JP2002064824A - Method and apparatus for encoding image as well as recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent a deterioration of a part of an important structure in an image when transforming and encoding image data by a wavelet transformation or the like. SOLUTION: A method for encoding the image comprises the step of wavelet transformation the image data S0 to obtain a coefficient signal K. The method further comprises the steps of forming a histogram H of the data S0, and calculating an entropy value E of the data S0 from the histogram H. The method also comprises the steps of deciding a quantization bit distribution. Accordingly, a dispersed value of the entropy value E is larger, the quantization bit distribution becomes greater, and quantizing the signal K by the decided quantization bit distribution to obtain a quantization coefficient signal RK. Thus, the signal RK is encoded to obtain encoded data F.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of transforming image data by a method such as a wavelet transform or a discrete cosine transform used to create a JPEG format file, obtaining a coefficient signal, and encoding the coefficient signal. The present invention relates to an image encoding method and apparatus, and a computer-readable recording medium on which a program for causing a computer to execute the image encoding method is recorded.

[0002]

2. Description of the Related Art Various compression algorithms have been proposed in the field of image data compression in image servers of medical networks and general data compression such as communication and filing. For example, image data is subjected to multi-resolution conversion by wavelet transform to obtain coefficient signals of respective resolutions, and bit allocation for quantizing each coefficient signal is determined based on the power (variance) of the coefficient signal. An encoding method has been proposed in which after quantizing a coefficient signal based on the above, each quantized coefficient signal is encoded and compressed as one file. In this manner, an encoding method for encoding image data by performing multi-resolution conversion is referred to as conversion encoding. The total code amount (a value obtained by multiplying the image information amount by the compression ratio) at the time of quantization is predetermined in accordance with the compression ratio.

In the general field of JPEG compression, discrete cosine transform (DCT) is performed after image data is divided into blocks, and bit allocation is determined and quantized to obtain quantized data. Data is encoded to obtain encoded data. In such an encoding method using DCT, data in each block is decomposed into a DC component and an AC component, and the AC component represents a structure from a low frequency to a high frequency in the block. The signal after the discrete cosine transform is in a state as if the image data was subjected to multi-resolution conversion. Therefore, an encoding method using the discrete cosine transform is one of the transform encoding methods.

[0004]

In the above-mentioned transform coding method, the compression ratio is determined by the bit distribution when quantizing the coefficient signal. This bit allocation is determined based on the power (variance) of the coefficient signal. Specifically, a large number of bits are assigned to a portion where the variance of the signal value in the coefficient signal is large, that is, a portion representing an important structure in the image, and a portion where the variance of the signal value is small, that is, the signal value is constant. A small number of bits will be assigned to a flat area.

However, if the image represented by the image data contains a background such as an artificial character or an irradiation field stop, the variance of the coefficient signal also becomes large in that part, so that the bit allocation is determined. In this case, a large number of bits is allocated to that part. As a result, there is a problem that the bit allocation to important structures in the image is reduced, and a portion of the important structures in the image is deteriorated by the encoding.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and provides an image encoding method and apparatus capable of encoding image data without deteriorating important structural parts in an image, and an image encoding method for a computer. It is an object of the present invention to provide a computer-readable recording medium on which a program to be executed is recorded.

[0007]

According to the present invention, there is provided an image encoding method for transforming and encoding image data to obtain encoded data, wherein the image data is converted into a multi-resolution coefficient signal. Based on the bias of the probability density distribution of the image data, a quantization bit distribution of the coefficient signal is determined, and the coefficient signal is quantized based on the bit distribution to obtain a quantization coefficient signal. It is characterized in that the encoded data is obtained by encoding.

As a method of converting image data into a multi-resolution coefficient signal, a wavelet transform, a discrete cosine transform, or the like can be adopted.

The bias in the probability density distribution refers to a bias in the degree of concentration of the size of image data, and can be specifically expressed by, for example, an entropy value of the image data.

[0010] In the image encoding method according to the present invention, the bias of the probability density distribution may be used as an entropy value of the image data.

In the image coding method according to the present invention, the coefficient signal may be obtained by performing a wavelet transform on the image data.

An image encoding apparatus according to the present invention is an image encoding apparatus for transform-encoding image data to obtain encoded data, wherein coefficient signal converting means for converting the image data into a multi-resolution coefficient signal; Bit allocation determining means for determining the quantized bit distribution of the coefficient signal based on the bias of the probability density distribution of data; and quantizing means for quantizing the coefficient signal based on the bit allocation to obtain a quantized coefficient signal And encoding means for encoding the quantized coefficient signal to obtain the encoded data.

