No description available.

Cette thèse se situe dans le cadre du projet MISS (Mathématiques de l'Imagerie Stéréoscopique Spatiale) monté par le CNES en collaboration avec plusieurs laboratoires universitaires en 2007. Ce projet se donne l'objectif ambitieux de modéliser un satellite stéréoscopique, prenant deux vues non simultanées mais très rapprochées de la Terre en milieu urbain. Son but principal est d'obtenir une chaîne automatique de reconstruction urbaine à haute résolution à partir de ces deux vues. Ce projet se heurte toutefois à des problèmes de fond que la présente thèse s'attache à résoudre. Le premier problème est le rejet des matches qui pourraient se produire par hasard, notamment dans les zones d'ombres ou d'occlusion, et le rejet également des mouvements au sol (véhicules, piétons, etc.). La thèse propose une méthode de rejet de faux matches basée sur la méthodologie dite a contrario. On montre la consistance mathématique de cette méthode de rejet, et elle est validée sur des paires simulées exactes, sur des vérités terrain fournies par le CNES, et sur des paires classiques de benchmark (Middlebury). Les matches fiables restants représentent entre 40% et 90% des pixels selon les paires testées. Le second problème de fond abordé est la précision. En effet le type de stéréoscopie envisagé exige un très faible angle entre les deux vues, qui sont visuellement presque identiques. Pour obtenir un relief correct, il faut effectuer un recalage extrêmement précis, et calibrer le niveau de bruit qui permet un tel recalage. La thèse met en place une méthode de recalage subpixélien, qui sera démontrée être optimale par des arguments mathématiques et expérimentaux. Ces résultats étendent et améliorent les résultats obtenus au CNES par la méthode MARC. En particulier, il sera montré sur les images de benchmark Middlebury que la précision théorique permise par le bruit correspond bien à celle obtenue sur les matches fiables. Bien que ces résultats soient obtenus dans le cadre d'un dispositif d'acquisition précis (stéréoscopie aérienne ou satellitaire à faible angle), tous les résultats sont utilisables en stéréoscopie quelconque, comme montré dans beaucoup d'expériences.

[pt] Foi desenvolvido pelos grupos MPEG (Moving Picture Expert Group) e VCEG (Video Coding Expert Group) um novo padrão de codificação de vídeo denominado AVC (Advanced Vídeo Coding). Este padrão fornece uma capacidade de compressão maior de vídeo se comparado com os padrões anteriores. A estimação de movimento é a fase da codificação de vídeo que demanda maior tempo de processamento. Estimação de movimento é computacionalmente custosa para H.264/AVC se o método Full Search [1] é usado. A fim de reduzir o tempo de codificação, o software de referência JM 9.8 adota um rápido método de estimação de movimento para pixel inteiro chamado UMHexagonS e um método para pixel fracionário chamado CBFPS. Um método proposto por Xiaoquan Yi, Jun Zhang, Nam Ling e Weijia Shang [2] chamado aqui, por simplicidade, de P021 apresenta, em comparação com o software de referência JM, uma melhor e simplificada forma de estimação de movimento para aumentar a velocidade do processo de codificação e manter a qualidade visual do vídeo. Visando diminuir ainda mais o tempo de processamento e o custo computacional apresentada pelos métodos citados anteriormente, o que é necessário principalmente para aplicações de codificação de vídeo móveis, como celulares e palm tops e aplicações de tempo real, como videoconferência, este trabalho propõe um método de estimação de movimento que consiste em evitar o processamento de blocos que não se beneficiam da pesquisa de metade de pixel. Após a execução dos testes, pode-se constatar que com uma qualidade de vídeo semelhante, o método proposto reduz o tempo de processamento em 76,17%, 75,95% e 11,74% em média quando comparado com os métodos Full Search, JM 98 e P021. Além disso, este método praticamente preserva a taxa de bits, apresentando um pequeno aumento de 8,72% , 8,06% e 8,02% em média quando comparado com os mesmos métodos. Isto representa, em nossa opinião, um pequeno preço a ser pago, considerando os benefícios em termos de tempo de processamento. / [en] The MPEG (Moving Picture Experts Group) and the VCEG (Video Coding Experts Group) have developed a new video compression standard entitled AVC (Advanced Video Coding). This standard offers the capacity of video compression greater than the previous standards.The motion estimation which is a phase of the video compression is extremely computer-intensive and therefore demands most of the processing time. When Full Search [1] method is used for H.264/AVC, this process is extremely expensive. In order to reduce encoding time, the reference software JM (throughout this work, JM98 is the version used) has respectively adopted a fast motion estimation method for the integer pixel called UMHexagonS and a method for the sub- pixel called CBFPS. A method proposed by Xiaoquan Yi, Jun Zhang, Nam Ling e Weijia Shang [2] called here P021 (as referenced by the Joint Video Team) has shown, when compared to the reference software JM, a simple way of motion estimation that increases the speed of coding process while maintaining the video visual quality. In order to decrease the processing time and computational cost of these methods, which are: i) particularly needed in mobile video coding applications such as mobile phones and palmtops; ii) in real time applications such as videoconference, this work proposes a motion estimation method that eliminates the processing of blocks that does not produce benefits for the half-pixel search. The simulation results show that the proposed method reduces the processing time in 76,17%, 75,95% and 11,74% in average when respectively compared with Full Search, JM98 and P021 methods, without relevant impact in video quality. Besides, this method produces a bits rate increase of 8,72% , 8,06% e 8,02% in average when compared with the same methods. This is in our opinion, an inexpensive price to be paid when the time benefits are considered.

