JPH0678290A - Picture coder - Google Patents

Abstract

(57) [Summary] [Object] It is an object of the present invention to provide an image coding apparatus capable of obtaining high coding efficiency even at a low bit rate. [Structure] A correction circuit 3 that corrects a pixel value at the lower right of a block so that the interpolation error power is minimized, and a prediction circuit 7 that performs extrapolation prediction of a pixel at the lower right of the block and interpolation prediction of other pixels.
, A subtractor 4 for calculating a prediction error, a scalar quantizer 6 for scalar quantizing an extrapolation prediction error, a vector quantizer 5 for vector quantizing an interpolation prediction error, and coding quantization information. An image encoding apparatus having an encoder 11 for

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding apparatus for efficiently coding an image signal.

[0002]

2. Description of the Related Art Image coding technology has been used for image communication,
It is used for broadcasting and storage. Image encoding is carried out for the purpose of transmitting as fast as possible in the case of still image transmission such as facsimile, and in the case of moving image communication such as video conferencing and videophone, it uses a narrow band or low bit rate as much as possible. It is done for the purpose of transmission. Further, when recording image information on a disk or a memory, an image encoding technique is used to efficiently record as many images as possible.

For image coding, methods such as predictive coding, transform coding and vector quantization are known. Predictive coding is a method of predicting a pixel value to be currently coded from surrounding already coded pixel values, and quantizing and predicting the prediction error. Transform coding is a method in which an image is subjected to orthogonal transform such as Fourier transform or cosine transform to cause energy deviation, and the energy deviation is used to encode only a large energy region. Vector quantization is a method in which patterns that frequently appear in images are registered in advance as a codebook, the pattern closest to the pattern in the image to be encoded is searched from the codebook, and the pattern number is encoded. is there.

Among these, the predictive coding is not sufficient in terms of coding efficiency in order to be able to transmit a high quality image at a low transmission rate. In comparison, transform coding and vector quantization have high coding efficiency, but they are still insufficient at low bit rates. The transform coding has a drawback that the subjective quality is greatly deteriorated due to block distortion when the transmission rate becomes low. Further, the vector quantization has a drawback that if the coding is directly performed on the level of the image signal, it becomes difficult to represent various images with a limited image pattern in the codebook, and the coding efficiency is deteriorated.

There is also known a method of improving the efficiency of vector quantization by performing vector quantization on a prediction residual signal by combining predictive coding and vector quantization.
However, this method is also inefficient in conventional predictive coding, and therefore cannot sufficiently bring out the performance of vector quantization.

Further, as a predictive coding method, an extrapolative interpolative predictive coding method with a very small amount of calculation and high coding efficiency (IEICE Spring National Convention, D-86, 198).
9) is proposed. This method can obtain high coding efficiency at high bit rates, but has a problem that it is weak against image noise and quantization noise at low bit rates.
Sufficient performance was not obtained.

[0007]

As described above, in the conventional image coding system, although a high coding efficiency can be obtained at a high bit rate, a prediction error is large due to image noise and quantization noise at a low bit rate. Therefore, there is a problem that the coding efficiency is reduced. The present invention has been made in order to solve the above-mentioned problems, and an object of the present invention is to provide an image coding apparatus that can obtain high coding efficiency even at a low bit rate.

[0008]

In order to solve the above-mentioned problems, the image coding apparatus of the present invention comprises a block dividing means for dividing an input image signal into blocks composed of a plurality of pixels,
Prediction means for predicting the first pixel value to be coded in the block of interest using a plurality of already coded second pixel values to obtain a prediction value, and the interpolation error in the block being the minimum To correct the first pixel value so that the difference between the predicted value obtained by the predicting means and the first pixel value obtained by the correcting means is quantized to obtain a quantized value. Scalar quantizing means, interpolation predicting means for interpolating and predicting other pixel values in the block using the quantized first pixel values and a plurality of already encoded second pixel values, A vector quantization means for vector-quantizing the prediction error of the prediction value obtained by the interpolation prediction means; and an encoding means for encoding the quantization information obtained by the scalar quantization means and the vector quantization means. It is characterized by having.

[0009]

According to the present invention, the pixel value at the lower right of the block of interest is not subjected to predictive coding as it is, but is corrected after being corrected so as to minimize the interpolation prediction error in the block of interest.
In the past, if the pixel value at the lower right of the block was affected by noise, the interpolation prediction error was increased. However, by correcting the interpolation prediction error to the minimum, the influence of noise can be reduced. Further, since the power of this interpolation prediction error signal is extremely low, it is possible to obtain higher coding efficiency by collectively performing vector quantization.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an image coding apparatus according to an embodiment of the present invention.