In the image coding apparatus according to the present invention, the bias of the probability density distribution may be used as an entropy value of the image data.

In the image coding apparatus according to the present invention, the coefficient signal converting means may be means for obtaining the coefficient signal by performing a wavelet transform on the image data.

[0015] A computer-readable recording medium may be provided as a program for causing a computer to execute the image encoding method according to the present invention.

[0016]

According to the present invention, image data is converted into a multi-resolution coefficient signal, and the quantization bit distribution of the coefficient signal is determined based on the bias of the probability density distribution of the image data. Here, the probability density distribution of the image data is distributed in a wider range for important structures in the image, and a background such as an artificial character or an irradiation field diaphragm is distributed in a narrower range. Therefore, based on the bias of the probability density distribution,
The important structure in the image is determined by determining the quantization bit distribution of the coefficient signal so that the quantization bit distribution is larger in the part with the higher probability density distribution and smaller in the part with the smaller probability density distribution. The coefficient signal can be quantized without deteriorating the object. Therefore, an image obtained by decoding the encoded data obtained by the present invention can be a high-quality image with little deterioration of important structures.

Further, by using the bias of the probability density distribution as the entropy value of the image data, the bias of the probability density distribution can be obtained by a relatively simple calculation.

[0018]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic block diagram showing a configuration of an image coding apparatus according to an embodiment of the present invention. As shown in FIG. 1, an image encoding device 10 according to the present embodiment performs a wavelet transform on image data S0 to obtain a multi-resolution coefficient signal K (a coefficient signal transforming unit) 1; Histogram H of image data S0
Histogram generating means 2 for calculating
An entropy value calculating means 3 for calculating an entropy value E for each signal value of the image data S0 based on the above, a quantization bit allocation determining means 4 for determining a quantization bit allocation of the coefficient signal K based on the entropy value E, Quantizing means 5 for quantizing coefficient signal K based on the determined quantization bit distribution to obtain a quantized coefficient signal RK, and coding means 6 for coding quantized coefficient signal RK to obtain coded data F
And

The wavelet transform means 1 performs a wavelet transform on the image data S0 as follows.
First, as shown in FIG. 2A, the image data S0 is subjected to wavelet transform to obtain four coefficient signals LL1, HL0, L
Decomposed into H0 and HH0. Here, the coefficient signal LL
Reference numeral 1 denotes an image obtained by reducing the length and width of the image to half, and coefficient signals HL0, LH0, and HH0 represent images of vertical edge, horizontal edge, and oblique edge components, respectively.
Then, as shown in FIG. 2B, the coefficient signal LL1 is further subjected to wavelet transform to obtain four coefficient signals LL2 and HL.
1, LH1 and HH1 are obtained. Here, the coefficient signal LL
2 represents an image obtained by further reducing the length and width of the coefficient signal LL1 to 1/2, and the coefficient signals HL1, LH1 and H
H1 represents an image of a vertical edge, a horizontal edge, and a diagonal edge component of the coefficient signal LL1, respectively. And
A coefficient signal K for each of a plurality of resolutions is obtained by repeating the wavelet transform a desired number of times for the coefficient signal LL obtained every time the wavelet transform is performed.

The histogram creating means 2 stores the image data S0
Is created. Here, the image data S0
Represents a tomographic image of the head as shown in FIG. 3A, the histogram of the image data S0 is shown in FIG.
It becomes what is shown in. On the other hand, the image data S0 corresponds to FIG.
As shown in FIG. 4B, when a tomographic image of the head obtained by imaging using the irradiation field aperture is shown, the histogram is as shown in FIG. 4B. That is,
In the case of the tomographic image of the head shown in FIG. 3A, the histogram H includes only the mountain A having a certain degree of variation, but in the case of the tomographic image of the head shown in FIG.
Is composed of a mountain B having variation and a mountain C having almost no variation. The peaks A and B represent the distribution of the signal values of the portion corresponding to the head image in the tomographic image of the head and the portion irradiated directly with radiation, and the peak C is the distribution of the signal value of the edge portion of the irradiation field stop. Represents

The entropy value calculating means 3 calculates the image data S based on the histogram H by the following equation (1).
An entropy value E for a signal value of 0 is calculated.

