosat.801927

Cited By