Validation: application to the Interleaved S+P

For validation purposes, the proposed QM coding scheme has been implemented in the Interleaved S+P codec. Each substream has been coded according to both symbol oriented and bitplane oriented coding schemes. Entropy coding is performed inline with the prediction process.

To compare both complexity and compression ratio, the set of images of the core experiment 2 of JPEG AIC standardization process has been used. For each image, 30 compressions have been performed from low to high bitrates. Both symbol and bit plane oriented coding have been used. Concerning the complexity benchmark, results have been obtained by performing each compression 50 times and taking the average time of these 50 compressions.

Compression ratio. Figure3.12presents the average compression ratio of both bit plane oriented and symbol oriented coding. An overhead between 0.5% and 3.5% can be observed for the proposed method against bit plane oriented coding. This result could be mainly improved by using a better context modeling. The context modeling proposed in this chapter is a preliminary study and additional work should be done to improve the results.

30 32 34 36 38 40 42 44 46 48 50 0 1 2 3 4 5 6 7 8 Rate (bpp) WPSNR (dB)

Symbol oriented coding Bit plane oriented coding

Figure 3.12: Compression ratios of QM encoding for both symbol oriented and bit plane oriented QM Coding

Complexity. Figure 3.13 presents the computational time of both bit plane oriented and symbol oriented coding for all images and three compressions. The average gain in computational time is

around 50% for the entropy coding process and 33% for the whole encoding process of the Interleaved S+P coder for low bitrate as well as for high bitrates.

0 500 1000 1500 2000 2500 3000 3500 0 50 100 150 200 250 Quantization factor

Encoding time (ms/Megapixel)

Symbol oriented coding Bit plane oriented coding

Figure 3.13: Compression ratios of QM encoding for both symbol oriented and bit plane oriented QM Coding

3.5

Conclusion

In this chapter, we presented three different generic tools for coding purposes. First, based on an inter- component adaptive decorrelation, multi-component compression framework has been presented. Results obtained show bit-rate saving from 0.20 up to 0.46 bpp have been observed. Considering colour spaces, using the RGB colour space together with the decorrelation process leads to the best compression results. However, for better scalability features, YDbDr colour space might be considered.

Secondly, a statistical analysis of predictive coders based on Laplacian distributions provides estimators of entropy and distortion. These tools can be then combined for advanced functionalities such as rate control and rate distortion optimization processes.

Finally, a symbol oriented QM coder was then proposed and compared to the classical bit plane oriented QM coder. Results in terms of complexity of the proposed scheme were shown to be better than bit plane oriented coding with a gain of about 60 %.

If the validation process has been conducted with the Interleaved S+P coder, as previously mentioned, the genericity of the presented techniques allows to envisage their integration into other predictive coders.

As we shown in this chapter, this generic coding toolbox should enhance compression perfor- mances as well as providing advanced Quality of Service (QoS) features based on statistical modeling of prediction errors. In particular, improving Rate / Distorsion features naturally enhances Quality of Experience. Even if the quality metric used in this study remains the debatable MSE, the related complexity insures realistic network based implementations.

Along with these coding-oriented tools, additional services has to be designed so that to guarantee an end-to-end QoS. In particular, the next chapter focuses on data integrity.

Content securization and Quality of

Service: preserving end-to-end data

integrity

Quality of Experience is not only based on the lone received image quality. Typically, medical images or high resolution art images usually embed private metadata that have to remain confidential. Huge amount of medical data are then stored on different media and are exchanged over various networks. If these embedded sensitive data are accidentally altered, even if the received image quality is sufficient, end-users can perceive the overall process as an inefficient one. As a consequence, techniques especially designed for these data are required so that to provide security functionalities such as privacy, integrity, or authentication. Multimedia security is thus aimed towards these technologies and applications [24]. In addition, to insure reliable transfers, flexible and generic scheduling and identification processes have to be integrated for database distribution purposes while taking into account secure remote network access together with future developments in network technologies.

Then, on one hand, content protection consists of preserving data integrity and masking data content. The commonly used methods to obtain these protections are respectively hashing and ciphering. On the other hand, the embedding of hired data aims to protect copyrights or add metadata to a document. Besides watermarking, steganography, and techniques for assessing data integrity and authenticity, providing confidentiality and privacy for visual data is among the most important topics in the area of multimedia security. Applications range from digital rights management to secured personal communications, such as medical materials.

Moreover, errors can occur in various ways when transmitting multimedia contents. Depending on the communication media, i.e. cable network, wireless network (wifi, cellphone network) or even during physical storage (hard disk, flash memory...), binary or packet errors can appear, the worst cases being potentially the wireless network followed by the cable network. In order to provide the best user experience, error concealment and robustness strategies are incorporated during both source coding and the transmission process. As for the channel coding, protection strategies depend on transmission conditions in order to ensure Quality of Service (QoS).

When considering the source coding side, this robustness is usually obtained by adding a given level of redundancy within the transmitted data. Under certain conditions this redundancy allows to recover the whole information even if some parts of the information are damaged. Hybrid approaches,

named joint source-channel coding, achieve robustness by combining protection strategies at both source and channel coding. However when the errors are too strong, or when data packets are definitely lost, some part of the information can be missing or unusable despite the transmission robustness and the added redundancy.

