JPH07250327A - Image coding method - Google Patents

Abstract

PURPOSE:To reduce quantization distortion by dividing digital image information into plural coding areas, applying prescribed conversion to the information, quantizing the result with a corresponding quantization width, decoding it and adjusting the quantization quantity based on an evaluation corresponding to an error with respect to original information and to a fluctuation in the coded areas. CONSTITUTION:Input digital image information divided into coded blocks is subjected to DC transformation and quantized at a decided quantization width and then decoded. Then an error amount calculation device 11 and a fluctuation calculation device 12 calculate respectively an error quantity epsilonk between original information and reproduced decoded information and a fluctuation amount alphak in a coded area, an evaluation quantity calculation device 13 decides an evaluation value gk. When the value is a threshold value or over, it is decided that the compression coding information has large distortion and the quantization width is adjusted to be small. Thus, the compression coding image information is obtained, in which quantization distortion is reduced even from a flat level portion such as a background.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compression coding method for storing or transmitting digital image data.

[0002]

2. Description of the Related Art A compression encoding process is required for transmitting and recording digital moving image data. Such moving image data compression methods include intraframe coding and interframe predictive coding. In the intra-frame coding, the compression process of one image is completed in one frame, whereas in the inter-frame coding, motion-compensated prediction is performed from adjacent frames to realize high efficiency compression. By applying these, intra-frame coding and inter-frame coding may be combined and coded. In some cases, a series of moving images may be intraframe-encoded, or the first image may be intraframe-encoded and all the remaining images may be interframe-encoded. In some cases, intra-frame coding is periodically performed, and inter-frame predictive coding is performed on an image located between two intra-frame coded images.

Intraframe coding removes redundancy in space, and is an orthogonal transform method represented by discrete cosine transform (hereinafter referred to as DCT), a wavelet transform or subband method for dividing into frequency bands. Is used.
The orthogonally transformed coefficients and the coefficients divided into frequency bands are quantized into a desired transmission amount or storage amount and variable-length coded. At this time, the quantization width must be encoded so that the reproduction side can decode it. When performing DCT, the image is divided into blocks of an appropriate size and then converted.

On the other hand, interframe predictive coding removes redundancy in the time direction. An image of one frame is divided into a plurality of adjacent blocks, a past or future frame is referred to for each block, and a motion vector is obtained under a predetermined evaluation function. Using the obtained motion vector, the reference block at the offset position is used as the prediction signal. Then, the difference between the target block and the block of the prediction signal is calculated, and the intra-frame redundancy is further removed by the above-described intra-frame coding method. A reproduced image is often used as the prediction signal. In that case, the encoded image must be decoded and reproduced.

In addition to the past and future prediction signals, a prediction signal may be created by an average or weighted average of motion-compensated past and future signals. Further, there may be a block that is encoded in the same manner as the intraframe encoding without obtaining the differential signal. That is, there are a plurality of coding modes in an inter-frame coded image. Therefore, in addition to the difference signal, the motion vector and the information about the coding mode must be coded. In the case of the wavelet transform or the sub-band method, the redundancy in the time direction may be removed by dividing into frequency bands, dividing them into blocks, and performing motion compensation.

As described above, when compressing and coding a digital moving image, quantized transform coefficients (or coefficients divided into frequency bands), quantization widths, coding methods (intraframe / interframe), Information such as a mode of a coding block and a motion vector is coded. In addition, information such as frame size, frame rate, and pixel aspect ratio must be transmitted or recorded. This series of encoded data streams is called a bitstream.
On the reproducing side, this bit stream is read and decoded to decompress and reproduce the image.

Quantization noise occurs when the transform coefficient and the like are quantized. A large amount of distortion appears when coarsely quantized with a large quantization width. According to human visual characteristics, the distortion appearing in the flat part of the image is easier to detect with eyes than the complicated part of the image. Therefore, when determining the quantization width, a method of reducing the quantization width of the flat portion and increasing the quantization width of the complicated portion is widely used. Such a technique is disclosed, for example, in US Pat. No. 5,144,424.

