Various Approaches to perceptual image hashing systems-A Survey*

The era of digital space has increased the chances of illicit use of digital content exponentially, thus posing innumerable threats to legit content. This issue is tackled efficiently using the perceptual image hashing systems. This study provides an overview of the previous and current literature on the variety of techniques that are employed for the creation of perceptual hash. It also categorizes the kind of manipulation that is investigated in the literature. Moreover, the framework, different evaluation criteria, and performance metrics are also discussed. The paper presents future recommendations and drawn conclusions from the surveyed techniques.

I. Introduction

Manipulation in digital images has been increasing day by day with the widespread access to social media and editing tools. The use of these tools makes editing and changing the content of the images much easier and less traceable. This poses an issue for the authenticity of both the content and the user. Hence, there needs some technique that ensures the validity of the content and more, to withdraw the right one for retrieval purposes. The usual cryptographic hash functions aren't the kind that can be used as it generates highly different hash vector for slight changes in the images. The other method such as digital watermarking [1] can be used, however, it needs to add a watermark in the host image that sometimes can be removed from the image and thus making it less applicable and need extra information for the image. The perceptual image hashing (PIH) algorithms are exploited for the same purpose. PIH algorithms outdraw the relevant features from the image itself and generate efficient hash vectors that are resistant to slight incidental changes, viz. filtering, noising, JPEG compression, etc. These hash vectors can be utilized in various applications, viz. image/video indexing [2], similar image detection [3], detection of tampering in images [4], [5], retrieval of near-duplicate images [6], etc. In particular, the hash functions can be unkeyed and keyed [7], where the keyed one requires the use of a key to create the hash vector.