This image coding apparatus includes a memory 2, a correction circuit 3, a subtractor 4, a vector quantizer 5, a scalar quantizer 6, a prediction circuit 7, a selector 8, an adder 9, and a memory 10.
And an encoder 11. The image signal input from the input terminal 1 is written in the memory 2 for one screen,
It is divided into a plurality of blocks and read one by one in the order of encoding.

The correction circuit 3 corrects the pixel value at the lower right of the block so that the interpolation error power becomes minimum. Figure 2
Will be used to explain the interpolation method. The figure shows an example in which the block size is 8 × 8. The shaded areas are pixels that have already been encoded. Interpolation is performed using the pixel X _{88 at the} lower right of the block and the pixel in the shaded area. First, the intermediate points X ₈₄ , X ₄₈ , and X ₄₄ are _calculated by the following equation.

[0013]

[Formula 1] X ₈₄ = (X ₈₀ + X ₈₈ ) / 2 X ₄₈ = (X ₀₈ + X ₈₈ ) / 2 X ₄₄ = (X ₈₄ + X ₄₈ + X ₄₀ + X ₀₄ ) / 4 Similarly, the intermediate point is further set. The pixels are calculated one after another, and all the pixels in the block are interpolated. At that time, the interpolation error power J is expressed by the following equation.

[0014]

## EQU2 ## J = Σ (Y _ij -C _ij X ₈₈ -D _ij ) ²

Where Y _ij is the pixel value of the input image, C _ij X
₈₈ is the contribution component of the pixel X ₈₈ , and D _ij is the contribution component of the pixel in the shaded area. Partial differentiation of J by X _{88 is performed} in order to obtain X ₈₈ that minimizes the interpolation error power J.

[0016]

[Equation 3]

[0017]

[Formula 4] X₈₈= ΣC_ij(Y_ij-D_ij) / ΣC_ij ²  Becomes In other words, the optimum X according to the above formula₈₈Ask for.

The pixel X88 in the lower right corner of the modified block is
The prediction circuit 7 performs extrapolation prediction using the pixels in the shaded area in FIG. Extrapolation prediction methods include, for example, a method of predicting with an average value of X ₈₀ and X ₀₈ and a method of weighted addition. The method by weighted addition is, for example,

[0019]

[Equation 5] P (X ₈₈ ) = (| X ₈₀ −X ₀₀ | X ₈₀ + | X ₀₈ −X ₀₀ | X ₀₈ ) / (| X ₈₀ −X ₀₀ | + | X ₀₈ −X ₀₀ |) To do.

The difference between the predicted value P (X ₈₈ ) and X ₈₈ is quantized by the scalar quantizer 6 and coded by the coder 11. Further, it is inversely quantized by an inverse quantizer (not shown), is added to the predicted value by the adder 9 through the selector 8, and is locally decoded and written in the memory 10.

Next, the pixel in the block is interpolated and predicted by the prediction circuit 7 using the locally decoded lower right pixel X ₈₈ and the pixel in the shaded area in FIG. 2, and the interpolated prediction error is a vector quantum. Vector quantization is performed by the quantizer 5. The vector quantizer has a codebook with various patterns,
A pattern that best matches the interpolation prediction error pattern is searched for, and the pattern number is encoded by the encoder 11. The vector quantization pattern is inversely quantized by an inverse quantizer (not shown), passes through the selector 8 and is added to the prediction value by the adder 9 to be locally decoded and written in the memory 10. FIG. 3 is a block diagram showing the basic configuration of an image coding apparatus according to another embodiment of the present invention.

This image coding apparatus has a class classification circuit 13 and an error power evaluation circuit 1 newly added to the image coding apparatus shown in FIG.
4 is added. In the image code or device of FIG. 1, the block size is fixed, but in this image code or device, the block of interest is divided into smaller blocks according to the class classification information of the block of interest and the error power evaluation result. .

The class classification circuit 13 calculates the variance of the pixel values of the block of interest and classifies it according to the magnitude of the variance. Depending on this class classification, it is determined how small the block should be divided into.

In the error power evaluation circuit 14, the interpolation prediction error power or the error power after vector quantization of the interpolation prediction error is calculated and compared with a predetermined value. When the error power is larger than a predetermined value, block division is performed according to the class classification.

