Feature Fusion Tactics in Gastrointestinal Endoscopic Image Retrieval Using Deep Features

Learning outcomes in the vast field of image retrieval (IR) can enhance the effectiveness of IR systems. Medical imaging is crucial because it can quickly and precisely diagnose a patient’s disorder, monitor their response to treatment, and manage their condition. Therefore, it is essential to develop a reliable and effective medical image retrieval system. Endoscopic colored images have a single mode of representation with identical attributes, making it difficult to exploit low-level features. In this work, feature vectors extracted from deep learning networks such as ResNet101 and DenseNet201 are fused together in two ways, namely concatenation and summation, using the canonical correlation analysis-based feature fusion technique. The dimensionality of the feature space chosen for the analysis was reduced using principal component analysis. The effectiveness of this methodology was examined using the well-known medical image dataset, Kvasir. The experimental performance analysis demonstrated that the retrieval strategy is extremely effective when integrating various feature spaces.