<p>The possibility to use real-time computer vision in video sequences gives many opportunities for a system to interact with the environment. Possible ways for interaction are e.g. augmented reality like in the MATRIS project where the purpose is to add new objects into the video sequence, or surveillance where the purpose is to find abnormal events.</p><p>The increase of the speed of computers the last years has simplified this process and it is now possible to use at least some of the more advanced computer vision algorithms that are available. The computational speed of computers is however still a problem, for an efficient real-time system efficient code and methods are necessary. This thesis deals with both problems, one part is about efficient implementations using single instruction multiple data (SIMD) instructions and one part is about robust tracking.</p><p>An efficient real-time system requires efficient implementations of the used computer vision methods. Efficient implementations requires knowledge about the CPU and the possibilities given. In this thesis, one method called SIMD is explained. SIMD is useful when the same operation is applied to multiple data which usually is the case in computer vision, the same operation is executed on each pixel.</p><p>Following the position of a feature or object in a video sequence is called tracking. Tracking can be used for a number of applications. The application in this thesis is to use tracking for pose estimation. One way to do tracking is to cut out a small region around the feature, creating a patch and find the position on this patch in the other frames. To find the position, a measure of the difference between the patch and the image in a given position is used. This thesis thoroughly investigates the sum of absolute difference (SAD) error measure. The investigation involves different ways to improve the robustness and to decrease the average error. One method to estimate the average error, the covariance of the position error is proposed. An estimate of the average error is needed when different measurements are combined.</p><p>Finally, a system for camera pose estimation is presented. The computer vision part of this system is based on the result in this thesis. This presentation contains also a discussion about the result of this system.</p> / Report code: LIU-TEK-LIC-2007:5. The report code in the thesis is incorrect.

Accurate areal measurements of snow-cover extent are important for hydrological and climate modeling. The traditional method of mapping snow cover is binary where a pixel is approximated to either snow-covered or snow-free. Fractional snow cover (FSC) mapping achieves a more precise estimate of areal snow-cover extent by determining the fraction of a pixel that is snow-covered. The two most common FSC methods using Moderate Resolution Imaging Spectroradiometer (MODIS) images are linear spectral unmixing and the empirical Normalized Difference Snow Index (NDSI) method. Machine learning is an alternative to these approaches for estimating FSC, as Artificial Neural Networks (ANNs) have been used for estimating the subpixel abundances of other surfaces. The advantages of ANNs over the other approaches are that they can easily incorporate auxiliary information such as land-cover type and are capable of learning nonlinear relationships between surface reflectance and snow fraction. ANNs are especially applicable to mapping snow-cover extent in forested areas where spatial mixing of surface components is nonlinear. This study developed an ANN approach to snow-fraction mapping. A feed-forward ANN was trained with backpropagation to estimate FSC from MODIS surface reflectance, NDSI, Normalized Difference Vegetation Index (NDVI) and land cover as inputs. The ANN was trained and validated with high spatial-resolution FSC derived from Landsat Enhanced Thematic Mapper Plus (ETM+) binary snow-cover maps. ANN achieved best result in terms of extent of snow-covered area over evergreen forests, where the extent of snow cover was slightly overestimated. Scatter plot graphs of the ANN and reference FSC showed that the neural network tended to underestimate snow fraction in high FSC and overestimate it in low FSC. The developed ANN compared favorably to the standard MODIS FSC product with the two methods estimating the same amount of total snow-covered area in the test scenes.

