Sun, Deqing. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2007. / Includes bibliographical references (leaves 86-91). / Abstracts in English and Chinese. / Abstract --- p.i / 摘要 --- p.iii / Contributions --- p.iv / Acknowledgement --- p.vi / Abbreviations --- p.xviii / Notations --- p.xxi / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Image compression and postprocessing --- p.1 / Chapter 1.2 --- A brief review of postprocessing --- p.3 / Chapter 1.3 --- Objective and methodology of the research --- p.7 / Chapter 1.4 --- Thesis organization --- p.8 / Chapter 1.5 --- A note on publication --- p.9 / Chapter 2 --- Background Study --- p.11 / Chapter 2.1 --- Image models --- p.11 / Chapter 2.1.1 --- Minimum edge difference (MED) criterion for block boundaries --- p.12 / Chapter 2.1.2 --- van Beek's edge model for an edge --- p.15 / Chapter 2.1.3 --- Fields of experts (FoE) for an image --- p.16 / Chapter 2.2 --- Degradation models --- p.20 / Chapter 2.2.1 --- Quantization constraint set (QCS) and uniform noise --- p.21 / Chapter 2.2.2 --- Narrow quantization constraint set (NQCS) --- p.22 / Chapter 2.2.3 --- Gaussian noise --- p.22 / Chapter 2.2.4 --- Edge width enlargement after quantization --- p.25 / Chapter 2.3 --- Use of these models for postprocessing --- p.27 / Chapter 2.3.1 --- MED and edge models --- p.27 / Chapter 2.3.2 --- The FoE prior model --- p.27 / Chapter 3 --- Postprocessing using MED and edge models --- p.28 / Chapter 3.1 --- Blocking artifacts suppression by coefficient restoration --- p.29 / Chapter 3.1.1 --- AC coefficient restoration by MED --- p.29 / Chapter 3.1.2 --- General derivation --- p.31 / Chapter 3.2 --- Detailed algorithm --- p.34 / Chapter 3.2.1 --- Edge identification --- p.36 / Chapter 3.2.2 --- Region classification --- p.36 / Chapter 3.2.3 --- Edge reconstruction --- p.37 / Chapter 3.2.4 --- Image reconstruction --- p.37 / Chapter 3.3 --- Experimental results --- p.38 / Chapter 3.3.1 --- Results of the proposed method --- p.38 / Chapter 3.3.2 --- Comparison with one wavelet-based method --- p.39 / Chapter 3.4 --- On the global minimum of the edge difference . . --- p.41 / Chapter 3.4.1 --- The constrained minimization problem . . --- p.41 / Chapter 3.4.2 --- Experimental examination --- p.42 / Chapter 3.4.3 --- Discussions --- p.43 / Chapter 3.5 --- Conclusions --- p.44 / Chapter 4 --- Postprocessing by the MAP criterion using FoE --- p.49 / Chapter 4.1 --- The proposed method --- p.49 / Chapter 4.1.1 --- The MAP criterion --- p.49 / Chapter 4.1.2 --- The optimization problem --- p.51 / Chapter 4.2 --- Experimental results --- p.52 / Chapter 4.2.1 --- Setting algorithm parameters --- p.53 / Chapter 4.2.2 --- Results --- p.56 / Chapter 4.3 --- Investigation on the quantization noise model . . --- p.58 / Chapter 4.4 --- Conclusions --- p.61 / Chapter 5 --- Conclusion --- p.71 / Chapter 5.1 --- Contributions --- p.71 / Chapter 5.1.1 --- Extension of the DCCR algorithm --- p.71 / Chapter 5.1.2 --- Examination of the MED criterion --- p.72 / Chapter 5.1.3 --- Use of the FoE prior in postprocessing . . --- p.72 / Chapter 5.1.4 --- Investigation on the quantization noise model --- p.73 / Chapter 5.2 --- Future work --- p.73 / Chapter 5.2.1 --- Degradation model --- p.73 / Chapter 5.2.2 --- Efficient implementation of the MAP method --- p.74 / Chapter 5.2.3 --- Postprocessing of compressed video --- p.75 / Chapter A --- Detailed derivation of coefficient restoration --- p.76 / Chapter B --- Implementation details of the FoE prior --- p.81 / Chapter B.1 --- The FoE prior model --- p.81 / Chapter B.2 --- Energy function and its gradient --- p.83 / Chapter B.3 --- Conjugate gradient descent method --- p.84 / Bibliography --- p.86

This thesis presents novel research work on copyright protection of grey scale and colour digital images. New blind frequency domain watermarking algorithms using one dimensional and two dimensional Walsh coding were developed. Handwritten signatures and mobile phone numbers were used in this project as watermarks. In this research eight algorithms were developed based on the DCT using 1D and 2D Walsh coding. These algorithms used the low frequency coefficients of the 8 × 8 DCT blocks for embedding. A shuffle process was used in the watermarking algorithms to increase the robustness against the cropping attacks. All algorithms are blind since they do not require the original image. All algorithms caused minimum distortion to the host images and the watermarking is invisible. The watermark is embedded in the green channel of the RGB colour images. The Walsh coded watermark is inserted several times by using the shuffling process to improve its robustness. The effect of changing the Walsh lengths and the scaling strength of the watermark on the robustness and image quality were studied. All algorithms are examined by using several grey scale and colour images of sizes 512 × 512. The fidelity of the images was assessed by using the peak signal to noise ratio (PSNR), the structural similarity index measure (SSIM), normalized correlation (NC) and StirMark benchmark tools. The new algorithms were tested on several grey scale and colour images of different sizes. Evaluation techniques using several tools with different scaling factors have been considered in the thesis to assess the algorithms. Comparisons carried out against other methods of embedding without coding have shown the superiority of the algorithms. The results have shown that use of 1D and 2D Walsh coding with DCT Blocks offers significant improvement in the robustness against JPEG compression and some other image processing operations compared to the method of embedding without coding. The originality of the schemes enables them to achieve significant robustness compared to conventional non-coded watermarking methods. The new algorithms offer an optimal trade-off between perceptual distortion caused by embedding and robustness against certain attacks. The new techniques could offer significant advantages to the digital watermark field and provide additional benefits to the copyright protection industry.

