A CNN-inspired Reverse Search Engine for Content-based Image Retrieval
Shilpa Marathe, Sirshendu Arosh, Tamal Mondal · 2022
The Reverse Search Engine enables the search and retrieval schemas of images which are visually comparable to a specific query image by utilizing the features of the images that illustrates the visual content of provided images. This contribution creates a contextual schema for recovering photographs that are indexed in distributed databases from the visual information without the requirements of linguistic annotations. We implement the generic Convolutional neural network (CNN) with deep learning strategies for retrieving feature vectors from visual information. To build the global descriptors, the pre-trained CNN architectures i.e., ResNet50, InceptionV3, and InceptionResNetV2 were recommended. These techniques outperformed in the existing methods like Metaheuristic Algorithm, Gabor Wavelet, CNN-SVM which have been compared with the proposed technique i.e., Cartoon Texture Algorithm in CBIR. We have observed that they are similar to others based on machine learning approaches, with comprising of the conventional state-of-the-art in the Content-dependent Information Retrieval.