Method for the recovery of indexed images in databases from visual content

Abstract

The techniques of content-based image recovery (CBIR) provide a solution to a problem of information retrieval that may arise as follows: from an image of interest to recover or obtain similar images from among those present in a large collection, using only features or features extracted from said images Banuchitra and Kungumaraj (Int J Eng Comput Sci (IJECS) 5 (2016) [1]). Similar images are understood as those in which the same object or scene is observed with variations in perspective, lighting conditions or scale. The stored images are preprocessed and then their corresponding descriptors are indexed. The query image is also preprocessed to extract its descriptor, which is then compared to those stored by applying appropriate similarity measures, which allow the recovery of those images that are similar to the query image. In the present work, a method was developed for the recovery of indexed images in databases from their visual content, without the need to make textual annotations. Feature vectors were obtained from visual contents using artificial neural network techniques with deep learning.