No / At present, almost all digital images are stored and transferred in their compressed format in which discrete cosine transform (DCT)-based compression remains one of the most important data compression techniques due to the efforts from JPEG. In order to save the computation and memory cost, it is desirable to have image processing operations such as feature extraction, image indexing, and pattern classifications implemented directly in the DCT domain. To this end, we present in this paper a generalized analysis of spatial relationships between the DCTs of any block and its sub-blocks. The results reveal that DCT coefficients of any block can be directly obtained from the DCT coefficients of its sub-blocks and that the interblock relationship remains linear. It is useful in extracting global features in compressed domain for general image processing tasks such as those widely used in pyramid algorithms and image indexing. In addition, due to the fact that the corresponding coefficient matrix of the linear combination is sparse, the computational complexity of the proposed algorithms is significantly lower than that of the existing methods.

Thesis (DSc (Mathematical Sciences. Applied Mathematics))--University of Stellenbosch, 2009. / The replacement of the standard discrete cosine transform (DCT) of JPEG with the windowed modifed DCT (MDCT) is investigated to determine whether improvements in numerical quality can be achieved. To this end, we employ an existing algorithm for optimal quantisation, for which we also propose improvements. This involves the modelling and prediction of quantisation tables to initialise the algorithm, a strategy that is also thoroughly tested. Furthermore, the effects of various window functions on the coding results are investigated, and we find that improved quality can indeed be achieved by modifying JPEG in this fashion.

Of the many existing eukaryotic gene finding software programs, none are able to guarantee accurate identification of genomic protein coding regions and other biological signals central to pathway from DNA to the protein. Eukaryotic gene finding is difficult mainly due to noncontiguous and non-continuous nature of genes. Existing approaches are heavily dependent on the compositional statistics of the sequences they learn from and are not equally suitable for all types of sequences. This thesis firstly develops efficient digital signal processing-based methods for the identification of genomic protein coding regions, and then combines the optimum signal processing-based non-data-driven technique with an existing data-driven statistical method in a novel system demonstrating improved identification of acceptor splice sites. Most existing well-known DNA symbolic-to-numeric representations map the DNA information into three or four numerical sequences, potentially increasing the computational requirement of the sequence analyzer. Proposed mapping schemes, to be used for signal processing-based gene and exon prediction, incorporate DNA structural properties in the representation, in addition to reducing complexity in subsequent processing. A detailed comparison of all DNA representations, in terms of computational complexity and relative accuracy for the gene and exon prediction problem, reveals the newly proposed ?paired numeric? to be the best DNA representation. Existing signal processing-based techniques rely mostly on the period-3 behaviour of exons to obtain one dimensional gene and exon prediction features, and are not well equipped to capture the complementary properties of exonic / intronic regions and deal with the background noise in detection of exons at their nucleotide levels. These issues have been addressed in this thesis, by proposing six one-dimensional and three multi-dimensional signal processing-based gene and exon prediction features. All one-dimensional and multi-dimensional features have been evaluated using standard datasets such as Burset/Guigo1996, HMR195, and the GENSCAN test set. This is the first time that different gene and exon prediction features have been compared using substantial databases and using nucleotide-level metrics. Furthermore, the first investigation of the suitability of different window sizes for period-3 exon detection is performed. Finally, the optimum signal processing-based gene and exon prediction scheme from our evaluations is combined with a data-driven statistical technique for the recognition of acceptor splice sites. The proposed DSP-statistical hybrid is shown to achieve 43% reduction in false positives over WWAM, as used in GENSCAN.

