CBIR Using Features Derived by Deep Learning

In a Content Based Image Retrieval (CBIR) System, the task is to retrieve similar images from large database given query image. The usual procedure extract some useful features image, and which have set of features. For this purpose, suitable similarity measure chosen, with high scores are retrieved. Naturally choice these play very important role in success system, level required reduce semantic gap. paper, we propose use derived pre-trained network models deep-learning convolution trained for image classification problem. This approach appears produce vastly superior results variety databases, it outperforms many contemporary CBIR systems. We analyse retrieval time method, also pre-clustering based on above-mentioned yields comparable much shorter most cases.