FIG. 4 is a diagram for explaining how blocks are divided for each class. The black points are the points where the extrapolation prediction error is scalar quantized. The block written as V is a block for vector-quantizing the interpolation prediction error. In the case of class 0, the prediction error of the pixel in the lower right of the block is scalar-quantized, and all the other pixels in the block are interpolated. That is, the case where the interpolation prediction error is 0 is shown. In the case of class 1, the prediction error of the lower right pixel of the block is scalar-quantized, the pixel in the block is interpolated, and the prediction error is vector-quantized. In the case of class 2, the block is divided into four, the prediction error of the pixel at the lower right of the block is scalar-quantized for each small block, and other pixels in the block are interpolated. . When the interpolation error power is larger than a predetermined value, the interpolation prediction error is further vector quantized. In the case of class 3, first, the block is divided into four, the prediction error of the pixel in the lower right of the block is scalar-quantized for each small block, and the pixel in the block is interpolated. When the interpolation error power is larger than a predetermined value, the small block is further divided into four, and the same operation as class 2 is performed. The scalar quantization information, the vector quantization information, the quars information, and the division information are encoded by the encoder 11 and transmitted.

The class may be classified not by the variance of the input image but by the first step interpolation prediction error power. At that time, the output of the subtractor 4 is input to the class classification circuit.

FIG. 5 is a graph showing the coding performance.
The image used is an image of a woman wearing a straw hat called "Lena". Extrapolation and interpolation VQ is the variable block size scheme of the present invention. JPEG is an international standard method for still image coding. Extrapolation and interpolation V at 0.5 bpp
Q has 0.7 dB SNR higher than that of JPEG,
At 2 bpp, a 2.0 dB higher result was obtained.

[0028]

According to the present invention, the pixel value in the lower right corner of the block of interest is not subjected to predictive coding as it is, but is corrected after being corrected so as to minimize the interpolation prediction error in the block of interest. In the past, if the pixel value at the lower right of the block was affected by noise, the interpolation prediction error was increased. However, by correcting the interpolation prediction error to the minimum, the influence of noise can be reduced. Also, this interpolated prediction error signal is
Since the power is extremely low, it is possible to obtain higher coding efficiency by performing vector quantization collectively.

[Brief description of drawings]

FIG. 1 is a block diagram of an image coding apparatus according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining a method of interpolation

FIG. 3 is a block diagram of an image coding apparatus according to another embodiment of the present invention.

FIG. 4 is a diagram showing a state of block division for each class.

FIG. 5 is a diagram showing encoding performance.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input terminal 2 ... Memory 3 ... Correction circuit 4 ... Subtractor 5 ... Vector quantizer 6 ... Scalar quantizer 7 ... Prediction circuit 8 ... Selector 9 ... Adder 10 ... Memory 11 ... Encoder 12 ... Output terminal 13 ... Class classification circuit 14 ... Error power evaluation circuit

Claims (2)

[Claims] 1. A block dividing means for dividing an input image signal into blocks made up of a plurality of pixels, and a plurality of second pixels in which a first pixel value to be encoded in a target block has already been encoded. A prediction unit that obtains a prediction value by performing prediction using a value; a correction unit that corrects the first pixel value so that the interpolation error in the block of interest is minimized; and a prediction value obtained by the prediction unit. A scalar quantizing means for quantizing the difference between the first pixel value and the modified value obtained by the modifying means to obtain a quantized value; and a value obtained by decoding the quantized value obtained by the scalar quantizing means. Interpolation predicting means for interpolating and predicting pixel values in the target block using the plurality of already encoded second pixel values, and prediction of the other pixel values obtained by the interpolation predicting means Vector quantization of the value prediction error And torr quantization means, the picture coding apparatus, wherein a quantization information obtained by said scalar quantization means and the vector quantization unit and a coding means for coding. 2. A block dividing means for dividing an input image signal into blocks composed of a plurality of pixels, a class classification means for classifying a block of interest by variance, and a first pixel value to be encoded in the block of interest. Predicting means for predicting ## EQU1 ## by using a plurality of second pixel values that have already been encoded, and correction for correcting the first pixel value so that the interpolation error in the target block is minimized. Means, a scalar quantization means for obtaining a quantized value by quantizing a difference between the predicted value obtained by the predicting means and the corrected value of the first pixel value obtained by the correcting means; Interpolation prediction means for interpolating and predicting other pixel values in the target block using the decoded value quantized by the means and the plurality of already-encoded second pixel values, With this interpolation prediction means Vector quantization means for vector quantizing the prediction error of the obtained prediction value, and the interpolation prediction error power or the error power after vector quantization of the prediction error is compared with a predetermined value. An error power evaluation unit for dividing the block, and an encoding unit for encoding the quantization information, the class classification information and the block division information obtained by the scalar quantization unit and the vector quantization unit. Image encoding device.