Image compression using stochastic artificial neural networks (SANNs) is studied. The ideal is to store an image in a stable distribution of a stochastic neural network. Given an input image f epsilon F, one can find ...
详细信息
Image compression using stochastic artificial neural networks (SANNs) is studied. The ideal is to store an image in a stable distribution of a stochastic neural network. Given an input image f epsilon F, one can find a SANN t epsilon T such that the equilibrium distribution of this SANN is the given image f. Therefore, the input image, f, is encoded into a specification of a SANN, t. This mapping from F (image space) to T (parameter space of SANN) defines the SANN transformation. It is shown that the compression ratio R of the SANN transformation is R=O(n/(K (log n)/sup 2/)) where n is the number of pixels. To complete a SANN transformation, SANN equations must be solved. Two SANN equations are presented. The solution of SANN is briefly discussed.< >
The authors consider the problem of optimal bit allocation in various forms of predictive coding, where the predictor itself has errors resulting from previous quantization. The solution to this problem has potential ...
详细信息
The authors consider the problem of optimal bit allocation in various forms of predictive coding, where the predictor itself has errors resulting from previous quantization. The solution to this problem has potential application to many forms of image and video coding where predictive coding is used. In predictive coding, the input to the quantizer can be decomposed into the innovation, i.e., the part of the quantizer input signal due to the quantization of the predictor. The natural question that arises is whether it is better to allocate more bits to the predictor, since quantization errors persist longer, or to allocate more bits to coding the total residual. This problem is analyzed for predictive video coding through the use of a simple parametric distortion-rate model for the propagation of quantization errors. This model provides a framework in which the optimal bit allocation problem can be solved in the presence of quantizer feedback. An exact MMSE (minimum mean-square error) solution is obtained that involves solving one nonlinear monotonic equation for one Lagrange multiplier, after which the bit allocation has a closed-form analytic solution. Since the MMSE solution does not produce equal distortion in all frames, the optimal MINMAX (minimize the maximum) bit allocation that minimizes the frame distortion subject to equal distortions per frame is also introduced.< >
A major source of audible distortion in current low-bit-rate speech coding algorithms is an inaccurate degree of periodicity of the voiced speech signal. If the correlations between neighboring pitch cycles are accura...
详细信息
A major source of audible distortion in current low-bit-rate speech coding algorithms is an inaccurate degree of periodicity of the voiced speech signal. If the correlations between neighboring pitch cycles are accurately reproduced, these audible distortions can be reduced significantly. To this purpose, a novel method of coding voiced speech is introduced, which transmits an encoded prototype waveform at 20-30 ms intervals. The prototype waveform describes a pitch cycle representative for the interval, and is quantized using analysis-by-synthesis methods. The speech signal is reconstructed by concatenation of interpolated prototype waveforms. The short-term and the long-term correlations between pitch cycles can be controlled explicitly. Unquantized reconstructed speech is virtually indistinguishable from the original signal. The method results in excellent speech quality at rates between 3.0 and 4.0 kb/s.< >
At this moment different speech coders at low bitrates are being defined and normalized. These speech coders are all based on an analysis by synthesis procedure. Different methods have been proposed in recent litterat...
详细信息
At this moment different speech coders at low bitrates are being defined and normalized. These speech coders are all based on an analysis by synthesis procedure. Different methods have been proposed in recent litterature to construct the excitation signal. The goal of this article is to present in a unified formalism these different algorithms and to estimate their costs in terms of complexity. In a first step the basic principle of these speech coders is described. It is shown that a generalized description of the excitation covers different classical coding techniques and results in a most general iterative standard algorithm. The complexity of this algorithm is evaluated in terms of the number of multiplications/accumulations per second and different possible simplifications are analyzed. This article concludes with a presentation of the coders that are actually submitted for normalization.
This paper presents a new method for efficient image coding, consisting of a cascade of the following processing stages: i) predictive ordering technique (POT);ii) feedback transform coding (FTC);iii) vertical subtrac...
