This paper addresses an approach to video coding utilizing an iterative nonorthogonal expansion technique called "matching pursuits" (MP) in combination with a new algorithm for selecting an appropriate coding technique at each frame in a sequence. This decision algorithm is called "selective update" and is based on an estimate of the amount and type of motion occurring between coded frames in a video sequence. This paper demonstrates that the matching pursuits approach is most efficient for video coding when motion compensation results in a prediction error which is well localized to the edges of moving objects. In the presence of global motion, such as panning and zooming, or in the presence of objects entering or leaving a scene, matching pursuits becomes less effective than orthogonal transform-based coding techniques like the block-based discrete cosine transform (DCT). The rate-distortion characteristics of matching pursuits and block-wise DCT coding are used to demonstrate how MP coding can be more efficient than block-wise DCT-based coding. When an appropriate combination of these nonorthogonal and orthogonal transforms are used for encoding a complete low bit-rate video sequence, improved overall compression efficiency can be achieved. Results are shown which demonstrate the effectiveness of a hybrid video codec based on this concept.

In this paper, we present a new spatially adaptive approach to the restoration of noisy blurred images, which is particularly effective at producing sharp deconvolution while suppressing the noise in the flat regions of an image. This is accomplished through a multiscale Kalman smoothing filter applied to a prefiltered observed image in the discrete, separable, 2-D wavelet domain. The prefiltering step involves constrained least-squares filtering based on optimal choices for the regularization parameter. This leads to a reduction in the support of the required state vectors of the multiscale restoration filter in the wavelet domain and improvement in the computational efficiency of the multiscale filter. The proposed method has the benefit that the majority of the regularization, or noise suppression, of the restoration is accomplished by the efficient multiscale filtering of wavelet detail coefficients ordered on quadtrees. Not only does this lead to potential parallel implementation schemes, but it permits adaptivity to the local edge information in the image. In particular, this method changes filter parameters depending on scale, local signal-to-noise ratio (SNR), and orientation. Because the wavelet detail coefficients are a manifestation of the multiscale edge information in an image, this algorithm may be viewed as an "edge-adaptive" multiscale restoration approach.

This paper introduces a method for spatially adaptive image restoration based on the detail coefficients of the wavelet transform of a noisy blurred image. A multiscale recursive smoothing filter is applied to the wavelet coefficients ordered onto quadtree structures along different orientations. These coefficients are first prefiltered by a constrained least squares filter in order to remove the spatial correlations due to the blur. An optimal way of choosing the regularization parameters used for this prefiltering operation is introduced here. This is based on an analysis of the detection operations performed by the multiscale filter in order to model edge and non-edge regions of the image differently. Results show that this approach offers a highly adaptive means of preserving edges in a restored image.>

In this paper, we present a novel coding technique that makes use of the nonstationary characteristics of an image sequence displacement field to estimate and encode motion information. We utilize an MPEG style codec in which the anchor frames in a sequence are encoded with a hybrid approach using quadtree, DCT, and wavelet-based coding techniques. A quadtree structured approach is also utilized for the interframe information. The main objective of the overall design is to demonstrate the coding potential of a newly developed motion estimator called the coupled linearized MAP (CLMAP) estimator. This estimator can be used as a means for producing motion vectors that may be regenerated at the decoder with a coarsely quantized error term created in the encoder. The motion estimator generates highly accurate motion estimates from this coarsely quantized data. This permits the elimination of a separately coded displaced frame difference (DFD) and coded motion vectors. For low bit rate applications, this is especially important because the overhead associated with the transmission of motion vectors may become prohibitive. We exploit both the advantages of the nonstationary motion estimator and the effective compression of the anchor frame coder to improve the visual quality of reconstructed QCIF format color image sequences at low bit rates. Comparisons are made with other video coding methods, including the H.261 and MPEG standards and a pel-recursive-based codec.

We present a new matrix vector formulation of a wavelet-based subband decomposition. This formulation allows for the decomposition of both the convolution operator and the signal in the subband domain. With this approach, any single channel linear space-invariant filtering problem can be cast into a multichannel framework. We apply this decomposition to the linear space-invariant image restoration problem and propose a family of multichannel linear minimum mean square error (LMMSE) restoration filters. These filters explicitly incorporate both within and between subband (channel) relations of the decomposed image. Since only within channel stationarity is assumed in the image model, this approach presents a new method for modeling the nonstationarity of images. Experimental results are presented which test the proposed multichannel LMMSE filters. These experiments show that if accurate estimates of the subband statistics are available, the proposed multichannel filters provide major improvements over the traditional single channel filters.

Much work in image coding has centered on wavelet transforms, which can be used to generate multiresolution representation of images. Wavelet transforms have been shown to achieve high compression ratios while maintaining very good image quality, because edge characteristics of images can be well preserved by this method at low bit rates. Quadtree decomposition techniques also utilize the multiresolution nature of images of a segmentation device before coding. The implementation of a novel hybrid technique for coding the coefficients of a wavelet transformation which makes use of a quadtree structure is discussed. This method is a scalar coding technique which takes advantage of the correlation between different scales in the wavelet decomposition, while requiring much less computational complexity than a comparable full search vector quantization technique described.>