Of the many existing eukaryotic gene finding software programs, none are able to guarantee accurate identification of genomic protein coding regions and other biological signals central to pathway from DNA to the protein. Eukaryotic gene finding is difficult mainly due to noncontiguous and non-continuous nature of genes. Existing approaches are heavily dependent on the compositional statistics of the sequences they learn from and are not equally suitable for all types of sequences. This thesis firstly develops efficient digital signal processing-based methods for the identification of genomic protein coding regions, and then combines the optimum signal processing-based non-data-driven technique with an existing data-driven statistical method in a novel system demonstrating improved identification of acceptor splice sites. Most existing well-known DNA symbolic-to-numeric representations map the DNA information into three or four numerical sequences, potentially increasing the computational requirement of the sequence analyzer. Proposed mapping schemes, to be used for signal processing-based gene and exon prediction, incorporate DNA structural properties in the representation, in addition to reducing complexity in subsequent processing. A detailed comparison of all DNA representations, in terms of computational complexity and relative accuracy for the gene and exon prediction problem, reveals the newly proposed ?paired numeric? to be the best DNA representation. Existing signal processing-based techniques rely mostly on the period-3 behaviour of exons to obtain one dimensional gene and exon prediction features, and are not well equipped to capture the complementary properties of exonic / intronic regions and deal with the background noise in detection of exons at their nucleotide levels. These issues have been addressed in this thesis, by proposing six one-dimensional and three multi-dimensional signal processing-based gene and exon prediction features. All one-dimensional and multi-dimensional features have been evaluated using standard datasets such as Burset/Guigo1996, HMR195, and the GENSCAN test set. This is the first time that different gene and exon prediction features have been compared using substantial databases and using nucleotide-level metrics. Furthermore, the first investigation of the suitability of different window sizes for period-3 exon detection is performed. Finally, the optimum signal processing-based gene and exon prediction scheme from our evaluations is combined with a data-driven statistical technique for the recognition of acceptor splice sites. The proposed DSP-statistical hybrid is shown to achieve 43% reduction in false positives over WWAM, as used in GENSCAN.

No description available.

L'athérosclérose est une maladie vasculaire complexe qui affecte la paroi des artères (par l'épaississement) et les lumières (par la formation de plaques). La rupture d'une plaque de l'artère carotide peut également provoquer un accident vasculaire cérébral ischémique et des complications. Bien que plusieurs modalités d'imagerie médicale soient actuellement utilisées pour évaluer la stabilité d'une plaque, elles présentent des limitations telles que l'irradiation, les propriétés invasives, une faible disponibilité clinique et un coût élevé. L'échographie est une méthode d'imagerie sûre qui permet une analyse en temps réel pour l'évaluation des tissus biologiques. Il est intéressant et prometteur d’appliquer une échographie vasculaire pour le dépistage et le diagnostic précoces des plaques d’artère carotide. Cependant, les ultrasons vasculaires actuels identifient uniquement la morphologie d'une plaque en termes de luminosité d'écho ou l’impact de cette plaque sur les caractéristiques de l’écoulement sanguin, ce qui peut ne pas être suffisant pour diagnostiquer l’importance de la plaque. La technique d’élastographie vasculaire non-intrusive (« noninvasive vascular elastography (NIVE) ») a montré le potentiel de détermination de la stabilité d'une plaque. NIVE peut déterminer le champ de déformation de la paroi vasculaire en mouvement d’une artère carotide provoqué par la pulsation cardiaque naturelle. En raison des différences de module de Young entre les différents tissus des vaisseaux, différents composants d’une plaque devraient présenter différentes déformations, caractérisant ainsi la stabilité de la plaque. Actuellement, les performances et l’efficacité numérique sous-optimales limitent l’acceptation clinique de NIVE en tant que méthode rapide et efficace pour le diagnostic précoce des plaques vulnérables. Par conséquent, il est nécessaire de développer NIVE en tant qu’outil d’imagerie non invasif, rapide et économique afin de mieux caractériser la vulnérabilité liée à la plaque. La procédure à suivre pour effectuer l’analyse NIVE consiste en des étapes de formation et de post-traitement d’images. Cette thèse vise à améliorer systématiquement la précision de ces deux aspects de NIVE afin de faciliter la prédiction de la vulnérabilité de la plaque carotidienne. Le premier effort de cette