Abstract

Today, there is a lot of digital image material on the Internet. As the need for better image search tools grows, new ways must be found to solve the problem of telling the difference between question pictures and received images. This study presents a new technique to enhance the searching of images in retrieval based on image content. The technique involves using several deep neural network methods, such as ResNet50, InceptionResNetV2,VGG19, InceptionV3, DenseNet121, Xception, and Improved CNN. These models make it easier to find interesting photos. The study tested the technique on the Dataset and a dataset from a Kaggle competition, using metrics like the confusion matrices, F1 scores, and mAP to assess the model performance. The tests show that the proposed system works very well, taking away some of the problems that come with picture searches being unclear and showing how well it can find appropriate images.