[0008]

Although a high compression rate can be realized by the inter-frame predictive coding, the quality of the prediction signal depends on two factors. That is, there are two factors: the performance of motion detection and the quality of the prediction signal.

The performance of motion detection depends on the model of motion adopted. For example, if the motion of an object is approximated by a model that moves only horizontally, the motion of an object that does not move horizontally cannot be detected accurately, and the performance of motion compensation will deteriorate.
The motion model determines the performance of motion detection.

On the other hand, the quality of the prediction signal is the quantization distortion included in the prediction signal. Since the past or future reconstructed image is used as the prediction signal so that the encoding unit and the decoding unit can be matched with each other, the prediction signal includes quantization distortion. In this case, even if the motion detection is accurate, if the prediction signal used for motion compensation contains a lot of distortion, this distortion is left in the difference signal and the quantization distortion is encoded instead of the desired signal. Will be done. In the case of a high compression rate, since the number of bits is small, this difference signal cannot be encoded sufficiently, and even if the difference signal reproduced on the decoding side is added to the prediction signal, the distortion of the prediction signal cannot be canceled. Distortion appears in the reproduced image. Such distortion is especially common when a hidden background appears on the previous screen. This will be described with reference to FIG.

FIG. 2 shows how the diamond-shaped pattern in the middle of the frame 1 spreads through the frames 2 and 3 and then to the frame 4. Consider a case where frame 4 is predicted from frame 1 and frame 2 and frame 3 are predicted from frame 1 and frame 4 for this series of images.

The block (i, j-) in the diamond of frame 4
1), (i-1, j), (i, j), (i + 1, j),
(I, j + 1) is a block (i, j) in the diamond of frame 1.
j) has the same luminance signal. Therefore, the block (i, j) of frame 1 is used as the prediction signal. Here, if a large amount of quantization distortion is included in the block (i, j) of the frame 1, the same distortion is included in the prediction signal of the block in the rhombus of the frame 4, and when the differential signal is obtained, it is almost the same. Becomes quantization distortion. In the case of a high compression rate, this difference signal cannot be sufficiently coded, and the distortion of the prediction signal cannot be canceled on the decoding side, so that the distortion appears in the rhombus. Further, since the prediction signals of the blocks in the rhombus of the frame 2 and the frame 3 are selected from the pixel data in the rhombus of the frame 1 or the frame 4, distortion appears in the reproduced image similarly. In this way, distortion propagates one after another.

As described above, in the conventional technique, the quantization width is determined without considering the distortion amount included in the reproduced image (the reproduced image of the image to be encoded, the reproduced reference image, etc.). Therefore, there is a drawback that a very noticeable distortion appears in the flat portion, especially in the hidden background.

[0014]

In order to solve the above-mentioned problems, the image coding method of the present invention provides a method for easily detecting the distortion of a flat portion, that is, a portion where distortion can be visually detected easily. Introduce and apply this distortion detection method at the time of encoding.

According to a first aspect of the present invention, an original image to be encoded is input, the original image is divided into a plurality of encoding areas, and the encoding area is converted into an encoding conversion area by a predetermined conversion method. After the transformation, the coefficient of the coding transformation area is quantized with the first quantization width, inversely quantized, and inverse transformation is performed to restore the coding area into the reproduction area, and then the following processing is performed. That is, the error amount ε _k is calculated from the reproduction region and the coding region, the variation amount α _k of the coding region is calculated, and the evaluation function f () increases as ε _k increases and decreases as α _k increases. Evaluation value g _{k by} ε _k , α _k ).
If the evaluation value g _k exceeds a predetermined threshold value, the first
The second quantization width is formed by multiplying the quantization width of 1 by a predetermined coefficient, and the coefficient in the coding conversion region is quantized by the second quantization width.

According to a second aspect of the present invention, an original image to be encoded is input, the original image is divided into a plurality of encoding areas, and the encoding area is converted into an encoding conversion area by a predetermined conversion method. After the transformation, the coefficient of the coding transformation area is quantized with the first quantization width, and dequantized to form the quantized transformation area, the following processing is performed. That is, the error amount ε _k is obtained from the quantized transformation region and the coding transformation region, the variation amount α _k of the coding region is obtained, and an evaluation function that increases as ε _k increases and decreases as α _k increases , F (ε _k , α _k ) to obtain the evaluation value g _k ,
When the evaluation value g _k exceeds a predetermined threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width, and the coefficient in the coding conversion region is quantized with the second quantization width. Turn into.

