Effective image retrieval based on an optimized algorithm utilizing a novel WOA-based convolutional neural network classifier

Abstract

Traditional image retrieval from database based on available specific algorithms has drawbacks, including laborious image annotation, poor feature extraction, an inability to handle complicated queries, longer processing times, and less precise results. The study of CBIR is a current area in image processing. The most effective and efficient method for identifying and extracting images or patterns is deep learning. Finding the ideal CNN hyper-parameter value is a significant challenge, and using nature-inspired methods and the Whale Optimization Algorithm (WOA), it is possible to identify the ideal capability (NIAs). With the help of the best feature extraction technique, this effort aims to efficiently retrieve photos. An MRFODE approach is utilized in the pre-processing of this study to eliminate the undesired data that was present in the dataset. Feature extraction is used to extract features like texture and colour after pre-processing. Here, the statistical and colour features are referred to as image intensity-based colour features, while the texture feature is classified as a grey-level co-occurrence matrix. K-means clustering is used to group these features into groups for label creation. The features are classified using a unique WOA-based convolutional neural network, and were optimized using an MRFODE. The performance assessed based on sensitivity, specificity, precision, recall, retrieval, and recognition rate utilized a novel deep learning technique for CBIR using a CNN optimized through WOA. The WOA applied in two levels of CNN at the convolutional layer and fully connected later on to the CIFAR10 dataset provide better performance results compared to other reported models.

URN:NBN:sciencein.jist.2023.v11.523