Content based Retrieval

Since decade, with rapid growth of available multimedia documents and increasing demand for information indexing and retrieval, much effort has been done on text extraction in images and videos. There are many challenges and difficulties for designer and developer of video extraction process. A lot of work has been done in the field of video extraction from multimedia data. But most of the work is application specific and there is still need of work in designing domain independent systems. This is because there are so many challenges when extracting video with variation in fonts, size, color, alignment, orientation, illumination and background. Problem of video extraction get very difficult because of these deviations. This paper presents a conceptual framework for video extraction derived from the research literature and used as a means for surveying the research literature.

The article is devoted to some ``strange'' phenomena of representation theory and their interrelations. Cross-projective representations of pairs of anticommutative algebras, alloys, their universal envelopping Lie algebras and their representations, quaternary algebras and their alloyability are discussed. Considered examples allow to conclude that new representations have some intriguing features (continuous moduli of finite-dimensional irreducible representations, sophisticated Clebsch-Gordan coefficient calculus, etc.).

We describe a system for querying 3D model databases using the spin image representation as a shape signature for objects depicted as triangular meshes. The spin image representation facilitates the task of aligning the query object with respect to matched models (coarsegrain registration). The main contribution of this work is the introduction of a three-level indexing schema based on artificial neural networks. The indexing schema improves significantly the efficiency in matching query spin images against those stored in the database. Our results are suitable for content-based retrieval in 3D general object databases. A particular application to molecular databases is also presented.

Images, music, video, and 3D scenes play a crucial role in visual information systems and multimedia. There is an extraordinary number of applications of such systems in entertainment, business, art, engineering, and science. Such applications often involve huge collections of media, so that efficient and effective searching in databases of these media is an important operation.

Given a similar query, users would have similar behaviors due to the similar expected result. Under the circumstances, this paper therefore presents a Similar Behavior Learning (SBL) model to learn behaviors of multiple users, thereby enhancing the accuracy of image retrieval (IR). In the SBL, we propose an IR approach consisting of two stages. In the first stage, given a user's query image, the approach retrieves some images based on the similarity of the low-level features and learns user behavior information through the SBL model as a preliminary query result. In the second stage, the user selects the preferred images from the preliminary query result, which is considered as user behavior. The SBL model instantly determines the similarity between the current user's behavior and those recorded in the SBL model database, thereby achieving a better final query result. Experimental results demonstrate the efficacy of the proposed approach.

We derive the integer part of the inverse fine structure constant from four structures, each following from the Dirac equation or the geometry of its rotor space S³. (i) The Larmor self-energy of the zitterbewegung current gives 32π²/3 ≈ 105. (ii) The Clifford grade-2 spin bivector adds a dipole selfenergy equal to 1/3 of the monopole, yielding the Poincaré factor 4/3. (iii) The projective cross-ratio in the rotor's Conformal Geometric Algebra, between the quantum unit and the classical period on S³, gives sinθ = 1/(2π) via the spinor double cover. (iv) The Hopf fibration quantizes 1/α to odd integers. The combined result is 1/α = 32(4π²-1)/9 = 136.81, which rounds to 137.

This interview with Thierry Dutoit, director of the numediart programme at UMONS, traces ten years of research at the intersection of art, science and technology in Mons. It covers the programme's origins, its research themes, its concrete achievements and its embeddedness in the broader Walloon cultural and economic landscape, through to its involvement in Mons 2015, European Capital of Culture.

A robust Neural Network based optimal dementia detection technique is presented in this study. Initially, the input brain images are improved in quality with an Adaptive Median Filter based preprocessing to reduce noise and enhance image quality. Following this, segmentation is performed using the Otsu method, which automatically thresholds the image to separate brain regions of interest from the background. For feature extraction, the Gray-Level Co-occurrence Matrix (GLCM) is utilized to capture essential texture characteristics that reflect subtle changes in brain structures indicative of dementia. The extracted features are then classified using a Dragonfly Algorithm-optimized Convolutional Neural Network (CNN), where the optimization process improves the model's classification accuracy by fine-tuning its parameters. Experimental outcomes demonstrate that the presented approach achieves high detection accuracy. In this paper has the comparison between the neural networks shows that the proposed Dragonfly Optimized CNN achieves 91.2% sensitivity and 93.4% for accuracy.

In the era of Big Data Mining, discovering patterns and relationships
within massive datasets has become essential, as processing such data
can reveal previously unknown knowledge. Currently, there is a growing
interest in tools and techniques that allow for the efficient and effective
analysis of Big Data Mining. The challenges of Big Data Mining involve
not only issues related to scale but also opportunities at every stage, from
data acquisition to data visualisation. Big Data Mining also represents an
important source of information and knowledge, benefiting both systems
and end users. However, managing such a large amount of data and
knowledge requires automation, which has led to the increased use of
machine learning techniques that leverage statistical and mathematical
methods. Machine learning, as one of the most popular techniques, allows
for efficient model training on massive datasets and aids in decisionmaking,
even in the absence of a "right way" in previous knowledge bases.
In our research problem, we focus on developing efficient optimisation
methods to address problems related to Big Data Mining. To this end,
we conduct an in-depth analysis of the best machine learning algorithms
and statistical techniques applied to Big Data Mining to identify one
or more robust optimisation methods. These methods help to overcome
challenges generated by Big Data Mining by utilising statistical techniques
and machine learning algorithms. Furthermore, our work contributes to
various scientific fields by enabling better analysis of large datasets and
uncovering new knowledge.

