Content-Based Image Retrieval using Convolutional Neural Networks

Searching a collection of images that have similarities with input images, without knowing the name of the image, makes a search system that applies the concept of content-based image retrieval (CBIR), is very necessary. In general, CBIR systems use visual features such as color, image edge, texture, and suitability of names in input images with images in the database. The method for classification is convolutional neural networks (CNN), while retrieval with cosine similarity. Dataset is divided into 5 masterclasses, each masterclass has 5 subclasses. The class used for retrieval is a masterclass, where the images of each large class are combined images of subclasses in the large class. From the experiments, we found that the CNN method has succeeded in supporting the retrieval task, by classifying image classes.