E (x) = − ΣP (x) · log ₂ P (x) (1) The quantization bit allocation determining means 4 calculates a variance value of the entropy value E, and calculates a coefficient signal based on the variance value. Determine the quantization bit allocation for K. Specifically, the larger the variance value, the larger the number of bits is allocated.

Here, FIG.
A case where the image data S0 representing the image shown in FIG. The coefficient signal K obtained by performing the wavelet transform on the image data S0 representing the image shown in FIG. 3A is as shown in FIG.
The coefficient signal K obtained by performing the wavelet transform on the image data S0 representing the image shown in FIG.
The result is as shown in FIG. Here, as shown in FIG. 5 (b), this coefficient signal K is not limited to the edge part of the head,
The edge portion of the irradiation field stop also has a signal value. Accordingly, in the coefficient signal K, since the variance value increases not only at the edge part of the head but also at the edge part of the irradiation field stop, when the coefficient signal K is quantized, Is also assigned a relatively large number of bits. The total code amount (a value obtained by multiplying the image information amount by the compression ratio) at the time of quantization is predetermined in accordance with the compression ratio. As described above, when a large bit distribution is assigned to the edge portion of the irradiation field stop, the bit distribution for the important head is reduced, and as a result, the image quality of the head image is degraded by encoding.

Therefore, in the present embodiment, the entropy value E and its variance are obtained from the histogram H of the image data S0, and a larger bit allocation is assigned to the coefficient signal K corresponding to the portion having a larger variance. It is. Specifically, in the image shown in FIG. 3B, the edge of the irradiation field stop is shown in FIG.
Since the variance value becomes smaller as shown in FIG. 4, the quantization bit distribution is made smaller, and the variance value becomes larger for the head image as shown in FIG. This is to increase the quantization bit distribution. In the quantization distribution, the quantization bit distribution is allocated to the total code amount so as to be proportional to the variance of the entropy value E.

The quantization means 5 calculates a quantization bit distribution based on the quantization bit distribution determined by the quantization bit distribution determination means 4,
The coefficient signal K is quantized. Specifically, the positions of the image represented by the coefficient signal K of each resolution and the image represented by the image data S0 are associated with each other, and quantization is performed on the signal value of the corresponding position of the image represented by the image data S0. The coefficient signal K is quantized based on the bit allocation. For example, the point P0 of the highest resolution coefficient signal (corresponding to LL0) of the coefficient signal K shown in FIG. 5B corresponds to the edge portion P0 'of the irradiation field stop in the image shown in FIG. 3B. Therefore, quantization is performed so that the quantization bit distribution is reduced. On the other hand, regarding the point P1 of the coefficient signal K shown in FIG. 5B, the point P1 ′ in the head in the image shown in FIG.
In order to cope with the above, quantization is performed so that the quantization bit distribution is increased. Note that scalar quantization can be used as a quantization method.

The encoding means 6 performs entropy coding on the quantized coefficient signal RK obtained by the quantization means 5 to obtain encoded data. Here, as an entropy coding method, bit plane binary arithmetic coding can be used. In bit-plane binary arithmetic coding, quantized data is decomposed into a plurality of bit planes and binarized, and binary arithmetic coding is performed on the data of each bit plane to code each output. Things. Also, multi-value arithmetic coding can be used as entropy coding.

Next, the operation of this embodiment will be described. FIG. 6 is a flowchart showing the operation of the present embodiment. A program for causing a computer to execute the processing shown in the flowchart of FIG.
It may be provided by being recorded on a computer-readable recording medium such as R, FD, and hard disk. First, the image data S0 is wavelet-transformed by the wavelet transform means 1 to obtain coefficient signals K of respective resolutions (step S1). On the other hand, a histogram H of the image data S0 is created by the histogram creating means 2 (step S2), and an entropy value E of the histogram H is calculated by the entropy value calculating means 3 (step S3),
Further, in the quantization bit allocation determining means 4, the coefficient signal K
Is determined at the time of quantizing (step S4). Step S1 and steps S2 to S2
The processing of step S4 may be performed in parallel.
The processing of steps 2 to S4 may be performed before the processing of step S1.

Then, the coefficient signal K is quantized by the determined quantization bit distribution (step S5), and the quantization coefficient signal K is encoded by the encoding means 6 (step S6), and the encoded data F is Then, the processing is completed.

