JPH03114381A - Picture coding and decoding system, picture coder and picture decoder - Google Patents

Abstract

PURPOSE:To improve coding efficiency by adapting a code word assignment in variable word length coding in the case of coding to a dynamic range of a current prediction error signal and adapting the decoding in the variable word length decoding to the dynamic range of the current input prediction error signal in the case of decoding. CONSTITUTION:A prediction coding circuit consists of a subtraction circuit 12, a prediction circuit 13, a quantization circuit 14, an inverse quantization circuit 15 and an adder circuit 16, and a prediction decoding circuit consists of an inverse quantization circuit 23, an adder circuit 24 and a prediction circuit 25. In the case of the coding, the code word assignment in the variable word length coding is adapted to a dynamic range of a current prediction error signal according to a picture element of a local decoded signal obtained in the process of the past prediction coding and the decoding in the variable word length decoding is adapted to a dynamic range of a current input prediction error signal according to a picture element of the decoded picture signal obtained in the past prediction decoding process in the decoding. Thus the coding efficiency is improved.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) This invention relates to an encoding/decoding method used to highly efficiently encode and transmit an image signal such as a television signal, and this method. This applies to image encoding devices and image decoding devices to which the method is applied.

(Prior art) Predictive coding, orthogonal transform coding, vector quantization coding, etc. are known as coding techniques for image signals such as television signals, and combinations of these known coding methods are also known. There is also a method to remove redundancy in image information. Furthermore, when aiming for higher efficiency, variable word length encoding (entropy encoding) is used. When a fixed rate transmission system is used for variable word length encoding, a method is used in which a smoothing buffer is used to maintain a constant transmission rate before encoding is performed. In encoding broadcast television signals, high image quality is often required while allowing a relatively high bit rate, and many encoding systems suitable for this have already been published.

By the way, in recent years, as seen in image transmission using the asynchronous transfer mode (packet transmission using fixed-length cells, hereinafter referred to as ATM) of wideband 15DN, instead of targeting fixed-speed transmission lines, variable-speed ( A transmission system using a transmission system (hereinafter referred to as rate-free) is about to be realized. In the transmission of television signals using ATM networks,
There is no need to send the encoded output to the transmission path at a constant rate, and the transmission rate of the encoded output can be changed depending on the image information. As a result, in principle, by increasing the amount of encoded output information for parts of the image signal where the encoding quality would deteriorate in a fixed speed transmission system, such as fine parts of the image and parts with intense movement, Transmission with constant quality is possible.

As mentioned above, encoding that allows transmission with constant quality (hereinafter referred to as constant quality encoding) is achieved by adaptively storing necessary information according to the characteristics of the image. A method such as this is generally used.

(1) Adaptively change encoding parameters (quantization step number, etc.).

(2) Use entropy encoding.

In (1), for example, the number of quantization steps is made different between fine parts and flat parts of an image, and adaptive processing is performed so as to keep the encoding distortion constant.

On the other hand, (2) takes advantage of the reversibility of variable word length coding (no coding distortion occurs) and removes redundancy while maintaining the quality of other coding that is normally performed in the previous stage. It is often used in combination with (1).

Among these, method (1) has a drawback in that it is extremely difficult to quantitatively evaluate the quality of the image. Usually, RMS of coding distortion (
root mean-square) value is often used. However, in an actual device, it is not easy to control the encoding parameter without calculating the RMS value.

On the other hand, according to variable word length encoding with reversibility (2), there is no need to calculate encoding distortion in real time before encoding, but there is a drawback that encoding efficiency is not sufficient ( Problems to be Solved by the Invention) As mentioned above, variable word length coding, which is one of the constant quality coding methods used for rate-free transmission lines, is ideal in that there is no coding distortion, but on the other hand, There was a problem that the encoding efficiency was not high.

The present invention aims to provide an image encoding/decoding method, an image encoding device, and an image decoding device that are more efficient than normal variable word length encoding while maintaining the reversibility of variable word length encoding. purpose.

[Structure of the Invention] (Means for Solving the Problem) In order to achieve the above object, the present invention encodes an image signal by a combination of predictive encoding and variable word length encoding, and encodes the encoded image signal. In an image encoding/decoding method that performs decoding by a combination of variable word length decoding and pre-Δaj decoding,
During encoding, codeword assignment in variable word length encoding is adapted to the dynamic range of the current prediction error signal, based on the pixel values of the locally decoded signal obtained in the past predictive encoding process, and decoding is performed. In this process, the decoding operation in variable word length decoding is adapted to the dynamic range of the current input prediction error signal according to the pixel values of the decoded image signal obtained in the past predictive decoding process. be.

