Title: Feasibility of conventional neural networks for content-based image retrieval in big data

Authors: M.A. Muthiah; E. Logashanmugam

Addresses: Department of Electronics and Communication Engineering, Sathyabama Institute of Science and Technology, Jeppiaar Nagar, Chennai 600 119, India ' Department of Electronics and Communication Engineering, Sathyabama Institute of Science and Technology, Jeppiaar Nagar, Chennai 600 119, India

Abstract: In real-time applications, content-based image retrieval (CBIR) has to be done with high level of accuracy and less latency. In this paper, feasibility of conventional neural networks for image retrieval is studied. A research database consisting of 8,000 images in 102 different categories is considered. Four sets of image features are obtained, i.e., moments (up to order 6) from the intensity of the images, moments from the discrete cosine transform (DCT) coefficients of the images, moments from each of the four sub-bands of the wavelet decomposed image (reverse biorthogonal wavelets) and moments from the discrete Walsh Hadamard transform (DWHT) coefficients of the images. Different sets of the dataset are used for training, testing and validation of the networks. The performance is measured in terms of accuracy. From the analysis, it is found that PNN provides the highest accuracy for DWT features.

Keywords: content-based image retrieval; CBIR; discrete cosine transform; DCT; discrete Walsh Hadamard transform; DWHT; discrete wavelet transform; DWT; back propagation network; BPN; general regression neural network; GRNN; probabilistic neural network; PNN; pattern recognition network; probability of detection.

DOI: 10.1504/IJPSPM.2021.118697

International Journal of Public Sector Performance Management, 2021 Vol.8 No.3, pp.253 - 263

Received: 29 Jun 2019
Accepted: 21 Nov 2019
Published online: 03 Nov 2021 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article