The problem of compressing multidimensional digital signals (images) stored as data arrays in the computer memory or transmitted through digital channels was stated more than 40 years ago. Since then, this type of data processing has always been in the focus of researchers' attention (see review articles in the Proceedings of IEEE, 3.1967, 7.1972, 3.1980, 3.1981, 4.1985; the collection Image Transmission Techniques: Redundancy Reduction, Moscow, Radio i Svyaz, 1983; the monograph Digital Image Processing by W. Pratt, Moscow, Mir, 1983; Data Compression Methods by D. Vatolin, А. Ratushnyak, M. Smirnov, and V. Yukin, Moscow, Dialog-MIFI, 2002, and others). To date, there have been several major classes of compression methods employed for solving different problems of image transferring and storage: differential coding methods, transform coding methods (including methods based on wavelet transforms), interpolation methods (including hierarchical ones), methods based on image structural decomposition (segmentation), methods based on vector quantization, and fractal methods. Each of the studied approaches of image compression has its advantages; however, neither of them is free of drawbacks. In practice, compression of two-dimensional images has been most widely implemented using the JPEG standard (method and data format), which is based on the cosine transform, and the JPEG2000 standard, which uses a hierarchical wavelet transform of the 2D signal. Current methods most widely used for processing 3D digital video signals include MPEG-2 and MPEG-4 standards, which combine adaptive inter-frame interpolation and compression of 2D (reference and difference) images using methods similar to JPEG and JPEG-2000 respectively. Except for some differential encoder versions, all the methods listed above have been initially developed for domestic or technical application, and therefore their characteristics are not quite suitable for ERS systems. Nonetheless, many of these methods have been implemented in the on-board hardware of the state-of-the-art Russian and foreign ERS satellites. For instance, the differential image encoding method is used on the satellites SPOT-1,2,3,4 (France, launched in 1986–1998), IKONOS (USA, 1999), QuickBird (USA, 2001), Resurs-DK (Russia, 2006), WorldView-1 (США, 2007). The JPEG method is used on the satellites SPOT-5 (France, 2002), IRS-P5 (India, 2005), Beijing-1 (China, 2005), Cartosat-1,2 (India, 2005, 2007), whereas JPEG-2000 is used on IMS-1 (India, 2008), X-SAT (Singapore, 2009), and PLEIADES-HR (France, 2010). Radar satellites Radarsat-2 (Canada, 2007) and TerraSAR (Germany, 2007) use vector quantisation methods. On-board video equipment of the above-listed satellites produce half-tone (panchromatic) or multi-range (3–8 spectral ranges) images of the Earth surface, so the use of the 2D image compression methods following necessary upgrades seems natural. However, with the advent of new satellites for hyperspectral ERS, the researchers are faced with entirely new challenges of developing dedicated methods for 3D digital signal compression.
Wide acceptance and use of digital images as carriers of ERS data, which form a basis for decision-making at municipal, regional, and state administration levels, poses the problem of revealing of possible data forgery as a way of reducing the decision-making risks. Digital imagery protection from unauthorized copying, interference, and distribution has been an important aspect in providing the security of information systems associated with processing and analysis of visualisable information. Currently, there have been two main techniques to identify the deliberate interference in digital imagery in general and in the ERS data in particular (M. Sridevi, C. Mala): active and passive ones. A key element of an active approach to identifying the deliberate interference is the use of digital water marks (DWM). With this approach, the DWM is supposed to be built into the image while the image is being recorded. As distinct from the active approach, the passive one makes no use of DWM, being based on the assumption that interference traces maybe identified through the computer analysis of the image. Within the active approach, two following protection techniques have been traditionally used [Cox, 1999], [Kutter, 1999].
1. Digital image protection from unauthorized copying. According to Pereira (2000) and Cox (2004), a major shortcoming of the existing methods is their computational complexity when processing large-format imagery. Besides, the algorithms’ robustness from the cryptography and steganography perspectives [Furht, 2005] has not been sufficiently studied.