Codeword assignment in variable word length encoding is determined, for example, according to the probability density within a predetermined distribution range of interest in the prediction error distribution of the prediction error signal. In that case, adapting the codeword assignment to the dynamic range of the current 7111 error signal is done by translating the distribution range of interest according to the pixel values of the locally decoded signal.

In the present invention, quantization in predictive encoding is particularly effective in the case of nonlinear quantization.

The image encoding device according to the present invention includes means for obtaining a prediction error signal that is a difference between an input image signal and a prediction signal predicted from a locally decoded signal obtained in a past pre-p1 encoding process; means for quantizing a signal to obtain a quantized prediction error signal; and variable word length encoding for outputting a coded image signal by variable word length encoding the quantized prediction error signal according to statistical properties modeled in advance. means for dequantizing a quantized prediction error signal to obtain a decoded prediction error signal; means for adding the recorded/decoded pre-4pj error signal and the prediction signal to obtain a locally decoded signal; means for generating a prediction signal from the encoded signal; and means for adapting the word length allocation in the variable word length encoding means to the dynamic range of the pre/fpj error signal in accordance with the locally decoded signal.

Further, the image decoding device according to the present invention includes a variable word length decoding means for obtaining a quantized prediction error signal by variable word length encoding an encoded image signal obtained by variable word length encoding of a prediction error signal, and a quantized prediction error signal. means for obtaining a decoded prediction error signal by dequantizing the error signal, and a sum signal of the decoder 7111 error signal and a prediction signal pre-Δallj calculated from pixel values of past decoded image signals as a decoded image signal. means for generating a prediction signal from the decoded image signal; and means for adapting the decoding operation in the variable word length decoding means to the dynamic range of the prediction error signal in accordance with the decoded image signal.

(Operation) It is generally known that the probability density of a pre/l1lj error signal obtained when an image signal such as a television signal is predictively encoded has a Laplace distribution. This property is preferable for coding in which the probability density must be known in advance, such as Huffman coding, which is one type of variable word length coding. In other words, the closer the probability density function model used to create the Huffman code is to the probability density of the input signal to the actual Huffman encoding circuit, the more ideal the encoding will be, and if they match perfectly, becomes optimal, and theoretically it cannot get any better.

Here, consider, for example, an 8-bit quantized digital image signal as the input image signal, and assume that the 28 pixel values included in this image signal are 0 to 255. At this time, the predicted ApJ error signal obtained by predictive encoding will exist in the range of -255 to 255. As is known in a reflexed quantizer, the pixel value X input at a certain time i is known. For example, if x is -128, the difference from the next input pixel value Xi+1 is at most -128~1
27, which is finite and 1/of the maximum prediction error signal range.
It is 2. Therefore, the number of quantization sweeps at x, -128 may be 1/2 of that when X is arbitrary (0 to 255), so the same quantization can be performed with fewer code words.

Furthermore, assuming that the value of It becomes possible to do so. This is also known as sliding quantization.

The present invention applies a similar principle to variable word length coding (particularly Huffman coding), and on the coding side, prediction error distribution is calculated according to the pixel value X of the locally decoded signal obtained at time i in predictive coding. Among them, the interest distribution range that is the target of code word assignment in variable word length coding is shifted by X, (
In variable word length coding, the prediction error signal for the next input image signal is assigned the optimal codeword (probabilistically shortest) by performing parallel translation). code word).

Figure 5 (a) (b) (c) shows the current plan 111 like this.
1j The state of codeword assignment in Huffman coding adapted according to the dynamic range of the error signal is shown on the prediction error distribution with X as a parameter. In both cases, the horizontal axis represents the prediction error signal e1, and the vertical axis A probability density function P (e) that is predetermined is taken respectively. X,
The overall shape of the prediction error distribution shown by the solid line and broken line is constant (Laplace distribution) with respect to changes in , but among these, the distribution range of interest shown by the solid line, that is, the range targeted for code word assignment in Huffman encoding. is shifted in the horizontal axis direction by an amount of X.

FIG. 5(a) shows the case of X-Δ (twilight 0). In this case, the distribution range of interest for the pre-1lFj error distribution is 1∆≦e≦255−∆, and otherwise the probability density is 0.
become. Therefore, a short code is not assigned to the prediction error signal in the range of e × Δ, which has a probability density of 0.
On the other hand, a shorter code can be given to a prediction error signal with a large probability density, improving coding efficiency.

