Coded data generation or conversion – Digital code to digital code converters – Adaptive coding
Reexamination Certificate
1999-06-07
2001-03-27
Tokar, Michael (Department: 2819)
Coded data generation or conversion
Digital code to digital code converters
Adaptive coding
C341S065000, C341S067000
Reexamination Certificate
active
06208272
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention concerns a method and device for decoding a coded digital signal, and more particularly a method and device for decoding a digital signal which was quantized during coding.
The coding of a digital signal is normally used for reducing the transmission time for this signal or to reduce the space required for storing such a digital signal. The decoding makes it possible to reconstitute, in order to use it, a digital signal which was previously coded.
The present invention concerns the decoding of digital signals coded by methods of compressing digital signals with loss, and is particularly related to compression methods in which quantization methods are used.
Quantization techniques make it possible, in a known fashion, to transform a continuous signal into a discrete signal, or to transform one discrete signal into another discrete signal which is less finely represented.
Such a quantization of the digital signal makes it possible to represent the signal by means of a finite number of bits; the larger the quantization step (in scalar quantization) or the smaller the dictionary (in vector quantization), the smaller the number of bits.
It is clear that, the larger the size of the quantization cells in which all the vectors of the signal are coded by the same quantization symbol, the greater the compression of the signal whilst, the smaller the size of the quantization cells, the smaller the distortion appearing on decoding.
Known techniques of dequantizing a coded signal are described in the book entitled “Vector Quantization and Signal Compression” by A. Gersho et R. M. Gray, published by Kluwer Academic Publishers, Boston, Mass., 1992, pages 173-185. These techniques use the centre of the quantization cell as the dequantization vector, or restoration level.
SUMMARY OF THE INVENTION
The object of the present invention is to improve the ratio of compression to distortion when decoding a quantized digital signal, and notably to reduce the distortion caused during the dequantization of the signal.
The method of decoding a digital signal to which the invention relates applies to a coded signal quantized by means of quantization symbols associated respectively with quantization cells.
According to the invention, this decoding method is characterised in that it includes the following steps:
reading the quantization symbols;
constructing an experimental distribution of the quantization symbols on the quantization cells by calculating the number of occurrences of the said quantization symbols in the coded signal;
approximating the said experimental distribution by means of a continuous distribution model, the continuous distribution model passing through values of the experimental distribution;
calculating, for at least one quantization cell, a dequantization vector equal to the centre of mass of the points of the said quantization cell weighted by the values of the continuous distribution restricted to the said quantization cell;
storing the dequantization vectors associated respectively with the quantization symbols; and
dequantizing the coded signal, representing each vector of the
signal by means of a dequantization vector associated with the quantization signal received.
The decoding method according to the invention makes it possible, by virtue of the estimation of the distribution of the signal within each quantization cell, to calculate an optimum dequantization vector, equal to the centre of mass of the points of each quantization cell weighted by the values of the estimated distribution on this quantization cell. This optimum quantization vector is closer to the mean value of the digital signal before coding than the centre of the quantization cell used in the conventional decoding methods.
The distortion caused during the decoding of the digital signal is consequently reduced, the decoding method according to the invention thus making it possible to increase the ratio of compression to distortion.
According to a preferred characteristic of the invention, which makes it possible to improve the approximation, the continuous distribution model passes through the values of the experimental distribution at the centre of each quantization cell.
According to a preferred version of the invention, the decoding method according to the invention also includes the following steps:
comparing the number of occurrences of the quantization symbols with a threshold value;
calculating, for the quantization cells in which the number of occurrences of the said quantization symbol is greater than the said threshold value, a dequantization vector equal to the centre of mass of the points of the said quantization cell weighted by the values of the continuous distribution restricted to the said quantization cell;
allocating, for the quantization cells in which the number of occurrences of the said quantization symbol is less than the said threshold value, a dequantization vector equal to the centre of the said quantization cell;
storing the dequantization vectors associated respectively with the dequantization symbols; and
dequantizating the coded signal, representing each vector of the signal by the dequantization vector associated with the quantization symbol received.
The decoding method thus makes it possible, for the quantization cells in which the values of the experimental distribution are low and are not sufficiently representative to effect a good approximation of the distribution of these cells, to use, as dequantization vectors, the centre respectively of these quantization cells.
According to an advantageous version of the invention, and in order to facilitate the calculations, the continuous distribution model is a real function with N variables, N being equal to the dimension of the quantization space.
In particular, the continuous distribution model is a linear function by pieces when the dimension of the quantization space is equal to one and is a plane function by pieces when the dimension of the quantization space is equal to two.
Correlatively, the invention proposes a device for decoding a coded digital signal, the said coded signal being quantized by means of quantization symbols associated respectively with quantization cells, characterised in that it has:
means of storing the form of the quantization cells;
means of reading the quantization symbols;
means of constructing an experimental distribution of the quantization symbols on the quantization cells by calculating the number of occurrences of the said quantization symbols in the coded signal;
means of approximating the said experimental distribution by means of a continuous distribution model, the continuous distribution model passing through values of the experimental distribution;
means of calculating a dequantization vector equal to the centre of mass of the points of the said quantization cell weighted by the values of the continuous distribution restricted to the said quantization cell;
means of storing the dequantization vectors associated respectively with the quantization symbols; and
means of dequantizing the coded signal, adapted to represent each vector of the signal by means of a dequantization vector associated with the quantization signal received.
This decoding device has advantages similar to those of the method according to the invention and thus makes it possible to improve the ratio of compression to distortion in the devices for compressing and decompressing digital signals with loss.
According to a preferred version of the invention, which makes it possible to reduce still further the distortion caused on decoding, notably for weakly represented quantization cells, the decoding device also has:
means of comparing the number of occurrences of the said quantization symbols with a threshold value, the said calculation means being adapted to calculate a dequantization vector equal to the centre of mass of the points of a quantization cell weighted by the values, of the continuous distribution restricted to the quantization cell for the quantization cells in
Canon Kabushiki Kaisha
Fitzpatrick ,Cella, Harper & Scinto
Tokar Michael
Tran Anh
LandOfFree
Decoding of a quantized digital signal does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Decoding of a quantized digital signal, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Decoding of a quantized digital signal will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2470577