Title:
Minimum Distortion Data Hiding for Compressed Images
Minimum Distortion Data Hiding for Compressed Images
Author(s)
Candan, Cagatay
Advisor(s)
Jayant, Nikil S.
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
We present a novel data hiding method for compressed images. The method is designed to minimize the quality loss associated with data embedding into a JPEG image. The described technique uses the objective criterion such as the mean square error and the human visual system based criterion such as the Just Noticable Distortion metric for distortion minimization. The hiding method is designed under the restrictions of the JPEG compression standard to develop new image applications without any modifications or additions to the existing standard. An application example is presented in the thesis. The performance of the technique is examined at different image sizes and resolutions. The cost of hiding in terms of file length extension is examined. Some subjective experiments to determine the zero-perceived distortion hiding capacity are made. An application illustrating the usage of the technique is given. The described application embeds check-bits into JPEG images to facilitate the verification of the sender identity and the authenticity of the transmitted image. In this thesis, we give a list of requirements on the data hiding methods to implement standard compliant applications; design a provably good hiding method operating under these requirements; determine the critical performance points of the method and propose an application based on the method.
We have performed some additional research to determine how our system works with high resolution images and existing other well-known algorithms for information hiding. The experiments on high-resolution images have shown that there exists a large embedding capacity for the high resolution images in spite of a loss of embedding density. The performance comparison experiments have shown that the spread spectrum technique offers a competitive but less efficient distortion performance.
Sponsor
Date Issued
2004-03-22
Extent
1697313 bytes
Resource Type
Text
Resource Subtype
Dissertation