In addition, Fig. 5(b) and (c) are the cases of x, -128° ,(
In e), a similar effect can be obtained by shifting -255+Δ≦e≦Δ.

On the other hand, on the decoding side, the decoding operation in variable word length decoding is adapted to the dynamic range of the current input prediction error signal according to the pixel value X of the decoded image signal obtained in the past predictive decoding process. let

Furthermore, if the parameter X is a pixel that has been encoded and decoded, the decoding side can also know this without additional information, so there is no increase in the amount of extra information, and it is completely reversible. Because it maintains the same characteristics, there is no increase in encoding distortion.
The necessary encoding efficiency can be improved.

(Example) Hereinafter, the present invention will be described in detail with reference to the drawings.

FIG. 1 and FIG. 2 are block diagrams showing an embodiment of an image encoding device and an image decoding device, respectively, to which the image encoding/decoding method of the present invention is applied.

In FIG. 1, an image signal input to an input terminal 11 is subtracted from a pre-71pj signal from a prediction circuit 13 by a subtraction circuit 12. The prediction error signal output from the subtraction circuit 12 is quantized by a quantization circuit 14. In addition, quantization circuit 1
Although the quantization characteristic of No. 4 may generally be a linear quantization characteristic, the effect of adaptive Huffman encoding becomes greater when a nonlinear quantization characteristic is used. This will be discussed later.

The quantization pre-7ipj error signal output from the quantization circuit 14 is dequantized by the dequantization circuit 15 to become a decoded prediction error signal. This decoded prediction error signal is sent to the adder circuit 16
It is added to the output of the prediction circuit 13 and becomes a locally decoded signal.

The subtraction circuit 12, the prediction circuit 13, the quantization circuit 14, the inverse quantization circuit 15, and the addition circuit 16 constitute a predictive encoding circuit.

The quantized prediction error signal from the quantization circuit 14 is also input to an adaptive Huffman encoding circuit 17, which is a variable word length encoding circuit. The characteristics (particularly code word assignment) of the adaptive Huffman encoding circuit 17 are controlled by the adaptive control circuit 18, and as shown in FIG.
The probability density function model in the error distribution, that is, the interest distribution range to which codeword assignment is targeted shifts.

Here, the probability density function model is determined by the input image signal and the quantization characteristics of the quantization circuit 14, and its shape is, for example, a Laplace distribution.

The encoded image signal output from the adaptive Huffman encoding circuit 17 is sent to a rate-free transmission path (not shown) via an output terminal 19.

Next, the image decoding device will be explained.

In FIG. 2, the encoded image signal input to the input terminal 21 via the transmission path is transmitted to the adaptive Huffman decoding circuit 2.
2. The adaptive Huffman decoding circuit 22 performs decoding corresponding to the adaptive Huffman encoding in the encoding device shown in FIG. An error signal is output. This quantization pre/Il error signal is applied to one input terminal of an adder circuit 24 via an inverse quantization circuit 23, which is the same as the inverse quantization circuit 15 in FIG. The other input terminal of the adder circuit 24 receives a prediction signal output from a prediction circuit 25 that is the same as the prediction circuit 13 in the image encoding device shown in FIG. The output signal of the adder circuit 24 is sent to the output terminal 27 as a decoded image signal output, and is also input to the adder circuit 25 and the adaptation control circuit 26.

Inverse quantization circuit 23, addition circuit 24 and pre-1 circuit 25
A predictive decoding circuit is configured by:

The adaptive control circuit 26 adaptively controls the encoding operation of the adaptive Huffman decoding circuit 22 as described later. Note that if the adaptive Huffman decoding circuit 22 can be controlled in response to 256 levels of pixel values, the adaptive control circuit 26 is not particularly necessary.

3 and 4 show specific configuration examples of the adaptive Huffman encoding circuit 17 and the adaptive Huffman decoding circuit 12 in FIGS. 1 and 2, respectively.

In FIG. 3, the pre-AP from the quantization circuit 14 of FIG.
The prediction error signal which is the I encoding result is sent to the code length generation circuit 3.
1 and is input to the code word generation circuit 3-2. The code length generation circuit 31 and the code word generation circuit 32 are, for example, a ROM.
(read-only memory).

The code length generation circuit 41 stores the input prediction error signal in the ROM.
is received as part of the address data, and generates data having the length of a code word corresponding to the value of the prell1lJ error signal and supplies it to the register circuit 33. On the other hand, the code word generation circuit 32 similarly receives the input prediction error signal as part of the address data of the ROM, and generates an n-bit long code word (n is an integer) corresponding to the value of the prediction error signal. , this n-bit long code word is output to the register circuit 33 in parallel.

