Article Preview
Top1. Introduction
With the diversification of data storage in public areas, it becomes more and more necessary for users to efficiently protect their own data (texts, images, or videos) by using their own tools in addition to the ones offered by service providers. Traditional standard encryption systems are commonly used to protect data (e.g. DES, 3DES, or its successor AES, etc.). However, it is argued (for instance, in (Massoudi et al., 2008)) that these encryption systems, which have been originally developed for text data are less suitable for securing images mainly because they consist in putting in the whole image data into a standard encryption system without considering its specific nature. One issue (Krikor et al., 2009) is that all symbols in the content are of equal importance are argued non optimal for securing images. Another issue which will be addressed in this paper is with performance. Full encryption algorithms can be time consuming once we require the speed of the process with limited calculation resources environment like the ones available in a laptop.
Other works are proposing other methods for securing images, in particular methods called selective encryptions (SE) which are the focus of this work. SE consists in applying encryption to a subset of the original content with or without a preprocessing step. The general approach in Figure 1 is to separate the image content into two fragments (or parts). The first fragment is to be confidential and will be encrypted. The main goal of SE methods is to reduce the amount of data to be encrypted and take as little storage as possible while achieving a required level of security. The tradeoff is to make the confidential fragment as small as possible in order to reduce processing time while keeping the image secure enough to comply with the requirements of a given specific use case. It is the task of the preprocessing step (in Figure 1) to resolve this tradeoff and separate the image in two fragments or parts. The second fragment is intended to be public and unencrypted as such, this fragment should not be sufficient to reveal or restore the full information. Moreover, it is intended to take most of storage space.
Figure 1. General concept of selective encryption
For uncompressed images like bitmaps, the most important visual characteristics of an image are to be found in the low frequencies while details are situated in the higher frequencies. Studies on the HVS (Human Visual System) have confirmed that human are more sensitive to lower frequencies than to higher ones (Puech & Rodrigues, 2005). This is why most SEs select and encrypt low frequencies in the confidential fragment rather than high frequencies. However, it is known that image sharp details reside in high frequencies; this means that sometimes, the public fragment can unveil information. We will discuss this point later since it led us to define a new design for SE with a strong level of protection.
There exist methods to selectively encrypt values in frequency domain (Krikor et al., 2009) to protect bitmap images. However, although providing a good level of protection, these methods actually suffer from performance issues, a major impediment which made SE difficult to use. These performance issues will be extensively addressed in this paper.
We are interested in image protection within limited calculation resources environments like desktops, laptops, or mobile devices such as latest tablets or even smartphones. In this work, bitmap files are initially stored on a laptop. Two new methods of selective encryption in the frequency domain are presented, called first level of protection and strong level of protection. They both use a GPU (Graphic Process Unit) to provide with the necessary acceleration, shifting the heavy computation burden from CPU to GPU. They aim at speeding up SE method by using calculation resources of both CPU and GPU available on a laptop.