thèse a été dédié à la formation d'images (Chapitre 5). L'imagerie par oscillations transversales a été introduite dans NIVE. Les performances de l’imagerie par oscillations transversales couplées à deux estimateurs de contrainte fondés sur un modèle de déformation fine, soit l’ « affine phase-based estimator (APBE) » et le « Lagrangian speckle model estimator (LSME) », ont été évaluées. Pour toutes les études de simulation et in vitro de ce travail, le LSME sans imagerie par oscillation transversale a surperformé par rapport à l'APBE avec imagerie par oscillations transversales. Néanmoins, des estimations de contrainte principales comparables ou meilleures pourraient être obtenues avec le LSME en utilisant une imagerie par oscillations transversales dans le cas de structures tissulaires complexes et hétérogènes. Lors de l'acquisition de signaux ultrasonores pour la formation d'images, des mouvements hors du plan perpendiculaire au plan de balayage bidimensionnel (2-D) existent. Le deuxième objectif de cette thèse était d'évaluer l'influence des mouvements hors plan sur les performances du NIVE 2-D (Chapitre 6). À cette fin, nous avons conçu un dispositif expérimental in vitro permettant de simuler des mouvements hors plan de 1 mm, 2 mm et 3 mm. Les résultats in vitro ont montré plus d'artefacts d'estimation de contrainte pour le LSME avec des amplitudes croissantes de mouvements hors du plan principal de l’image. Malgré tout, nous avons néanmoins obtenu des estimations de déformations robustes avec un mouvement hors plan de 2.0 mm (coefficients de corrélation supérieurs à 0.85). Pour un jeu de données cliniques de 18 participants présentant une sténose de l'artère carotide, nous avons proposé d'utiliser deux jeux de données d'analyses sur la même plaque carotidienne, soit des images transversales et longitudinales, afin de déduire les mouvements hors plan (qui se sont avérés de 0.25 mm à 1.04 mm). Les résultats cliniques ont montré que les estimations de déformations restaient reproductibles pour toutes les amplitudes de mouvement, puisque les coefficients de corrélation inter-images étaient supérieurs à 0.70 et que les corrélations croisées normalisées entre les images radiofréquences étaient supérieures à 0.93, ce qui a permis de démontrer une plus grande confiance lors de l'analyse de jeu de données cliniques de plaques carotides à l'aide du LSME. Enfin, en ce qui concerne le post-traitement des images, les algorithmes NIVE doivent estimer les déformations des parois des vaisseaux à partir d’images reconstituées dans le but d’identifier les tissus mous et durs. Ainsi, le dernier objectif de cette thèse était de développer un algorithme d'estimation de contrainte avec une résolution de la taille d’un pixel ainsi qu'une efficacité de calcul élevée pour l'amélioration de la précision de NIVE (Chapitre 7). Nous avons proposé un estimateur de déformation de modèle fragmenté (SMSE) avec lequel le champ de déformation dense est paramétré avec des descriptions de transformées en cosinus discret, générant ainsi des composantes de déformations affines (déformations axiales et latérales et en cisaillement) sans opération mathématique de dérivées. En comparant avec le LSME, le SMSE a réduit les erreurs d'estimation lors des tests de simulations, ainsi que pour les mesures in vitro et in vivo. De plus, la faible mise en oeuvre de la méthode SMSE réduit de 4 à 25 fois le temps de traitement par rapport à la méthode LSME pour les simulations, les études in vitro et in vivo, ce qui pourrait permettre une implémentation possible de NIVE en temps réel. / Atherosclerosis is a complex vascular disease that affects artery walls (by thickening) and lumens (by plaque formation). The rupture of a carotid artery plaque may also induce ischemic stroke and complications. Despite the use of several medical imaging modalities to evaluate the stability of a plaque, they present limitations such as irradiation, invasive property, low clinical availability and high cost. Ultrasound is a safe imaging method with a real time capability for assessment of biological tissues. It is clinically used for early screening and diagnosis of carotid artery plaques. However, current vascular ultrasound technologies only identify the morphology of a plaque in terms of echo brightness or the impact of the vessel narrowing on flow properties, which may not be sufficient for optimum diagnosis. Noninvasive vascular elastography (NIVE) has been shown of interest for determining the stability of a plaque. Specifically, NIVE can determine the strain field of the moving vessel wall of a carotid artery caused by the natural cardiac pulsation. Due to Young’s modulus differences among different vessel tissues, different components of a plaque can be detected as they present different strains thereby potentially helping in characterizing the plaque stability. Currently, sub-optimum performance and computational efficiency limit the clinical acceptance of NIVE as a fast and efficient method for the early diagnosis of vulnerable plaques. Therefore, there is a