Near Duplicate Image Pairs Detection Using Double-Channel Convolutional Neural Networks

Abstract

Measuring the image pair similarity is a fundamental task in computer vision. This paper illustrates a neural network model to accomplish the task and decide if the input pair is a near duplicate pair. Authors explore several convolutional neural networks and adopt the double-channel network on this task. The model achieves comparable results on benchmark datasets and well performs on the closely similar images pairs among them. Comparing with the conventional approaches, the network provides a straightforward function to measure the pair-wise similarity and utilizes the strong correlation meanwhile.