The invention according to claim 3 is different from the invention according to claim 1 or 2 in that instead of the evaluation value g _k of the current coding region, the evaluation value g _m (of the past coding region) g _m ( m
When <k) exceeds a predetermined threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width, and the current coding area is converted into a coding conversion area by a predetermined conversion method. This is an image coding method of transforming and quantizing the coefficient of the coding transform region with the second quantization width.

According to a fourth aspect of the present invention, an original image to be encoded and a reference image are input, the original image is divided into a plurality of encoding regions, and a prediction signal of the encoding region is calculated from the reference image. A region is detected, a difference signal region is obtained from the coding region and the prediction signal region, the difference signal region is converted into a difference signal conversion region by a predetermined conversion method, and the coefficient of the difference signal conversion region is converted into a first quantization width. Quantize, dequantize, and inverse transform to reproduce the difference signal area, add the predicted signal area data to the reproduced difference signal area data, restore the coding area to the reproduction area, and then perform the following processing. . That is, the error amount ε _k is calculated from the reproduction region and the coding region, the variation amount α _k of the coding region is calculated, and the evaluation function f () increases as ε _k increases and decreases as α _k increases. epsilon _k, we obtain the evaluation value g _k with alpha _k), if the evaluation value g _k exceeds a predetermined threshold value, forming a second quantization width by multiplying a predetermined coefficient to the first quantization width, the difference The coefficient in the signal conversion region is quantized with the second quantization width.

According to a fifth aspect of the present invention, an original image to be encoded and a reference image are input, the original image is divided into a plurality of encoding regions, and a prediction signal region from the reference image to the encoding region. Is detected, the difference signal region is obtained from the coding region and the prediction signal region, the difference signal region is converted into the difference signal conversion region by a predetermined conversion method, and the coefficient of the difference signal conversion region is converted to the first quantization width. The following processing is performed after quantized and dequantized to form a quantized differential signal conversion region. That is, the error amount ε _k is calculated from the quantized difference signal conversion region and the difference signal conversion region.
Then, the variation amount α _k of the coding region is obtained, and an evaluation function that increases with an increase of ε _k and decreases with an increase of α _k ,
f (ε _k, α _k) obtains the evaluation value g _k, if the evaluation value g _k exceeds a predetermined threshold value, the second to form a quantization width by multiplying a predetermined coefficient to the first quantization width , The coefficient in the differential signal transform region is quantized with the second quantization width.

Further, the invention according to claim 6 is different from the invention according to claim 4 or claim 5 in that instead of the evaluation value g _k of the current coding area, the evaluation value of the past coding area is used. If g _m (m <k) exceeds a predetermined threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width,
This is an image coding method for quantizing the coefficient of the difference signal conversion region of the current coding region with the second quantization width.

The invention according to claim 7 is an object of encoding.
Input the original image, the reference image, and the original image of the reference image.
The original image to be encoded is divided into multiple coding areas,
After detecting the prediction signal region of the coding region from the reference image
The following processing is performed. The pixels of the prediction signal area and the reference image
The image at the same position as the pixel in the predicted signal area in the original image
From the prime value and the error amount ε_kAnd the variation amount α of the coding area_kTo
Find, ε_kIncreases with the increase of α_kDiminishes with increasing
Evaluation function f (ε_k, Α_k) Is the evaluation value g_kSeeking
It Evaluation value g_kIf the value does not exceed the predetermined threshold,
The difference signal area is obtained from the range and the prediction signal area, and the difference signal
Convert the area to the difference signal conversion area by the specified conversion method,
The coefficient in the split signal conversion region is quantized with the first quantization width.
Evaluation value g _kIs greater than a predetermined threshold, the first quantization width
Is multiplied by a predetermined coefficient to form the second quantization width,
The coefficient in the signal transform domain is quantized with the second quantization width.