need to further develop NIVE as a non-invasive, fast and low computational cost imaging tool to better characterize the plaque vulnerability. The procedure to perform NIVE analysis consists in image formation and image post-processing steps. This thesis aimed to systematically improve the accuracy of these two aspects of NIVE to facilitate predicting carotid plaque vulnerability. The first effort of this thesis has been targeted on improving the image formation (Chapter 5). Transverse oscillation beamforming was introduced into NIVE. The performance of transverse oscillation imaging coupled with two model-based strain estimators, the affine phase-based estimator (APBE) and the Lagrangian speckle model estimator (LSME), were evaluated. For all simulations and in vitro studies, the LSME without transverse oscillation imaging outperformed the APBE with transverse oscillation imaging. Nonetheless, comparable or better principal strain estimates could be obtained with the LSME using transverse oscillation imaging in the case of complex and heterogeneous tissue structures. During the acquisition of ultrasound signals for image formation, out-of-plane motions which are perpendicular to the two-dimensional (2-D) scan plane are existing. The second objective of this thesis was to evaluate the influence of out-of-plane motions on the performance of 2-D NIVE (Chapter 6). For this purpose, we designed an in vitro experimental setup to simulate out-of-plane motions of 1 mm, 2 mm and 3 mm. The in vitro results showed more strain estimation artifacts for the LSME with increasing magnitudes of out-of-plane motions. Even so, robust strain estimations were nevertheless obtained with 2.0 mm out-of-plane motion (correlation coefficients higher than 0.85). For a clinical dataset of 18 participants with carotid artery stenosis, we proposed to use two datasets of scans on the same carotid plaque, one cross-sectional and the other in a longitudinal view, to deduce the out-of-plane motions (estimated to be ranging from 0.25 mm to 1.04 mm). Clinical results showed that strain estimations remained reproducible for all motion magnitudes since inter-frame correlation coefficients were higher than 0.70, and normalized cross-correlations between radiofrequency images were above 0.93, which indicated that confident motion estimations can be obtained when analyzing clinical dataset of carotid plaques using the LSME. Finally, regarding the image post-processing component of NIVE algorithms to estimate strains of vessel walls from reconstructed images with the objective of identifying soft and hard tissues, we developed a strain estimation method with a pixel-wise resolution as well as a high computation efficiency for improving NIVE (Chapter 7). We proposed a sparse model strain estimator (SMSE) for which the dense strain field is parameterized with Discrete Cosine Transform descriptions, thereby deriving affine strain components (axial and lateral strains and shears) without mathematical derivative operations. Compared with the LSME, the SMSE reduced estimation errors in simulations, in vitro and in vivo tests. Moreover, the sparse implementation of the SMSE reduced the processing time by a factor of 4 to 25 compared with the LSME based on simulations, in vitro and in vivo results, which is suggesting a possible implementation of NIVE in real time.

No / As existing video processing technology is primarily developed in the pixel domain yet digital video is stored in compressed format, any application of those techniques to compressed videos would require decompression. For discrete cosine transform (DCT)-based MPEG compressed videos, the computing cost of standard row-by-row and column-by-column inverse DCT (IDCT) transforms for a block of 8 8 elements requires 4096 multiplications and 4032 additions, although practical implementation only requires 1024 multiplications and 896 additions. In this paper, we propose a new algorithm to extract videos directly from MPEG compressed domain (DCT domain) without full IDCT, which is described in three extraction schemes: 1) video extraction in 2 2 blocks with four coefficients; 2) video extraction in 4 4 blocks with four DCT coefficients; and 3) video extraction in 4 4 blocks with nine DCT coefficients. The computing cost incurred only requires 8 additions and no multiplication for the first scheme, 2 multiplication and 28 additions for the second scheme, and 47 additions (no multiplication) for the third scheme. Extensive experiments were carried out, and the results reveal that: 1) the extracted video maintains competitive quality in terms of visual perception and inspection and 2) the extracted videos preserve the content well in comparison with those fully decompressed ones in terms of histogram measurement. As a result, the proposed algorithm will provide useful tools in bridging the gap between pixel domain and compressed domain to facilitate content analysis with low latency and high efficiency such as those applications in surveillance videos, interactive multimedia, and image processing.