By Topic

Vision, Image and Signal Processing, IEE Proceedings -

Issue 6 • Date Dec 1999

Filter Results

Displaying Results 1 - 9 of 9
  • Adaptive lossless coding scheme of lattice vector quantisation

    Page(s): 317 - 325
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (368 KB)  

    In the process of quantisation, a lattice vector quantiser (LVQ) generates radius and index sequences. In lossless coding, the radius sequence is run-length coded and then Huffman or arithmetic coded, and the index sequence is represented by fixed binary bits. The author has improved the LVQ lossless coding by removing the redundant information between radius sequence and index sequence. An algorithm is developed that redistributes radius and index sequences. The algorithm adaptively shifts down large indices to smaller values and reduces the index bits. Hence, the proposed LVQ lossless coding method reduces the gap between actual coding bit rates and the optimal bit rate boundary. For a Laplacian source the proposed lossless coding scheme achieves more than 10% of bit reduction at bit rates higher than 0.7 bits/sample over the traditional lossless coding method View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Proposal for transparent cascading of 3:4 up and subsequent 4:3 down conversion filtering

    Page(s): 334 - 338
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (216 KB)  

    A proposal is described which is based on a 3:4 FIR Nyquist up conversion filter and a corresponding 4:3 FIR Nyquist down conversion filter that on cascading result mathematically in the identity operator. The proposed filters are applied horizontally to critical video signals. Experimental results show that, despite the limited numerical precision of the filter coefficients and filter calculations, an almost (55 dB) transparent cascade can be achieved. The small amplitude errors are in the order of ±1 unit and found to be invisible when the pictures are displayed on a grade 1 studio monitor of standard TV resolution. Additionally, a high subjective quality for up converted pictures is achieved and maintained over several cascades View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Analysis of spatial transform motion estimation with overlapped compensation and fractional-pixel accuracy

    Page(s): 339 - 344
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (328 KB)  

    The authors investigate how the performance of spatial transform motion estimation can be significantly improved by incorporating overlapped compensation and fractional-pixel accuracy. An overlapped spatial transformation (OST) motion model is developed, which successfully addresses the inability of the conventional block matching technique to compensate for complex motion and inside-block motion field discontinuities. Simulation results show that the motion compensated prediction error of this method is reduced by 1.1 dB, compared with the conventional overlapped block matching motion estimation, for the same generated motion vectors overhead. To improve the performance, the overlap must be used in both the motion estimation and compensation processes. Further improvement can be obtained using half-pixel precision motion vectors. However, this improvement is comparatively less than the gain of conventional block matching from a half-pixel search View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Estimating the distortion and entropy-coded bit rate of lattice vector quantiser for a Laplacian source

    Page(s): 326 - 333
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (300 KB)  

    For practical use of lattice vector quantisers (LVQs), the codebook sizes are pre-decided and the lattice is scaled to control quantisation distortion and bit rate. Formulas are derived, estimating both the distortion and the bit rate as functions of a scaling factor of a LVQ, while a lossless coding method is specified as the rotational coordinate method and input sources are assumed to be Laplacian sources. Unlike conventional approaches based on asymptotical assumptions, the codebook size is limited first, and then the vector space is partitioned according to the characteristic of the source distribution. The proposed estimates are more realistic for practical applications because they handle practical codebook sizes. The lattice points are also counted at each surface on which lattice points exist. The Monte Carlo simulation verifies that the proposed estimates are accurate enough for determining the optimal scaling factor View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Cumulant-based LP method for two-dimensional spectral estimation

    Page(s): 307 - 312
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (224 KB)  

    A cumulant-based linear prediction (CBLP) method for two-dimensional (2-D) spectral estimation is presented. The main idea of the method is to compute the coefficients of two different single-quadrant prediction filters by applying the LP theory to a selected 2-D fourth-order mixed cumulant slice of the noisy signal. These coefficients are employed in formulating two different autoregressive spectral models. Both spectral models are combined to obtain the desired spectral estimate. The effectiveness of the proposed CBLP method is demonstrated through computer simulation View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Rotation invariant classification of rough surfaces

    Page(s): 345 - 352
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (432 KB)  

    Rotation of a rough, textured surface will not produce a simple rotation of the image texture. It follows that where image texture is a function of surface topography, existing rotation invariant texture classification algorithms are not robust to surface rotation. The effect of surface rotation on the observed image is analysed using an existing theory, a novel scheme to stabilise the classification accuracy is proposed and evaluated. The scheme uses photometric stereo to estimate the surface derivatives, which are then used as the input to a classifier. Simulations indicate that, where the level of image noise is moderate or low, the approach is successful in maintaining classification accuracy. Furthermore, in some circumstances, the extra information used by the algorithm allows classification accuracy superior to that based on one image alone, even without rotation View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A posteriori error learning in nonlinear adaptive filters

    Page(s): 293 - 296
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (172 KB)  

    The authors provide relationships between the a priori and a posteriori errors of adaptation algorithms for real-time output-error nonlinear adaptive filters realised as feedforward or recurrent neural networks. The analysis is undertaken for a general nonlinear activation function of a neuron, and for gradient-based learning algorithms, for both a feedforward (FF) and recurrent neural network (RNN). Moreover, the analysis considers both contractive and expansive forms of the nonlinear activation functions within the networks. The relationships so obtained provide the upper and lower error bounds for general gradient based a posteriori learning in neural networks View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Digital algorithm for maximisation of symmetric ambiguity functions and application to signal time-delay estimation

    Page(s): 313 - 316
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (152 KB)  

    Ambiguity functions are usually symmetric around their maximum. In such a case, a consistent estimator of their median value can also detect their maximum. A digital algorithm searches for the zero-crossing point of the Hilbert transform of the estimated function samples. The performance of such a method is analysed by a reduced Taylor expansion, depending on the second-order statistics of the estimated ambiguity samples. The accuracy is explicitly provided in the case of time-delay estimation between random Gaussian signals, corrupted by Gaussian noises. The optimal length of an FIR implementation of the Hilbert filter is also discussed with reference to the generalised cross-correlation method View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Combining linear filtering and radial basis function networks for accurate profile recovery

    Page(s): 297 - 305
    Save to Project icon | Click to expandQuick Abstract | PDF file iconPDF (312 KB)  

    The efficient method presented for the accurate approximation of signal profiles corrupted by noise is based on a principled combination of linear and nonlinear processing. The nonlinear processing is realised using a radial basis network which is designed, trained and validated within the strict time constraints set by instrumentation requirements. The quality of profile approximation and the decision to use either linear or nonlinear processing are set by confidence limits which, in turn, are set by the best estimate of current system noise. The approach is described in terms of a novel capillary electrophoresis instrument with all processing implemented on a dedicated DSP subsystem View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.