Content-based video retrieval technology holds the key to the efficient management and sharing of video content from different sources, in different scales, across different platforms, and over different communication channels. In this work fast video shot retrieval algorithms based on the geometry of video sequence traces in the principal component space are presented. Techniques to address scale (spatial and temporal) issues, in addition to noise and other possible distortions, such as frame dropping, are discussed. Experimental results demonstrate the effectiveness of the proposed approach.

An algorithm is introduced that performs stereophonic acoustic echo cancellation (SAEC) for systems using pairwise panning of a single monophonic source to provide the effect of spatialisation. The technique exploits the inherent high correlation between the loudspeaker signals, unlike other general SAEC techniques, which try to utilise any small uncorrelated features in the signals. The algorithm maintains a single aggregate echo path estimate that is updated using normalised least mean square (NLMS) and the knowledge of any change in the spatialisation. Consequently, it achieves a computational complexity that is of the same order as a single channel NLMS algorithm.

Results are presented of an analysis of land clutter from the fixed site high-resolution BYSON radar at Malvern. Digital terrain elevation data are used to determine the imaged ground regions and the normalised-log estimator U is measured to investigate the stability of the long-tailed clutter distribution. Operational target detectors must maintain a constant false alarm rate over the entire scene, but small areas of non-Gaussian backscatter cause the majority of false alarms. To overcome this, a variable CFAR detector is tested and compared to standard CA and CAGO processors. The performance of the empirically determined variable CFAR is superior, but results suggest that there is no simple statistical model for land clutter of this kind as the false alarms within a particular scene are strongly correlated with the environment.

It is a common practice in transform-domain fragile watermarking schemes for authentication purposes to watermark some selected transform coefficients so as to minimise embedding distortion. The author points out that leaving most of the coefficients unmarked results in a wide-open security gap for attacks to be mounted on them. A fragile watermarking scheme is proposed to implicitly watermark all the coefficients by registering the zero-valued coefficients with a key-generated binary sequence to create the watermark and involving the unwatermarkable coefficients during the embedding process of the embeddable ones. Non-deterministic dependence is established by involving some of the unwatermarkable coefficients selected according to the watermark from a nine-neighbourhood system in order to thwart different attacks, such as cover-up, vector quantisation and transplantation. No hashing is needed in establishing the non-deterministic dependence.

MPEG-4 CELP speech coders configured with multiple bit rates and coding layers provide SNR scalability, and hence may be operated effectively in variable bandwidth environments. A generalised candidate (GC) scheme is proposed to reduce the computational complexity of the stochastic codebook search of CELP coders. The experimental results demonstrate that the proposed GC scheme, incorporated with the verification model (VM) of the MPEG-4 CELP coder, enables a reduction of over 50% of the computational load without suffering any subjective quality degradations. The proposed GC scheme is not only suitable for the base layer and enhancement layers, but also facilitates computational scalability, and is therefore suitable for different working platforms and integrated media source services.

The authors propose a multiscale Bayesian texture segmentation algorithm that is based on a complex wavelet domain hidden Markov tree (HMT) model and a hybrid label tree (HLT) model. The HMT model is used to characterise the statistics of the magnitudes of complex wavelet coefficients. The HLT model is used to fuse the interscale and intrascale context information. In the HLT, the interscale information is fused according to the label transition probability directly resolved by an EM algorithm. The intrascale context information is also fused so as to smooth out the variations in the homogeneous regions. In addition, the statistical model at pixel-level resolution is formulated by a Gaussian mixture model (GMM) in the complex wavelet domain at scale 1, which can improve the accuracy of the pixel-level model. The experimental results on several texture images are used to evaluate the algorithm.

A new method of video scene content characteristic detection is proposed. This method can be applied in conjunction with digital watermarking schemes in order to improve the transparency. In addition, new methods of video watermarking are also proposed. Three related original aspects are reported. First, the authors' previous robust image watermarking methods, which consider the block texture are extended into video. In the detection process, watermark extraction with or without the original frame is provided. Secondly, a method is proposed to improve imperceptibility by reducing the flickering effect. Finally, a method is introduced to detect video scene characteristics in order to suit the particular embedding scheme. By doing so, the embedding method can be chosen adaptively in accordance with the video scene. The subjective tests illustrate an imperceptibility improvement and experiments with various attacks show the watermarking robustness.

Two novel data-hiding schemes for secure data communications are presented. These two schemes perform on the pixel differences between the original (host) image and the decompressed image, which is inspired from Wu's and Chao's methods. The first scheme, referred to as ‘scheme-1’, utilises the differences between two similar images and then embeds the secret data into the different parts. In the embedding process, two auxiliary techniques, zero-replacement and complement-bit, will also be incorporated. In addition, to enhance the hiding efficiency of scheme-1, the authors have followed the bit-plane hiding strategy and created scheme-2. The proposed schemes have two advantages. First, the error distortion can be measured before the hiding process begins. Second, there is no complex computation, which means the hiding process can be executed very efficiently. The experimental results show that the two proposed schemes can achieve better image quality than other data-hiding schemes. At the same time, the new schemes can also satisfy the three basic requirements a data hiding scheme should live up to: invisibility, capacity and data security.

An efficient smoothing scheme for the real-time transmission of MPEG-1 transcoded video over 'best-effort' IP networks is presented. The scheme uses intelligent partitioning and multiplexing of the packetised bit stream. Bit-rate smoothing is achieved by partitioning packets according to their picture type (I, P or B). Subsequently, the partitioned packets are multiplexed in such a way that each packet from an anchor (I and P) picture is followed by two packets from B-pictures. The proposed scheme smooths the bit rate of the encoded video, making it more suitable for adaptive video streaming applications. In such applications, the transmission bit rate is varied so as to adapt to the available network bandwidth. The scheme reorganises the transmission order of the packets, spreading the less important packets from the B-pictures into the more important packets from the anchor pictures. This indicates a reduction in the likelihood of losing more important anchor packets as the bandwidth bottleneck increases, implying an improvement in the quality of transmitted video. A variety of simulations results are presented to demonstrate these points.

The current need for large multimodal databases to evaluate automatic biometric recognition systems has motivated the development of the MCYT bimodal database. The main purpose has been to consider a large scale population, with statistical significance, in a real multimodal procedure, and including several sources of variability that can be found in real environments. The acquisition process, contents and availability of the single-session baseline corpus are fully described. Some experiments showing consistency of data through the different acquisition sites and assessing data quality are also presented.