Near duplicate detection and retrieval of images using machine learning techniques

Abstract

Tremendous advancement in the growth of image editing / manipulating software leads to reproduction of thousands of images from genuine images resulting in numerous Near-Duplicate (ND) images. ND images are variants of original image with some transformations/ manipulations/forgeries in it. The illegal copies of images should be identified to protect copyright enforcement and to reduce redundancy. The detection of ND images is challenging due to the multitude of possible variations of images. It is a challenging task because the target images are usually modified by editing the original image through series of photometric and geometric transformations. These modifications make it difficult to establish correspondences between two ND images. Another important challenge in Near-Duplicate Detection (NDD) is the problem of multiple manipulations/ tampering. NDD methods achieve their goals by measuring the similarity between the features of the query and the target. So, exploiting effective features is one of the most fundamental tasks. The key issues of identifying suitable features are robustness, discriminability, variations due to transformation, scaling and rotation, storage requirements and computation complexity. Until now, a single image feature has hardly been able to achieve good performance in all these aspects at the same time. Given a query image, the objective is to find its ND versions in a large scale image database. The best solution is offered by extracting the features using PCNN. Also PCNN based NDD method is tested with the biomedical images. For the proper management of medical images and for clinical decision making, Content-Based Medical Image Retrieval (CBMIR) system emerged. Here, the physician can point out the disorder present in the patient report by retrieving the most similar report from related reference newline