Extraction of a target speech signal from the convolutive mixture of multiple sources observed in a cocktail party environment is a challenging task, especially when the room acoustic effects and background noise are present in the environment. Such acoustic distortions may further degrade the separation performance of many existing source separation algorithms. Algorithmic solutions to this problem are likely to have strong impact on many applications including automatic speech recognition, hearing aids and cochlear implants, and human-machine interaction. In such applications, to extract the target speech, it is usually required to deal with not only the interfering sound, but also the room reverberations and background noise. To address this problem, several methods are developed in this thesis. For the blind separation of a target speech signal from the convolutive mixture, a multistage algorithm is proposed in which a convolutive independent component analysis (leA) algorithm is applied to the mixture, followed by the estimation of an ideal binary mask (IBM) from the separated sources obtained with the convolutive leA algorithm. In the last step, the errors introduced due to estimation of the IBM are reduced by cepstral smoothing. The separation performance of the above algorithm, however, deteriorates with the increase in surface reflections and background noise within the room environment. Two different methods are therefore developed to reduce such effects. In the first method which is also a multistage method, acoustic effects and background' noise are treated together using an empirical-mode-decomposition (EMD) based algorithm. The noisy reverberant speech is decomposed adaptively into oscillatory components called intrinsic mode functions (IMFs) via an EMD algorithm. Denoising is then applied to selected high frequency IMFs using an EMD- based minimum mean squared error (MMSE) filter, followed by spectral subtraction of the resulting denoised high and low-frequency IMFs. The second method is a two-stage dereverberation algorithm in which the smoothed spectral subtraction mask based on a frequency dependent model is derived and then applied to the reverberant speech to reduce the effects of late reverberations. Wiener filtering is then applied such that the early reverberations are attenuated. Finally, an algorithm is developed for joint blind separation and blind dereverberation. The proposed method consists of a step for the blind estimation of reverberation time (RT). The method is employed in three different ways. Firstly, the available mixture signals are used to estimate blindly the RT, followed by the dereverberation of the mixture signals. Then, the separation algorithm is applied to these resultant mixtures. Secondly, the separation algorithm is applied first to the mixtures, followed by the blind dereverberation of the segregated speech signals. In the third scheme, the separation algorithm is split such that the convolutive leA is first applied to the mixtures, followed by the blind dereverberation of the signals obtained from convolutive leA. Then, the T-F representation of the dereverberated signals is used to estimate the IBM followed by cepstral smoothing.

Cartoon images are often stored as pixel-based, raster data. This format is not the most appropriate for representing cartoon images that are initially drawn as geometric areas of colour. Raster images cannot be scaled without creating pixelisation artefacts and modification of such images must be performed pixel by pixel. A vector-based representation is a more suitable format for storing cartoon images. This thesis investigates a novel process for converting colour raster images into colour vector images. The process has been refined to facilitate the conversion of colour cartoon images, natively stored as raster data, into a scalable vector representation consisting of areas of uniform colours. The process is composed of a number of stages which are reviewed in some detail. These include; colour quantisation, colour decomposition, shape outlining, path tracing, path simplification, rendering and storage. An investigation into the use of colours in different types of images has been performed and used to improve the colour quantisation stage of the process which was noted to be problematic. Using the enhanced colour quantisation scheme together with additional novel optimisations an efficient colour area-based vectorisation system has been produced. The system has been extended to process multiple sequential images to support efficient conversion of cartoon video sequences into scalable vector animation. The developed colour area-based vectorisation system has been applied to a database of raster cartoon images. The generated vector representation is shown to offer a number of benefits including reduced storage requirements, the ability to render at higher resolutions without creating pixelation artefacts and simplified image manipulation for future modifications.

No description available.

No description available.

No description available.

The growing demand for voice-over-packer (VoIP) services and multimedia-rich applications has made increasingly important the efficient, real-time implementation of low-bit rates speech coders on embedded VLSI platforms. Such speech coders are designed to substantially reduce the bandwidth requirements thus enabling dense multichannel gateways in small form factor. This however comes at a high computational cost which mandates the use of very high performance embedded processors. This thesis investigates the potential acceleration of two major ITU-T speech coding algorithms, namely G.729A and G.723.1, through their efficient implementation on a configurable extensible vector embedded CPU architecture. New scalar and vector ISAs were introduced which resulted in up to 80% reduction in the dynamic instruction count of both workloads. These instructions were subsequently encapsulated into a parametric, hybrid SISD (scalar processor)–SIMD (vector) processor. This work presents the research and implementation of the vector datapath of this vector coprocessor which is tightly-coupled to a Sparc-V8 compliant CPU, the optimization and simulation methodologies employed and the use of Electronic System Level (ESL) techniques to rapidly design SIMD datapaths.

To understand a real-world scene from several multiview pictures, it is necessary to find the disparities existing between each pair of images so that they are correctly related to one another. This process, called image registration, requires the extraction of some specific information about the scene. This is achieved by taking features out of the acquired images. Thus, the quality of the registration depends largely on the accuracy of the extracted features. Feature extraction can be formulated as a sampling problem for which perfect re- construction of the desired features is wanted. The recent sampling theory for signals with finite rate of innovation (FRI) and the B-spline theory offer an appropriate new frame- work for the extraction of features in real images. This thesis first focuses on extending the sampling theory for FRI signals to a multichannel case and then presents exact sampling results for two different types of image features used for registration: moments and edges. In the first part, it is shown that the geometric moments of an observed scene can be retrieved exactly from sampled images and used as global features for registration. The second part describes how edges can also be retrieved perfectly from sampled images for registration purposes. The proposed feature extraction schemes therefore allow in theory the exact registration of images. Indeed, various simulations show that the proposed extraction/registration methods overcome traditional ones, especially at low-resolution. These characteristics make such feature extraction techniques very appropriate for applications like image super-resolution for which a very precise registration is needed. The quality of the super-resolved images obtained using the proposed feature extraction meth- ods is improved by comparison with other approaches. Finally, the notion of polyphase components is used to adapt the image acquisition model to the characteristics of real digital cameras in order to run super-resolution experiments on real images.

Hyperspectral imaging is a technique which combines spectral and spatial imaging methods. The technology is used in remote sensing, medicine, agriculture and forensics just to mention a few. Non-remote systems are developed by using sensor designs different from push-broom and whisk-broom methods, commonly found in remote sensing hyperspectral imaging systems. Images are commonly acquired by mounting various electronically tunable filters in front of monochromatic cameras and capturing a range of wavelengths to produce a spectral image cube. Illumination plays a major role during imaging, as both the camera and electronically tunable filter may suffer low transmission at the ends of the visible spectrum, resulting in a low signal to noise ratio. The work described in this thesis attempts to address two key objectives. The first was to identify the main sources of errors in a common design of focal-plane hyperspectral imaging system and devise ways of compensating for these errors. Calibration and characterization of a focal-plane hyperspectral imaging system included system noise characterization, stray-light compensation, flat field correction, image registration, input-output function characterization and calibration verification. The other was to apply imaging techniques to hyperspectral images. This included scene recognition using ratio indexing and spectral gradients. This comes from the underlying idea that due to the large number of bands contained in hyperspectral images, more information is available so better recognition results compared to RGB images. A novel approach for obtaining ratios for ratio indexing is proposed in this thesis. The imaging of archived materials from University of Manchester's John Rylands Library was also done. The aim was to produce high resolution hyperspectral images that will help in identifying accurate matches for colours used in document restoration at the Library.

No description available.

No description available.