Leith Abed


With the capability of storing huge volumes of data over the Internet, cloud storage has become a popular and desirable service for individuals and enterprises. The security issues, nevertheless, have been the intense debate within the cloud community. Significant attacks can be taken place, the most common being guessing the (poor) passwords. Given weaknesses with verification credentials, malicious attacks have happened across a variety of well-known storage services (i.e. Dropbox and Google Drive) – resulting in loss the privacy and confidentiality of files. Whilst today's use of third-party cryptographic applications can independently encrypt data, it arguably places a significant burden upon the user in terms of manually ciphering/deciphering each file and administering numerous keys in addition to the login password. The field of biometric cryptography applies biometric modalities within cryptography to produce robust bio-crypto keys without having to remember them. There are, nonetheless, still specific flaws associated with the security of the established bio-crypto key and its usability. Users currently should present their biometric modalities intrusively each time a file needs to be encrypted/decrypted – thus leading to cumbersomeness and inconvenience while throughout usage. Transparent biometrics seeks to eliminate the explicit interaction for verification and thereby remove the user inconvenience. However, the application of transparent biometric within bio-cryptography can increase the variability of the biometric sample leading to further challenges on reproducing the bio-crypto key. An innovative bio-cryptographic approach is developed to non-intrusively encrypt/decrypt data by a bio-crypto key established from transparent biometrics on the fly without storing it somewhere using a backpropagation neural network. This approach seeks to handle the shortcomings of the password login, and concurrently removes the usability issues of the third-party cryptographic applications – thus enabling a more secure and usable user-oriented level of encryption to reinforce the security controls within cloud-based storage. The challenge represents the ability of the innovative bio-cryptographic approach to generate a reproducible bio-crypto key by selective transparent biometric modalities including fingerprint, face and keystrokes which are inherently noisier than their traditional counterparts. Accordingly, sets of experiments using functional and practical datasets reflecting a transparent and unconstrained sample collection are conducted to determine the reliability of creating a non-intrusive and repeatable bio-crypto key of a 256-bit length. With numerous samples being acquired in a non-intrusive fashion, the system would be spontaneously able to capture 6 samples within minute window of time. There is a possibility then to trade-off the false rejection against the false acceptance to tackle the high error, as long as the correct key can be generated via at least one successful sample. As such, the experiments demonstrate that a correct key can be generated to the genuine user once a minute and the average FAR was 0.9%, 0.06%, and 0.06% for fingerprint, face, and keystrokes respectively. For further reinforcing the effectiveness of the key generation approach, other sets of experiments are also implemented to determine what impact the multibiometric approach would have upon the performance at the feature phase versus the matching phase. Holistically, the multibiometric key generation approach demonstrates the superiority in generating the bio-crypto key of a 256-bit in comparison with the single biometric approach. In particular, the feature-level fusion outperforms the matching-level fusion at producing the valid correct key with limited illegitimacy attempts in compromising it – 0.02% FAR rate overall. Accordingly, the thesis proposes an innovative bio-cryptosystem architecture by which cloud-independent encryption is provided to protect the users' personal data in a more reliable and usable fashion using non-intrusive multimodal biometrics.

Document Type


Publication Date