详细信息
This paper presents a new method for efficient image coding, consisting of a cascade of the following processing stages: i) predictive ordering technique (POT);ii) feedback transform coding (FTC);iii) vertical subtraction of quantized coefficients (VSQC);iv) predictive coding refinements in the signal space consisting of either overshoot suppression (OS) as a first variant or hybrid block truncation coding (HBTC) as a second one. The POT algorithm uses the vertical correlation between adjacent pels to change the relative order of elements along a scan line, by putting them in decreasing order of amplitudes (taking as reference the previously received scan line);this ordering concentrates the signal energy into "low generalized frequency" regions. The FTC method is an iterative procedure for increasing with a given step the number of nonzero elements that belong to the orthogonally transformed picture vector, until the mean square error criterion is satisfied (the error representing the difference between the original image vector and the last reconstructed iteration). The VSQC computes the discrete differences between the quantized transform coefficients of the same order belonging to adjacent scan lines. The OS algorithm detects and eliminates the spatial reconstruction errors, whose absolute values exceed a given threshold. The HBTC uses a one-bit nonparametric quantization of the error vector representing the difference between the original picture vector and the last reconstructed vector (satisfying the FTC criterion), so that the first two sample moments are preserved. The reconstructed pictures are presented with their coding fidelity performances (mean square quantization error, mean absolute error and signal-to-noise ratio), using as test pictures a portrait and a LANDSAT image. Good quality images at low bit rate (0.55-1.1) bits/pixel have been obtained.
Studies adaptive predictive coding of digitized images using novel two-dimensional multiplicative autoregressive (MAR) models. A general stability theorem for 2-D MAR image models with causal nonsymmetric half-plane (...
详细信息
Studies adaptive predictive coding of digitized images using novel two-dimensional multiplicative autoregressive (MAR) models. A general stability theorem for 2-D MAR image models with causal nonsymmetric half-plane (NSHP) supports is presented, and two different models are introduced. A major advantage of using 2-D MAR models, as opposed to general 2-D models, is the ability to guarantee predictor stability for NSHP support regions. The performance of 2-D MAR predictive coders is tested on five different images. The results of these studies demonstrate that, for coding of digitized images, 2-D MAR model-based predictive coders offer an effective alternative to currently available techniques.< >
Linear predictive coding (LPC) computes predictive parameters using autocorrelation coefficients. Due to the demand of a large amount of computation time, the method has significantly reduced its value in real-time ap...
详细信息
Linear predictive coding (LPC) computes predictive parameters using autocorrelation coefficients. Due to the demand of a large amount of computation time, the method has significantly reduced its value in real-time applications. This paper describes a new method for speedy computation of the autocorrelation coefficients using Fermat number transform (FNT). It is found that there exists a fast computational algorithm for FNT which has a computational structure similar to the fast Fourier transform's (FFT). Since the fast Fermat number transformation (FFNT) and FFT have similar computational structures, we may adopt readily available FFT VLSI hardware structure for real-time implementation of the FFNT. A verification of the FFNT on an MC 68000 single-board computer has been performed and the result was quite satisfactory.
The arithmetic coding of the bit planes of an image represented either in the conventional weighted binary form or the Gray code is investigated. Two-dimensional contexts based on the pixels from the current bit plane...
详细信息
The arithmetic coding of the bit planes of an image represented either in the conventional weighted binary form or the Gray code is investigated. Two-dimensional contexts based on the pixels from the current bit plane and three-dimensional contexts based on pixels from both the current and previous bit planes are studied. Results indicate that for binary bit planes, a three-dimensional template results in a significant bit rate reduction compared to a two-dimensional template of the same size. The savings are much less for Gray encoded bit planes;since the final bit rates are similar to the binary case, the need for Gray encoding is eliminated. The effect of the order of encoding the bit planes is also studied. Comparisons to conventional lossless schemes, such as predictive coding followed by Huffman or arithmetic coding, are presented.
in this paper, median based predictors for predictive coding of image sequences are presented. Principles of constructing predictors are discussed after which two approaches of constructing multilevel structures are p...
详细信息
in this paper, median based predictors for predictive coding of image sequences are presented. Principles of constructing predictors are discussed after which two approaches of constructing multilevel structures are presented. Simulation results on coding gain, prediction signal entropy and transmission error propagation are presented and compared to an adaptive linear predictor.
暂无评论