Visual Speech Recognition
Automatic System for Lip Reading of Dutch
Alin G. Chitu, Leon J.M. Rothkrantz
Abstract:
Visual speech recognition is the next step towards robust and ubiquitous speech recognition. This paper gives detailed information about the work done at Delft University of Technology towards a robust automatic lip reading system. We proudly introduce the data corpus for the Dutch language we have built for the present research, with details on its development and characteristics. In the current research we consider a large spectrum of visual features, namely motion description features based on optical flow, shape description features based on key point detection and a statistical approach, and appearance description features. This paper introduces the use of Active Appearance Models for the detection of landmarks on the speaker face for lip reading. The inference is made based on the Hidden Markov Models approach using the HTK Toolkit.

Product Recommendations in Mobile Environments
K. Dimitrov, K. Kanev, S. Kimura
Key Words:
Mobile product awareness; QR-code; color barcode;surface encoding; Cluster Pattern Interface (CLUSPI).
Abstract:
In this work we propose a client server model of a product information support system for product recommendations in mobile environments. The system supports clients installed on mobile phones with embedded cameras that are used as optical input devices. Product identification based on 1D barcodes, 2D color codes, and QR-codes is discussed and extensions for product surface mapping with CLUSPI are considered. The server side of the system is envisaged as an ASP.NET application with access control and account management, web and database search frontend, and adaptive smart recommendation facilities.

Parallel Algorithm for Integer Sorting with Multi-core Processors
S. Stoichev, S. Marinova
Key Words:
Parallel algorithm; sorting; counting sort; multi-core processor; shared-memory.
Abstract:
This paper investigates the problem of optimizing sorting of large data sets by applying parallelism. Six variants of a parallel algorithm based on Counting-sort, designed for shared-memory parallel architectures are implemented. Counting-sort is applicable for sorting integers in a given range and is faster than all known sorting algorithms. A speedup from 1.1 times to 4.8 times with respect to the sequential algorithm, achieved by the implemented algorithms, which we called P-CountingSort, PaddingSort, StepPaddingSort, BlockSort, LoopBSort, PNLoopBSort, is demonstrated. Optimization techniques for multi-core processors of the contemporary PCs and investigated the speedup after applying the proposed techniques are applied. The performance and scalability of the implemented algorithms are measured under different conditions and distributions of the input data. Based on the experimental results, the PaddingSort algorithm achieves maximal speed-up for significant key ranges and a multi-processor system. We made recommendations for future research based on the conducted experiments.

Deploying Trusted Computing on TPM Enabled Mobile Computers
E. Vila, N. Sinjagina, P. Borovska
Key Words:
Data security; secure box; computer attacks; trusted computing; trusted platform module; AES; RSA; daisy chain encryption.
Abstract:
Several research studies are devoted to improving the security of personal and confidential information which is stored on a wide variety of data resources. The ongoing new threats such as viruses or malicious codes circulating around should be countered not only from the current methods and tools available but also from new techniques capable to overcome numerous existing problems on contemporary computer systems. This paper presents an experimental study of the trusted computing technology implementation as an eminent alternative to actual approaches with respect to data security. Our work focuses on the deployment and properties of systems based on trusted computing technology principles. Apart from facilitating deployment of forthcoming applications, the ability to enable trustworthy computing at low cost can also address many of existing data security concerns. We base our analyses on practical framework and data collected from investigation of experiments with various data file formats in respect of some performance evaluation parameters. Several references for the future work are also made.

RESCUER Technology for Incident Area Mapping M. Mladenov, M. Mustafa

Key Words:
Incident area mapping; mobile robot and objects localization; stereo vision; map building.
Abstract:
This paper presents a technology for dangerous areas mapping based on a remote controlled mobile robot with stereo vision and sensors for different hazard sources detection. The main goal is to build a map of the incident area, which is conformable to the specific rescue specialists’ requirements. The main steps of the developed technology include robot localization, analysis of data from SH and sensors/detectors for different objects identification, extracting depth information from SH about the identified objects and incident area map building. The procedure for mobile robot, hazard sources and survivors localization is based on information from a stereo head and 3D compass. Information from the stereo head, different sensors and specialized information systems is used for risk sources identification. Image processing techniques are used for incident area map building.