Sketch – An investigation into feature extraction in compressed domain / Kazuki Minemura

Kazuki , Minemura (2017) Sketch – An investigation into feature extraction in compressed domain / Kazuki Minemura. PhD thesis, University of Malaya.

Preview

PDF (Thesis PhD) Download (11Mb) | Preview

Abstract

The recent advancements of multimedia signal processing techniques enable us to conveniently edit/modify multimedia contents, which are usually available in the compressed form for transmission and storage purposes. Among the compression standards, JPEG for still image and H.264/AVC for video are the common formats that we handle in our daily life. As the application of compressed image/video widens, practical security tools for compressed image/video also become increasingly important. In that regard, encryption is the process designed to transform a content into an unintelligible form. Classical encryption approaches (e.g., DES, AES, RSA) handle the bitstream of a compressed image/video directly, but this approach invalidates many features of the compressed image/video, including hierarchical decoding and transcoding. In addition, the computational time of conventional encryption is also high. To overcome the aforementioned problems, the selective-encryption approach, which encrypts a subset of a compressed content, has been pursued by many researchers. Although many were proven to be robust against traditional cryptanalysis, the conventional selective-encryption methods are vulnerable to non-traditional forms of attack called sketch attack. In this thesis, five sketch attacks, i.e., four of them are based on DCT coefficients while another one exploits the statistical features of a coding block, are proposed. Their performances are evaluated and then compared. Furthermore, as a mean to evaluate the sketch image, an objective evaluation metric called outline clearness assessment (OCA) is put forward. Moreover, a selective encryption method for JPEG compressed image is put forward to address the problem of bitstream size overhead while being robust against both the traditional and non-traditional attacks. As an application, sketch attack is utilized as a feature extraction process in H.264/AVC compressed video for text detection purpose. Finally, contributions and limitations of this study are summarized, and future works are discussed.

Actions (For repository staff only : Login required)