The register circuit 33 performs parallel-to-serial conversion on the code word input from the code word generation circuit 32 in units of code length (n bits) and outputs the converted code word to the output terminal 19 . In this case, since the register circuit 33 is informed of the length of the code word from the code word generation circuit 32 by the code length generation circuit 31, such parallel-to-serial conversion for each code word is possible.

Then, a code length generation circuit 31 and a code word generation circuit 32
The output of the adaptation control circuit 18 is manually input to the other part of the address input of the ROM, and by switching the address space of the ROM by the output of the adaptation control circuit 18, adaptation of code word assignment is performed. It is done.

In FIG. 4, a serial encoded signal input is input to an input terminal 21. In FIG. This encoded signal input is, for example, RO
A code length detection circuit 41 composed of M and a register circuit 4
given to 5.

The code length detection circuit 41 receives a variable length code constituting a serially input encoded signal as part of an address input,
The length of each variable length code (n bit length) is detected and notified to the register circuit 42. Based on the code length information from the code length detection circuit 41, the register circuit 42 performs serial-to-parallel conversion on the encoded signal input from the input terminal 21 in units of n bits, and outputs the converted signal to the decoding circuit 43. The decoding circuit 43 is composed of, for example, ROhl, receives the n-bit parallel variable length code from the register circuit 42 as part of the address input of the ROM, and outputs decoded data to the inverse quantization circuit 13 shown in FIG. .

Here, the output of the adaptation control circuit 26 is input to other address inputs of the code length detection circuit 41 and the decoding circuit 436, and the output of the adaptation control circuit 26 causes the ROM to be
Adaptation of the decoding operation is achieved by switching the address space of the decoder.

The adaptive control circuits 18 and 26 have the same configuration, and input the local decoded signal obtained in the process of predictive encoding and the final decoded image signal, respectively, and calculate the interest in the prediction error distribution shown in Fig. 5. Generate data on the amount of shift in the distribution range,
This circuit controls adaptation of the adaptation/1 Huffman encoding circuit 17 and the adaptive Huffman decoding circuit 22.

Here, the shift a may be the locally decoded signal or the decoded image signal itself, or may be limited to several stages. In the former case, the adaptation control circuit 18.26 is not required.

The adaptive variable word length encoding (for example, Huffman encoding) according to the present invention is more effective when combined with nonlinear quantization as described above, and this will be explained below.

Since variable word length encoding is reversible, it is usually used in combination with linear quantization.

However, in the case of image coding, complete reversibility is often not required (particularly in the case of television signals, there are many examples of lossy coding), and the effect of masking visual distortions (for example, at the edges of the image) (distortion is an effect that is difficult to detect), the compression effect is further enhanced by lossy encoding.

Focusing on this point, in the present invention, the characteristics of the quantization circuit 14 are changed to nonlinear quantization, which is an irreversible characteristic, in predictive encoding before adaptive variable word length encoding (which is reversible encoding). By using the characteristics, it is possible to significantly compress the amount of encoded output information while keeping visual image quality deterioration to a low level.

The effectiveness of using nonlinear quantization in predictive coding before adaptive variable word length coding will be explained. FIG. 6 shows the prediction error distribution in predictive encoding. (a) is an example of a discrete prediction error distribution when linear quantization is performed with a relatively large number of steps,
This is almost the same as the prediction error distribution shown by the broken line before quantization. (b) is a discrete prediction error distribution when nonlinear quantization is performed. In the nonlinear quantization shown in FIG. 6(b), as the absolute value let of the prediction error increases, the quantization threshold interval (quantization step width) becomes wider, so the prediction error distribution changes even when let increases. Convergence is slow. In other words, the prediction error distribution is widening.

In order to variable word length encode a prediction error signal having a prediction error distribution that does not converge much when el is large, as in the case of using this nonlinear quantization, adaptive variable word length encoding in the present invention is used. is particularly effective. In this adaptive variable word length encoding, Fig. 5 (a) (b) (c
), the wider the prediction error distribution, the less the proportion of short code words assigned to areas outside the dynamic range of the current API error signal, that is, the range of the prediction error signal that does not originally exist. As a result, a relatively short code word can be allocated without waste even to a larger prediction error signal with a relatively low degree of occurrence among the originally existing pre/3pJ error signals, improving coding efficiency. We can expect more.