Pattern recognition is very important in automatic industry. The automation machinery vision system must exchange information very fast with the object we need. So the machinery vision system must have powerful recognition ability. There are more important on image processing, lately. But most researches of image processing are developed on high-resolution image. However, in same situation, for increasing the processing speed, reducing the saving space. Low-resolution image are the only way to achieve the above condition up to now. For the purpose of quickly recognition, we construct the recognition system on low-resolution image. From observing the characteristic of hexagonal grid, we knew the hexagonal grid have greater angular resolution and better image performance than rectangular grid. Therefore, we apply the hexagonal grid on low-resolution image, and using Curve Bend Function (call CBF) on hexagonal gird system; for promoting the accuracy of recognition.We presented an technique of subpixel on low-resolution hexagonal image to obtain better results.

Kassel, University, Diss., 2007.

Zugl.: Passau, University, Diss., 2007.

Although many devices can be used to capture images of high resolution, there is still a need to show these images on displays with low resolution. Existing methods of subpixel-based down-sampling are reviewed in this thesis and their limitations are described. A new approach to optimizing sampling structure conversion for color mosaic displays is developed. Full color images are filtered by a set of optimal filters before down-sampling, resulting in better image quality according to the SCIELAB measure, a spatial extension of the CIELAB metric measuring perceptual color difference. The typical RGB stripe display pattern is tested to get the optimal filters using least-squares filter design. The new approach is also implemented on a widely used two-dimensional display pattern, the Pentile RGBG. Clear images are produced and color fringing artifacts are reduced. Quality of down-sampled images are compared using SCIELAB and by visual inspection.

A pixel in remotely sensed hyperspectral imagery is typically a mixture of multiple electromagnetic radiances from various ground cover materials. Spectral unmixing is a quantitative analysis procedure used to recognize constituent ground cover materials (or endmembers) and obtain their mixing proportions (or abundances) from a mixed pixel. The abundances are typically estimated using the least squares estimation (LSE) method based on the linear mixture model (LMM). This dissertation provides a complete investigation on how the use of appropriate features can improve the LSE of endmember abundances using remotely sensed hyperspectral signals. The dissertation shows how features based on signal classification approaches, such as discrete wavelet transform (DWT), outperform features based on conventional signal representation methods for dimensionality reduction, such as principal component analysis (PCA), for the LSE of endmember abundances. Both experimental and theoretical analyses are reported in the dissertation. A DWT-based linear unmixing system is designed specially for the abundance estimation. The system utilizes the DWT as a pre-processing step for the feature extraction. Based on DWT-based features, the system utilizes the constrained LSE for the abundance estimation. Experimental results show that the use of DWT-based features reduces the abundance estimation deviation by 30-50% on average, as compared to the use of original hyperspectral signals or conventional PCA-based features. Based on the LMM and the LSE method, a series of theoretical analyses are derived to reveal the fundamental reasons why the use of the appropriate features, such as DWT-based features, can improve the LSE of endmember abundances. Under reasonable assumptions, the dissertation derives a generalized mathematical relationship between the abundance estimation error and the endmember separabilty. It is proven that the abundance estimation error can be reduced through increasing the endmember separability. The use of DWT-based features provides a potential to increase the endmember separability, and consequently improves the LSE of endmember abundances. The stability of the LSE of endmember abundances is also analyzed using the concept of the condition number. Analysis results show that the use of DWT-based features not only improves the LSE of endmember abundances, but also improves the LSE stability.