Content Based Image Retrieval using VGG19 KPCA and ELM

Abstract

Content-Based Image Retrieval (CBIR) systems aim to retrieve visually similar images based on low-level image features. However, a major challenge in conventional CBIR methods is the semantic gap, disconnect between low-level visual features and the high-level semantic meaning perceived by humans. This research tackles the semantic gap and retrieval inefficiency by proposing a novel CBIR framework based on deep learning. The proposed method works like this: First, it uses a well-known convolutional neural network called VGG19 to extract rich, abstract visual features from images. Next, it uses an algorithm called Kernel Principal Component Analysis (KPCA) to reduce the dimensionality of the feature vectors, which are huge and difficult to manage. Finally, it uses another machine learning algorithm called Extreme Learning Machine (ELM) to classify images based on the features extracted by VGG19. The experimental details of the proposed technique indicate that it outperforms state of the art CBIR methods in terms of the performance evaluation metrics.