In this chapter, we address these two different services, namely content protection solutions through Interleaved S+P based mechanisms, and Quality of Services tools designed for MPEG-4 SVC or any scalable image coder. This chapter is organized as follows: section 4.1defined first the securization oriented application contexts within these works have been realized. Then we focus on two aspects of the data integrity. Section4.2provides joint data hiding and cryptography processes, while section 4.3described two specific tools among Quality of Service ones.

4.1

Application contexts and related ANR projects

I was involved in two different ANR (French National Research Agency) projects, both relative to securization and image coding topics. These projects, namely TSAR and CAIMAN deeply influenced my research when considering the content protection domain. The application contexts, inherent to these collaborations, have raised some concrete issues that we tended to alleviate. In this section, I briefly describe the purposes of TSAR and CAIMAN projects and my associated involvement.

TSAR project: Safe Transfer of high Resolution Art images - 2006/2009. The protection of digitized works of art still pose security problems when they broadcasted on-line. The goal of the project TSAR is then to transmit in a secure way high quality images. Within the TSAR project, five laboratories were involved, namely the C2RMF (Louvre, Paris), IETR (Rennes), IRCCyN (Nantes), LIRMM (Montpellier) and LIS (Grenoble).

Museums are supposed to undertake at least two essential missions [31]. First, they have to preserve their huge number of items and save them from damage. At the same time, museums play an active role in the spread of cultural knowledge and this educational objective leads them to widely communicate these materials. However, these two missions are somewhat contrary in nature because handling art items inevitably causes damage. To solve this major problem, museum research centers have introduced the “digital museum” concept [116]: digital versions of the original art items are collected in a database on a server accessible via the Internet.

For example, the National Gallery in London, the Tokyo University Digital Museum (through the Digital Museum 2000 project [167]) or the Chinese University Museum Grid [31] provide public access to their databases. However, users can only download low-resolution images. The application has been actually designed to prevent illegal copies of digitized data. The best way to achieve this consists basically of not transmitting high-resolution images. High-resolution information is therefore stored separately and reserved for a limited number of people.

In France, the C2RMF laboratory, connected to the Louvre museum, has digitized more than 300, 000 documents taken from French museums, in high resolution (up to 20, 000 × 30, 000 pixels). The resulting EROS database [149] is for the moment only accessible to researchers whose work is connected with the C2RMF. To widely open the database, the idea is to create a framework to integrate the different security solutions in order to secure the access to the images. This project then aims at transmitting hidden data in images of important sizes by means of steganography and watermarking techniques. This data hiding scheme will also be combined with cryptography and

compression methods in order to both increase the transfer security and to minimize the transmis- sion time. The application concerns secure transmission of high-resolution images of painting and archaeological objects (several gigabytes of data).

As far as the French TSAR project is concerned, the Interleaved S+P image coding method takes part into the framework. The objective is to design an art image database accessible through a client-server process that includes and combines a hierarchical description of images and dedicated content securization processes. Jean Motsch, who I co-supervised, has defended his dissertation in 2009 in connection with the TSAR project. I was nominated as the scientific responsible of IETR works.

CAIMAN project: Codage Avanc´e d’IMAges et Nouveaux services - 2009/2012. In collab- oration with Thal`es Communications, and ETIS, XLIM-SIC, and IETR laboratories, the CAIMAN project aims at providing new image coders able to reach a widespread use such as JPEG does, together with advanced functionalities. In particular, a user-friendly solution should be designed, while keeping a low computational complexity and more advanced features in the direction of some tools of JPEG2K.

Not only the compression efficiency will drive the adoption of the future generation of image codecs but more the new services it will provide. These considerations have been taken into account by the JPEG Committee for JPEG advanced Image Coding to which CAIMAN will contribute. This CAIMAN project is thus naturally closely linked to JPEG-AIC work (see section2.1).

The main objective of CAIMAN consists of studying a still image coder that jointly integrates in its design security aspects such as steganography, error-resilience, adaptation to the network and robustness to losses and Quality of Experience issues, without sacrifying to compression efficiency. The Interleaved S+P framework has been then proposed to the JPEG committee and related work led to contributions to the JPEG-AIC standard.

In particular, medical image concerns were addressed during this project. The Interleaved S+p coding scheme has been developed to face the secure transmission issues. Embedded functionalities such as adapted selective cryptography, human vision-based steganography coupled with Unequal Error Protection and error resilience tools have been designed so that to maintain good coding properties together with embedded Quality Of Service oriented system.

MPEG4-SVC video coding standard. The standard MPEG4-SVC was designed so that to add scalability. Technicolor took part to the standardization process of the framework, based in MPEG4- AVC coding scheme. However, the complexity of the proposed solution prevent from a widespread use of it. One of the identified bottleneck relies in the rate control mechanism. In collaboration with Technicolor, we provide an efficient one-pass rate control solution. Yohann Pitrey has then designed during his PhD work a low-complex solution, able to address both AVC and SVC frameworks.