Deep CNN based online image deduplication technique for cloud storage system

Online image detection is one of the most critical components of an image deduplication technique for an efficient cloud storage system. Although extensive research has been conducted in this field, the problem still remains challenging. Deep learning techniques have achieved significant success in solving a variety of computer vision issues and have high potential in image deduplication techniques. Deduplication is an efficient method in a cloud storage system that minimizes redundant data at the file or sub-file level using cryptographic hash signatures. Although significant research on offline image deduplication techniques have been reported, yet limited research is available on online image deduplication techniques. Online image matching accuracy and performance has been a major challenge for online image deduplication techniques to detect exact or near-exact images using feature extraction techniques. These first use feature extraction techniques to extract image features and then match these image features to detect duplicate images. In this paper, we have proposed a Deep CNN based online image deduplication technique for a cloud storage system to detect exact and near-exact images using cross-domains, even in the presence of perturbations in the form of blur, noise, compression, lighting variations and many more. The experimental results show that our proposed deep CNN for online image deduplication technique outperforms in terms of image matching accuracy and performance. The paper also proposed a Hot Decomposition Vector (HDV) for image patch generation to store efficiently dissimilar parts of near-exact images. The experimental results demonstrate that HDV exhibits higher and stable image matching accuracy in all three types of image deformations with relatively small computation time.