2. Digital image protection from modification. Major shortcomings of the existing methods include the lack of research concerning their resistance to cryptographic attacks; the lack of adapted methods of cryptoanalysis and steganoanalysis, which would allow the resistance of the said DWM-imbedding algorithms to common types of attacks to be reliably evaluated.
The majority of well-known works (S. Prasad, B. Mahdian, A. C. Popescu, and others) dealing with the passive image protection are aimed at identifying a specific type of image distortion – attacks. A dedicated identification algorithm is specifically developed for each attack. While not putting in doubt the importance of this research area, below we mention its drawbacks.
Since the algorithms are developed for arbitrary images, they disregard specific image acquisition techniques. In particular, alongside imagery, the ERS data contain additional information relating to the remote sensing geography and time, satellite or recording hardware type, camera orientation, etc.
A great variety of different algorithms for deliberate image interference fail to answer which algorithm(s) and which way/when should be utilized in order to identify the deliberate interference in the specific ERS data or an attack.
Specifically, the proposed project aims at developing and studying methods, algorithms, and information technology for two classes of lossless and near-lossless image compression:
1. An authoring method for hierarchical grid interpolation, which will be modified and optimised for processing ERS data and include adaptive (learned) procedures for nonlinear processing in the feedback loop between the hierarchical levels. The method is expected to have low computational complexity of both compression and decompression, and thus be suited for both on-board use and in the ground data processing and archiving complexes. In the on-board version, the method will be seamlessly complemented with algorithmic tools for unbuffered stabilisation of output generation speed and low-redundancy compressed data protection from random noise in a digital channel.
2. A compression method based on authoring fast algorithms of 3D discrete orthogonal transforms (DOT), adaptive selection and quantisation of transforms, and controllable ‘squeezing’ of the compression error field in order to ensure the lossless or near-lossless process. It is expected that with the proper choice of the DOT basis, the method will offer a record-breaking efficiency in terms of ‘compression degree – retrieval error’, also providing accelerated (comparing to decompression) extraction of arbitrary cross-sections of the 3D cube under compression.
Regarding the data protection, it can be noted that the compression procedures themselves may be looked upon as means of cryptographic ERS data protection because they generate datasets structured in a complex way. However, this type of protection is removed following the decompression, which inadequate for a variety of threads associated with unauthorized copying, distribution, or modification of data. Because of this, the project will also study special steganographic methods of data protection in which identifying information in the form of DWM is embedded into the image. The image protection DWMs must satisfy the following basic requirements:
- DWMs must be invisible and/or have no effect on the results of automatic image analysis (including recognition)
- be cryptographically strong, unable to be accidentally or maliciously deleted, i.e. be resistant to the attacks they are supposed to protect from
- be distributed over a 3D data volume while allowing image fragments and arbitrary 2D sections to be protected.
Within the proposed project, unique authoring approaches and new methods for DWM embedding and retrieval are developed based on spectral recursive and block image processing, allowing parallel processing. To protect data from unauthorized copying, reliable DWMs are proposed, with the protection information embedded into the 3D image spectrum and being resistant to basic data transformations, including compression, filtering, radiometric and geometric correction, fragmentation, and 2D sectioning. At the same time, insecure DWMs are introduced to identify local interference into the protected data and indicating possible image forgery, thus showing that the initial data underwent distorting procedures of filtering, radiometric or geometric transformation, retouch, patching, recompression, etc.
Overall work plan and expected results for the term of the project:
- 2014: development of new information technology for ERS data compression, experimental software implementation and efficiency assessment
- 2015: development of new information technology for ERS imagery protection, experimental software implementation, and efficiency assessment
- 2016: adaptation, refinement, and experimental study of the effectiveness of methods, algorithms, and information technology for ERS data compression and protection with regard to different variants of their use in aerospace systems of Earth remote sensing and in applied problems addressed with use of hyperspectral data.