According to an eighth aspect of the present invention, an original image to be encoded and a reference image are input, the original image is divided into a plurality of encoding regions, and a prediction signal region of the encoding region is divided from the reference image. Following detection is performed. The error amount ε _k is calculated from the coding region and the prediction signal region, and the variation amount α _{k of the} coding region
Then, the evaluation value g _k is calculated with an evaluation function f (ε _k , α _k ) that increases with increase in ε _k and decreases with increase in α _k . When the evaluation value g _k does not exceed the predetermined threshold value, a difference signal area is obtained from the coding area and the prediction signal area, and the difference signal area is converted into the difference signal conversion area by a predetermined conversion method,
The coefficient in the differential signal transform region is quantized with the first quantization width. When the evaluation value g _k exceeds a predetermined threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width, and the coefficient in the differential signal transform region is quantized with the second quantization width. Turn into.

Further, in the invention described in claim 9, in the invention described in claims 1 to 8, when the evaluation value exceeds a predetermined threshold value, the first quantization width is multiplied by a coefficient smaller than 1 This is an image coding method for forming a quantization width of 2.

According to a tenth aspect of the invention, in the invention of the fourth to eighth aspects, when the evaluation value exceeds a predetermined threshold value, the coding area is coded by a predetermined conversion method instead of the differential signal area. This is an image coding method of transforming into the transform domain and quantizing the coefficient of the coding transform domain.

The invention described in claim 11 is the same as claim 1.
Therefore, in the invention of claim 10, the evaluation function f
(Ε _k , α _k ) is proportional to ε _k / α _k .

According to the twelfth aspect of the invention, in the invention according to the first to eleventh aspects, the variation amount α _k is normalized by the average value of the variation amounts of all the reference regions.

[0027]

According to human visual characteristics, even if the amount of error in the evaluation area is the same, the error is easier to detect visually in the flat portion than in the portion with high spatial frequency.

The present invention utilizes this characteristic to perform weighting so as to enhance distortion in a flat region and weight so as to suppress distortion in a region having a high spatial frequency, so that the strain can be easily detected visually. To detect. By finely quantizing the detected region or by encoding the original signal instead of the differential signal, it has the effect of reducing the distortion that appears in the flat region.

[0029]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Before describing the embodiments of the present invention, common premise of the embodiments described below will be described. First, an image to be coded is divided into a plurality of coding areas. The coding area may have any shape. For example, it is not necessary that the shape and size of a quadrangle block or triangle segment composed of N × M pixels, or the coding area in one screen be the same. Then, the coding area thus divided is converted into a conversion area by a predetermined conversion method. For this conversion, as described in the prior art, a conversion method such as DCT conversion or wavelet method is used. The following embodiment is not limited to a specific conversion method or division method, but for simplification, here, an example is shown in which an image is divided into encoded blocks of N × M pixels by using DCT transformation. Will be described as.

First, a first embodiment of the image coding method of the present invention will be described. FIG. 1 is a flow chart for explaining the first embodiment. First, in step 15, the coding block is DCT-transformed. In step 16, the first quantization width is determined, and the transform coefficient in step 15 is quantized with this first quantization width. Next, the quantized transform coefficient is inversely quantized (step 17) and the inverse DCT transform (step 1) is performed.
8) Then, the encoded block is restored to the reproduction block. Then the reclaim block is evaluated. In step 19, the error amount ε _k is obtained from the reproduced block and the encoded block, the variation amount α _k of the encoded block is obtained, and the evaluation function f (ε _k , α _k ) is obtained from the above ε _k and α _k. The evaluation value g _k is _obtained by

FIG. 3 shows a block diagram for _obtaining the evaluation value g _k in this embodiment. It is composed of an error amount calculator 11, a fluctuation amount calculator 12, and an evaluation value calculator 13. The data of the reproduction block is input to the terminal 9, and the data of the encoding block is input to the terminal 10. In the error amount calculator 11,
The error amount between the reproduction block and the coding block is calculated. This error amount represents quantization distortion. Preferably, the sum of squares or the sum of absolute values of the differences between the pixels may be obtained.

