Performance Evaluation of GaussVLAD for Efficient Image Retrieval in E-Health Applications

As electronic health (e-Health) systems advance, efficient and accurate image retrieval is crucial for applications like visual navigation, patient monitoring, and resource management. Traditional retrieval methods often fall short in meeting the demands for real-time performance and robustness in healthcare environments where signal stability may be compromised. This paper presents GaussVLAD, a streamlined feature encoding approach tailored for efficient image retrieval in e-Health applications. GaussVLAD models local features as Gaussian mixture distributions and employs the Wasserstein distance for effective similarity measurement, yielding high-precision feature encoding. By assuming feature dimensional independence, it constructs independent Gaussian components to enhance encoding efficiency. Built upon the EfficientNet-B3 backbone and augmented with a cross-attention mechanism, GaussVLAD demonstrates significant improvements in retrieval speed and accuracy in complex visual scenarios. Experimental evaluations confirm GaussVLAD's effectiveness in providing reliable support for e-Health systems in challenging indoor environments.