The affordability and accessibility of digital image sensors and internet technologies have led to the establishment of extensive picture databases for various applications. The demand for efficient picture retrieval techniques that fulfil user requirements is underscored by the proliferation of these image repositories. Considerable efforts have been invested in refining content-based, or semantic-based, image retrieval methods, aiming to overcome the semantic gap between the attributes of low-level image and human visual perception. This study introduces the development and implementation of Semantic-Based Image Retrieval (CBIR) employing the VGG19 Model and ELU (Exponential Linear Unit Activation Function) in response to the burgeoning research interest in this domain. Additionally, to stimulate further research, this investigation provides CBIR architecture overview, contemporary methods of low level feature extraction, machine learning techniques, similarity metrics, and performance evaluation metrics. Recent advancements in deep learning have yielded remarkable outcomes. The effectiveness of a classification system hinges on the feature extraction quality of an image; whereby higher-quality features correspond to improved accuracy. Despite the impressive performance of many deep learning-based systems in image classification tasks, their capacity to extract comprehensive image information remains limited due to various inherent challenges, consequently compromising overall classification accuracy. Leveraging the Corel image dataset, this study to refine semantic-based image retrieval and classification methodologies.