As described above, according to the present embodiment, based on the variance of the entropy value E of the image data S0, the larger the variance, the larger the quantization bit allocation, and the smaller the variance, the more the quantization bit allocation. Since the quantization bit distribution of the coefficient signal K is determined so as to be small, the coefficient signal K is quantized without deteriorating important structures in the image represented by the image data S0 like the head image. Can be Therefore, an image obtained by decoding the encoded data F obtained according to the present embodiment can be a high-quality image with little deterioration of important structures.

In the above embodiment, the distribution of the quantized bits of the coefficient signal K is determined based on the variance of the entropy value E of the image data S0. However, the present invention is not limited to this. Any value can be adopted as long as it is based on the bias of the probability distribution of the data S0.

In the above embodiment, the image data S0 is transformed and encoded by the wavelet transform. However, the image data S0 is transformed and encoded by the discrete cosine transform (DCT). Is also good. In this case, in the DCT, the image data S0 is divided into a plurality of blocks, and in each block, a coefficient signal including a DC component and an AC component is created. Here, the AC component represents a structure from a low frequency to a high frequency in the block. Then, the variance value of the entropy value of the coefficient signal obtained by DCT is obtained for each block, and the quantization bit distribution is determined for each block according to the variance value.

For example, as shown in FIG. 7, a cumulative histogram of the variance of the entropy value in each block is obtained, and the variance is classified into four classes I to IV according to the value of the cumulative frequency in the cumulative histogram. Note that the quantization bit distribution is determined in advance for each class. Then, the coefficient signal of each block is quantized based on the quantization bit distribution according to the classification result.

Here, as in the above-described embodiment, since the variance of the entropy value of the important part in the image is large, the more important the part in the image represented by the image data S0 is, the larger the quantization bit allocation is. Quantization will be performed. Therefore, even when image data is transformed by DCT and encoded, an image obtained by decoding the encoded data F has high quality with little deterioration of important structures. be able to.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram illustrating a configuration of an image encoding device according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining wavelet transform on image data.

FIG. 3 is a diagram illustrating an example of an image represented by image data.

FIG. 4 is a diagram showing an example of a histogram of image data.

FIG. 5 is a diagram showing an example of a coefficient signal.

FIG. 6 is a flowchart showing the operation of the embodiment.

FIG. 7 is a diagram showing a cumulative histogram of variance values of entropy values;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Wavelet transformation means 2 Histogram creation means 3 Entropy value calculation means 4 Quantization bit allocation determination means 5 Quantization means 6 Encoding means 10 Image encoding device

Claims (9)

[Claims] 1. An image encoding method for transform-encoding image data to obtain encoded data, wherein the image data is converted into a multi-resolution coefficient signal, and based on a bias of a probability density distribution of the image data, Determining a quantized bit distribution of the coefficient signal, quantizing the coefficient signal based on the bit distribution to obtain a quantized coefficient signal, and encoding the quantized coefficient signal to obtain the encoded data. Image encoding method. 2. The image encoding method according to claim 1, wherein the bias of the probability density distribution is an entropy value of the image data. 3. The image encoding method according to claim 1, wherein the coefficient signal is obtained by performing a wavelet transform on the image data. 4. An image encoding apparatus for transform-encoding image data to obtain encoded data, wherein: a coefficient signal converting means for converting the image data into a multi-resolution coefficient signal; and a bias in a probability density distribution of the image data. Bit allocation determining means for determining a quantized bit allocation of the coefficient signal based on the following: quantization means for quantizing the coefficient signal based on the bit allocation to obtain a quantized coefficient signal; Encoding means for encoding the data to obtain the encoded data. 5. The image encoding apparatus according to claim 4, wherein the bias of the probability density distribution is an entropy value of the image data. 6. The image coding apparatus according to claim 4, wherein said coefficient signal converting means is means for obtaining said coefficient signal by performing a wavelet transform on said image data. 7. A computer-readable recording medium in which a program for causing a computer to execute an image encoding method for transform-encoding image data to obtain encoded data is recorded, wherein the program converts the image data to a multi-resolution image. Converting the coefficient signal into quantized bits based on the bias of the probability density distribution of the image data; andquantizing the coefficient signal based on the bit allocation to perform quantization. A computer readable recording medium comprising: a step of obtaining a quantized coefficient signal; and a step of coding the quantized coefficient signal to obtain the coded data. 8. The computer-readable recording medium according to claim 7, wherein the bias of the probability density distribution is an entropy value of the image data. 9. The computer-readable recording medium according to claim 7, wherein the step of converting into the coefficient signal is a step of obtaining the coefficient signal by performing a wavelet transform on the image data.