Por favor, use este identificador para citar o enlazar este ítem:
http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/2503| A very deep super resolution approach for blind tampering reconstruction based on reversible image authentication | |
| Gabriel Melendez | |
| RENE ARMANDO CUMPLIDO PARRA | |
| Acceso Abierto | |
| Atribución-NoComercial-SinDerivadas | |
| Reversible image authentication Compressed sensing Tampering detection with reconstruction Very deep super resolution Deep Learning | |
| Reversible Image Authentication (RIA) involves a data hiding process in which watermarks are imperceptibly embedded into an image to protect it against tampering attacks. If a marked image is not tampered, then watermarks can be erased from the marked image and the original cover image is obtained without distortion. This is possible because Reversible Data Hiding (RDH) techniques are used to strategically embed the watermark bits into the image to allow reversibility. Recovery information can be afforded as part of the watermark for reconstruction of tampered image regions. However, most state-of-the-art works discard the reconstruction problem as a consequence of limited embedding capacity offered by existing RDH methods. Although some RIA methods have addressed the tampering reconstruction problem, most of them provide low-quality reconstructed images. To overcome this issue, recovery information must be carefully selected and used for reconstruction to provide images of better quality. In this thesis, we propose a new RIA scheme with reconstruction capabilities. The proposed scheme introduces a reconstruction strategy based on the Very Deep Super Resolution (VDSR) network. In this way, sparse signal representations of down-scaled image blocks are obtained to create the recovery information using compressed sensing theory. Simultaneously, a coefficient elimination phase is applied in order to enhance the sparse signal representation of each image block. Watermarks are reversibly embedded into the high-frequency Integer Wavelet Transform (IWT) coefficients of the blocks using companding technique. The proposed scheme is evaluated over a dataset of 135 benchmark images in terms of watermark imperceptibility, tampering detection accuracy, and reconstruction quality, employing Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) metrics. Experimental results indicate that the proposed scheme outperforms previous state-of-the-art methods; in both, reconstruction quality and runtime. The proposed scheme achieves a PSNR improvement of up to +8 dB in terms of image reconstruction quality while runtime is significantly reduced in sparse signal approximation phase, running ≈ 64% faster than previous methods, which is a result of dealing with downscaled image blocks. | |
| Instituto Nacional de Astrofísica, Óptica y Electrónica | |
| 2023-06 | |
| Tesis de doctorado | |
| Inglés | |
| Estudiantes Investigadores Público en general | |
| Melendez Melendez, G., (2023), A very deep super resolution approach for blind tampering reconstruction based on reversible image authentication, Tesis de Doctorado, Instituto Nacional de Astrofísica, Óptica y Electrónica | |
| OTRAS ESPECIALIDADES TECNOLÓGICAS | |
| Versión aceptada | |
| acceptedVersion - Versión aceptada | |
| Aparece en las colecciones: | Doctorado en Ciencias Computacionales |
Cargar archivos:
| Fichero | Tamaño | Formato | |
|---|---|---|---|
| MELENDEZMG_DCC.pdf | 9.15 MB | Adobe PDF | Visualizar/Abrir |