Assuming that an image to be coded is divided into K coded blocks, the k-th (k = 1, 2, ...
.., K), and the error amount between the encoded block and the reproduced block is ε _k . In addition, the coding block is input to the fluctuation amount calculator 12 to obtain the fluctuation amount α _k . It is preferable to obtain the AC energy of the pixels of the entire encoded block or a part thereof. Next, the obtained ε _k and α _k are input to the evaluation amount calculator 13. The function f (ε _k , α _k ) for calculating the evaluation value preferably increases with increasing ε _k and decreases with increasing α _k . This is based on the following ideas.

According to human visual characteristics, the flatness of the image
The distortion that appears in a region is easy to detect, but it can
Even if distortion appears in the high frequency range)
There are many. Therefore, the error amount ε_kEven if the values of
If the area is not flat, distortion will be difficult to see, but flat
If so, the distortion becomes significant. On the other hand, the variation amount α in the flat part_kIs
Since it is smaller than the variation of the complicated part, α_kIs small (flat
Part) sometimes ε_kTo strengthen α _kIs large (complex part), ε_k
Ε to suppress_kHumans by weighting
It is possible to detect the distortion that matches the visual characteristics of. This
Una evaluation function f (ε _k, Α_k) As an example_k/ Α_kTo
There is a function that is proportional. This function satisfies the above conditions
You can add and calculate easily.

By the way, whether or not a certain area of an image is flat is relative. Only when a human sees the entire screen of an image, the flat portion and the complex portion can be distinguished. That is, the variation amount α _k described above is not absolute. In order to relatively represent the variation amount, it is preferable to normalize α _k with the average value <α> of the variation amounts of all reference regions. As an example, g _k = f (ε _k , α _k ) = ε _k / (α _k / <α>) can be considered. However, it is not limited to this.
Note that <α> may be an average value of variation amounts of all regions of the past encoded image, instead of an average value of variation amounts of all regions of the current encoded image. The calculation of the evaluation value described above applies to all the following examples, and therefore will not be repeated below unless particularly necessary.

Now, the evaluation value g _k obtained by the above method is compared with a predetermined threshold value (step 20). g _k
Is greater than the threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width (step 21), and the DCT coefficient of the coding block is quantized by the second quantization width. (Step 22). The predetermined coefficient is preferably a value smaller than 1. For example, 0.3 to 0.5 may be used. By reducing the first quantization width, the DCT coefficient can be finely quantized, which leads to improvement in image quality.
The predetermined coefficient may be a coefficient larger than 1. In this case, since the DCT coefficient is roughly quantized, the number of generated bits is reduced. The next encoded block can be finely quantized by that amount. Finally, the quantized coefficient and other information are variable length or fixed length coded (step 23). If the evaluation value g _k does not exceed the threshold value in step 20, the transform coefficient quantized in step 16 and other information are encoded (step 23).

Although not shown in FIG. 1, there is a case where a process of dequantizing the transform coefficient quantized with the second quantization width and performing inverse DCT transform to restore the coded block to the reproduced block is performed. . This is because in the case of the inter-frame predictive coding method, it is used as a reference image for coding the next image. It is not necessary when only intraframe coding is performed.

In the above description, the error amount ε _k is obtained from the reproduced block and the encoded block. However, the error amount ε _k is obtained from the DCT coefficient before quantization and the DCT coefficient after dequantization.
You may ask for _k . Preferably, the sum of squares or the sum of absolute values of the differences between the coefficients may be obtained. Variation amount α _{k in} this case
Can also be obtained from the DCT coefficient before quantization. Preferably, the sum of squares or the sum of absolute values of the other coefficients (AC components) excluding the DC component of the DCT coefficient may be obtained. That is, in FIG. 3, the DCT coefficient after dequantization is input to the terminal 9, and the DCT coefficient before quantization is input to the terminal 10. In step 19 of FIG. 1, an evaluation value g _k is obtained from the DCT coefficient before quantization and after inverse quantization, and compared with a threshold value in step 20, and the subsequent processing is the same as in FIG.