Content-based retrieval requires the choice of distance functions for determining inter-image distances. Distance functions considered to be desirable for computing inter-image distances are often too expensive, computationally, to be used for on-line retrieval from large image databases. In this paper, we propose a generic and e cient contentbased image retrieval architecture where the original images are mapped on to an abstract feature space such that the desired (or real) inter-image distances correspond to the distances between the vector representations of the images in the feature space. It is shown that it is more e cient to compute distances between these feature vectors and use them as estimates of the real distances. We have conducted experiments using color as the low-level feature. The results show a substantial reduction in the size of the feature space. The experimental results also indicate that high accuracy is achieved for the set of queries tested.

This article explores group manifolds which are efficiently expressed in lower dimensional (Clifford) geometric algebras. The spectral basis of a geometric algebra allows the insightful transition between a geometric algebra of multivectors and its representation as a matrix over the real or complex numbers, or over the quaternions or split quaternions. Whereas almost all of the ground covered is well known, our approach is novel and lays down the fundamental ideas of Lie groups and algebras for group manifolds that are important in mathematics and physics, including the 3-sphere in 4-D Euclidean space, the 3-hyperboloid hypersurface in the neutral 4-D pseudo-Euclidean space, and the Lie group structure GL(2, C) of complex Minkowski spacetime. Other topics covered are the Haar measure of SU(2), and the Riemannian geometry imposed by the group structure at the identity. Mathematics Subject Classification (2010). 15A66, 22E99, 53C15.

The real number system is geometrically extended to include three new anticommuting square roots of plus one, each such root representing the direction of a unit vector along the orthonormal coordinate axes of Euclidean 3-space. The resulting geometric (Clifford) algebra provides a geometric basis for the famous Pauli matrices which, in turn, proves the consistency of the rules of geometric algebra. The flexibility of the concept of geometric numbers opens the door to new understanding of the nature of space-time, and of Pauli and Dirac spinors as points on the Riemann sphere, including Lorentz boosts.

In October 2012 a research project on the statistical analysis of jazz solos started operation at the HfM “Franz Liszt” in Weimar 1 . The aim of this contribution is to introduce this project, its goals and methods, and to place it in the context of jazz research, cognitive music psychology and computational (ethno-)musicology with special consideration of oral traditions in jazz and its implications for creativity research (cf. Pfleiderer & Frieler (2010)).

In this paper, we present a conceptual data model for multimedia database applications based on ExIFO 2 model. The ExIFO 2 data model is chosen as the conceptual model since it handles both complex objects along with their uncertain and imprecise properties. We enhanced this conceptual model in order to meet the multimedia data requirements. In addition to uncertain and imprecise information, we present a way of handling relationships among objects of multimedia database applications. Events that might be extracted from video or audio are also considered in this study. Finally, the conceptual model is mapped to a logical model, which the fuzzy object oriented data (FOOD) model is chosen, for storing and manipulating the multimedia objects. This mapping is done in a way that it preserves most of the information represented at the conceptual level.

DelosDLMS is a prototype of a next-generation Digital Library (DL) management system. It is the result of integrating various specialized DL services provided by partners of the DELOS network of excellence into the OSIRIS platform. OSIRIS is a middleware environment for the reliable and scalable distributed execution of processes. Processes, in turn, are DL applications that are built from the specialized services available in the integrated system. DelosDLMS provides support for content-based retrieval in image, audio, video, and 3D collections and combination of any of these media types with keyword queries. It allows annotating retrieved information, provides a rich set of advanced graphical user interfaces to browse and explore large collections, and supports users in interacting with the system using a speech interface and interactive paper. Finally, it can even be applied in settings where DL content is available in a distributed P2P system.

Although solutions to the challenge of binaural artificial recreation of audio spatialisation exist in the Computer Music domain, a review of the area suggests that a comprehensive, generic, accurate and efficient toolset is required; hence this paper will deal with automated sound creation using sound synthesis. The entire setup was implemented using Csound. Csound implements many synthesis techniques unavailable through other means, such as user-controlled physical modeling .A number of Csound opcodes is presented to satiate this necessity. It provides a platform for creating customized instruments to produce the required the sounds and a player that renders the sound according to the requirements .It provides a powerful base for performing signal processing functions like envelope shaping, filtering and mixing(additive and subtractive synthesis of sounds with disparate parameters).The final output is the product of the filter output that produces the sound of required frequencies and envelope shaping function that controls the amplitude. The utmost use of this research work is the optimum reduction in human interference without compensating with any of the efficient results obtained.