Experiments of Image Classification Using Dissimilarity Spaces Built with Siamese Networks.

Loris Nanni,Sheryl Brahnam,Alessandra Lumini,Giovanni Minchio,Gianluca Maguolo

doi:10.3390/s21051573

Loris Nanni, Sheryl Brahnam + Show 3 more

Open Access

https://doi.org/10.3390/s21051573

Copy DOI

Abstract

Traditionally, classifiers are trained to predict patterns within a feature space. The image classification system presented here trains classifiers to predict patterns within a vector space by combining the dissimilarity spaces generated by a large set of Siamese Neural Networks (SNNs). A set of centroids from the patterns in the training data sets is calculated with supervised k-means clustering. The centroids are used to generate the dissimilarity space via the Siamese networks. The vector space descriptors are extracted by projecting patterns onto the similarity spaces, and SVMs classify an image by its dissimilarity vector. The versatility of the proposed approach in image classification is demonstrated by evaluating the system on different types of images across two domains: two medical data sets and two animal audio data sets with vocalizations represented as images (spectrograms). Results show that the proposed system’s performance competes competitively against the best-performing methods in the literature, obtaining state-of-the-art performance on one of the medical data sets, and does so without ad-hoc optimization of the clustering methods on the tested data sets.

Highlights

The image classification system presented here trains classifiers to predict patterns within a vector space by combining the dissimilarity spaces generated by a large set of Siamese Neural Networks (SNNs)
The most common image classification strategy involves extracting features from samples and training classifiers to discriminate them within the selected feature space
The generic image classification system proposed here is tested and compared with the standalone classifiers and the state-of-the-art using four data sets representing two classification tasks: audio classification, with audio represented by spectrograms, and two medical data sets

Summary

Introduction

The most common image classification strategy involves extracting features from samples and training classifiers to discriminate them within the selected feature space. Another less common method involves training patterns within one or more (dis)similarity spaces. The idea of (dis)similarity, or semblance, is grounded in human learning and plays a fundamental role in theories of knowledge and behavior [1]. For this reason, (dis)similarity provides a sound theoretical basis for building learning algorithms. Because classification within (dis)similarity spaces does not require access to a sample’s features, the sample space can be any set and not limited to the Euclidean space as long as the (dis)similarity function is well defined for any pair of samples [8]

Methods

Results

Conclusion