In the description of the above embodiment, Huffman coding was used for variable word length coding, but the present invention is also effective when using other variable length coding.

Furthermore, when all of these variable word length encodings are entropy encoding, the present invention becomes adaptive entropy encoding.

[Effects of the Invention] According to the present invention, when an image signal such as a television signal is encoded by a combination of predictive encoding and variable word length encoding, additional information can be stored without increasing encoding distortion. It is possible to effectively reduce the amount of encoded output information without giving it to the decoding side, and the encoding efficiency is significantly improved.

Moreover, this effect is not impaired even when combined with nonlinear quantization, which is generally used for predictive coding in conventional methods, and in fact, it is possible to further improve coding efficiency.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an image encoding device according to an embodiment of the present invention, FIG. 2 is a block diagram similarly showing the configuration of the image decoding device, and FIG. 3 is an adaptive Huffman code in FIG. 1. Block diagram showing the specific configuration of the conversion circuit, No. 4
The figure is a block diagram showing the specific configuration of the adaptive Huffman decoding circuit in Figure 2, and Figure 5 (a) (b) (c).
FIG. 6 is a diagram showing how code words are assigned on a prediction error distribution using encoded output values as parameters to explain the principle of adaptive Huffman encoding according to an embodiment of the present invention.
a) and (b) are diagrams showing prediction error distributions during linear quantization and nonlinear quantization for explaining the effectiveness when nonlinear quantization is combined with the present invention. 11... Image signal input terminal, 12... Subtraction circuit, 1
3...Pre-Jl11 circuit, 14...Quantization circuit, 15
...Dequantization circuit, 16...Addition circuit, 17...
adaptive Huffman encoding circuit, 18... adaptive control circuit,
1...Encoded image signal output terminal, 21...Encoded image signal input terminal, 22...Adaptive Huffman decoding circuit, 23...Dequantization circuit, 24...Addition circuit, 25
...Prediction circuit, 26...Adaptation control circuit, 27...
-Decoded image signal output terminal.

Claims (5)

[Claims] (1) An image signal is encoded by a combination of predictive coding that quantizes a prediction error signal for a prediction signal predicted from a locally decoded signal and variable word length coding, and the encoded image signal is converted into a variable word length In an image encoding/decoding method that performs decoding by a combination of decoding and predictive decoding that performs decoding from a prediction signal predicted from a decoded image signal and an input prediction error signal, when encoding, past prediction codes are used. According to the pixel values of the locally decoded signal obtained in the decoding process, the code word assignment in variable word length coding is adapted to the dynamic range of the current prediction error signal, and during decoding, the code word assignment in the variable word length encoding is adapted to the dynamic range of the current prediction error signal. An image encoding/decoding method characterized in that a decoding operation in variable word length decoding is adapted to the dynamic range of a current input prediction error signal according to a pixel value of a decoded image signal. (2) The code word assignment in the variable word length encoding is determined according to the probability density within a predetermined distribution range of interest in the prediction error distribution of the prediction error signal, and the code word assignment is determined based on the current prediction error signal. 2. The image encoding/decoding method according to claim 1, wherein the operation of adapting to the dynamic range of the image is performed by translating the distribution range of interest in accordance with pixel values of the locally decoded signal. (3) The image encoding/decoding method according to claim 1, wherein the quantization in the predictive encoding is nonlinear quantization. (4) means for obtaining a prediction error signal that is the difference between an input image signal and a prediction signal predicted from a locally decoded signal obtained in a past predictive encoding process; and quantizing the prediction error signal. means for obtaining a prediction error signal; variable word length encoding means for outputting a coded image signal by variable word length encoding the quantized prediction error signal according to pre-modeled statistical properties; and the quantization. means for dequantizing a prediction error signal to obtain a decoded prediction error signal; means for adding the decoded prediction error signal and the prediction signal to obtain a locally decoded signal; and calculating the prediction from the locally decoded signal. means for generating a dynamic signal of the prediction error signal;
An image encoding device comprising: means for adapting to a range. (5) variable word length decoding means for obtaining a quantized prediction error signal by variable word length decoding an encoded image signal obtained by variable word length encoding of the prediction error signal; and dequantizing the quantized prediction error signal. means for obtaining a decoded prediction error signal; means for obtaining a sum signal of the decoded prediction error signal and a prediction signal predicted from pixel values of past decoded image signals as the decoded image signal; It is characterized by comprising: means for generating the prediction signal from the image signal; and means for adapting the decoding operation in the variable word length decoding means to the dynamic range of the prediction error signal according to the decoded image signal. Image decoding device.