An Optimized Feature for Content Based Multimedia Image Retrieval System Using Deep Learning Approaches

The World Wide Web and developments in computer and multimedia technologies have led to increased picture databases and collections such as digital libraries, medical imageries, and art galleries, which collectively contain millions of pictures. Developing an efficient image retrieval system that can manage these enormous volumes of pictures at once is essential. The major goal of this study was to create a reliable system that could efficiently create, manage, and react to data. An effective tool for retrieving images was found to be the content-based image retrieval (CBIR) system, which allows users to query the system to retrieve their desired image from the image collection. In addition, the variety of pictures that users can access, and the expansion of online development and transmission networks have continued to increase. In this paper, we proposed employing an Improved Mobilenetv3 method for picture retrieval. To preprocess the images, we applied noise reduction with a median filter, normalization using the min-max normalization method, and contrast enhancement using Adaptively Clipped Contrast Limited Adaptive Histogram Equalization (ACCLAHE). Then, a Modified ResNet152V2 model was employed to extract detailed features related to shape, texture, and color. After that, the Quantum Chaotic Honey Badger Algorithm (QCHBA) was utilized to select the most relevant features, improving computational efficiency and performance. Finally, the images were classified using the Improved MobileNetV3 technique, which was optimized for high accuracy and efficiency. The performance of the image retrieval framework for content-based retrieval was improved by combining these techniques. Furthermore, the precision-recall value of the outcomes was computed to assess the effectiveness of the system.