Content-based image retrieval using feature weighting and C-means clustering in a multi-label classification framework
- PDF / 1,475,322 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 69 Downloads / 210 Views
THEORETICAL ADVANCES
Content‑based image retrieval using feature weighting and C‑means clustering in a multi‑label classification framework Samaneh Ghodratnama1 · Hamid Abrishami Moghaddam2 Received: 27 January 2018 / Accepted: 8 June 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract In this paper, a novel learning algorithm based on feature weighting is proposed to improve the performance of image classification or retrieval systems in a multi-label framework. The goal is to exploit maximally the beneficial properties of each feature in the system. Since each feature can separate more effectively some of the image classes, it is hypothesized that the weights of various features at some states can be traded off against each other. The training phase of the suggested algorithm is performed in two stages: (1) The input images are clustered using a supervised C-means method iteratively; (2) image features are weighted using a local feature weighting method in each cluster. These weights are determined by considering the importance of each feature in minimizing the classification error on each cluster. In the testing phase, the cluster corresponding to the query is found first. Then, the most similar images are retrieved in the multi-label framework using the feature weights assigned to that cluster. Experimental results on three well-known, public and international image datasets demonstrate that our proposed method leads to significant performance gains over existing methods. Keywords CBIR · Multi-label · Feature weighting · C-means clustering · KNN classification
1 Introduction Content-based image retrieval (CBIR) is a method to retrieve the most similar images to a query image. This system searches similar images using features such as color, texture, and shape. One of the most common methods that can be used for this purpose is the K-nearest neighbors algorithm (KNN) [1–3]. This algorithm returns the most similar image using a distance function within feature space. Retrieving similar images using only low-level features is one of the main shortcomings of the conventional content-based image retrieval (CBIR) systems. Among various techniques, machine learning such as deep learning [4–10], sparse coding for bag-of-words (BoW)-based approach [11, 12], Fisher vectors [13], etc., have been actively investigated as possible directions to bridge the semantic gap in the long term. This paper presents an effort to overcome * Hamid Abrishami Moghaddam [email protected] 1
Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran
Faculty of Electrical and Computer Engineering, K. N. Toosi University of Technology, Tehran, Iran
2
this drawback and proposes a CBIR approach in which the retrieved labels of images in the multi-label classification framework satisfy user expectations. It is also noteworthy to consider that images are in different classes and each of them would be more separable based on one particular kind of feature. For insta
Data Loading...
