WATERMARKING CAPACITY IMPROVEMENT BY LOW DENSITY PARITY CHECK CODES
Tez Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Boğaziçi Üniversitesi, Mühendislik Fakültesi, Elektrik-Elektronik Mühendisliği Bölümü, Türkiye
Tez Danışmanı: Mehmet Bülent Sankur
Tezin Onay Tarihi: 2002
Tezin Dili: İngilizce
Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
Özet:
Digital image watermark is an imperceptible, robust, secure message embedded into
the image, which identifies one or more of the owner, distributor or recipient of the image,
origin or status of the data or transaction dates. Watermarking is also used for data hiding,
content labeling, broadcast monitoring and integrity control applications. Digital image
watermarking resembles communication systems. Watermark is the sent message. Image is
the watermark channel or carrier. Image pixels and possible attacks on the marked image
constitute the noise. Only the authorized parties extract the watermark message from the
marked image by using detectors.
Digital image watermarking has three major requirements. Watermark should be
robust against noise and attacks, imperceptible and carry the required number of bits.
These three requirements conflict with each other. To illustrate, increasing the watermark
strength makes the system more robust but unfortunately decreases the perceptual quality.
As a second example, increasing the number of embedded bits increases the capacity but
decreases the robustness.
In this thesis, the goal was to investigate the contribution of the error correcting
codes. More specifically, we studied the error correcting codes as a means to increase the
watermarking capacity of an image or conversely to decrease the embedding strength,
hence to decrease the visual impact of the watermark. We had two watermark channel
models, namely the DFT domain and the 8x8 block DCT domain. First, we compared the
performance of maximum likelihood (ML) detector vis-à-vis correlation and covariance
detector. Second, we compared the performance of LDPC codes vis-à-vis BCH codes and
pure repetition codes. We showed that ML detectors are slightly better than covariance
detectors and LDPC codes outperform BCH codes and repetition codes by a large margin.