In step 20 of FIG. 1, "g _k >
Threshold? Instead of ", it is possible to set" g _m > threshold? " However, m <k. That is, the evaluation value of the past encoded block is used as the determination condition. Preferably, m = k-1. The process after step 20 is shown in FIG.
Is the same as. Note that the mth coded block and the kth coded block
It is preferable that the variation amount of the second coded block is approximately the same. This is because even if the kth coded block is flat, the mth coded block is not always flat. If the m-th coding block is complex, it is not necessary to quantize it in detail. Therefore, in step 20 of FIG. 1, it is more preferable to use the determination condition that “g _m > threshold value 1 and (α _m / α _k ) <threshold value 2”. However, the threshold 2
= (1 ± λ), λ << 1.

The second embodiment of the image coding method of the present invention will be described below. FIG. 4 is a flow chart for explaining the second embodiment. This embodiment is an embodiment of an interframe predictive coding system, and a prediction signal is required for each coding block.

Therefore, it is assumed that the optimum prediction signal block of the coding block is detected from the reference image. First,
In step 31, a difference signal block is obtained from the coded block and the prediction signal block. This difference signal block is DCT transformed (step 32). In step 33, the first quantization width is determined, and the DCT coefficient is quantized with this first quantization width. Next, the quantized coefficient is inversely quantized (step 34) and inverse DCT transform (step 35) is performed. In step 36, the prediction signal block is added to the reproduced difference signal block to restore the coded block to the reproduced block.

Next, the reproduction block is evaluated. In step 37, the error amount ε _k is obtained from the reproduced block and the encoded block, the variation amount α _k of the encoded block is obtained, and the evaluation function f (ε _k , α _k ) described in the first embodiment is used. To obtain the evaluation value g _k . Then, in step 38,
The evaluation value g _k is compared with a predetermined threshold value. If the evaluation value exceeds the threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width (step 40), and the DCT coefficient of the difference signal block is set to the second quantization width. To quantize (step 41). The predetermined coefficient is preferably a value smaller than 1. For example, 0.3 to 0.5 may be used. By reducing the first quantization width, the DCT coefficient can be finely quantized, which leads to improvement in image quality. The predetermined coefficient may be a coefficient larger than 1. In this case, since the DCT coefficient is roughly quantized, the number of generated bits is reduced. The next encoded block can be finely quantized by that amount. Finally, the quantized coefficient and other information are encoded in variable length or fixed length (step 4
2) Do. In step 38, when the evaluation value g _k does not exceed the threshold value, the transform coefficient quantized in step 33 and other information are encoded (step 42).

Although not shown in FIG. 4, the coefficient quantized with the second quantization width is inversely quantized and inversely DCT-transformed,
A case may be considered in which a prediction signal block is added to a reproduced difference signal block to restore a coding block to a reproduction block. This is because in the case of the inter-frame predictive coding method, it is used as a reference image for coding the next image.

[0043] Note that to determine the amount of error epsilon _k from the reproducing block and the encoded block, and a DCT coefficient before quantization and inverse quantization after DCT coefficients may be obtained error amount epsilon _k.
Preferably, the sum of squares or the sum of absolute values of the differences between the coefficients may be obtained. In this case, since the DCT coefficient is a difference signal, its variation amount is different from that of the coding block (original image). If the variation amount of the difference signal is small, the coding block is not always flat. Therefore, the variation amount α _k is obtained from the pixels of the encoded block. It is preferable to obtain the AC energy of the pixels of the entire encoded block or a part thereof. That is, in FIG. 3, the terminal 9 is the DCT coefficient after dequantization, and the terminal 10 is the DCT coefficient before quantization.
The coefficient is input, but the input of the variation calculator 12 is the terminal 10
Input the coded block. In step 37 of FIG. 4, an evaluation value g _k is obtained from the DCT coefficients before quantization and after inverse quantization, and compared with a threshold value in step 38, and the subsequent processing is the same as in FIG.

In step 38 of FIG. 4, "g
_k > threshold? Instead of ", it is possible to set" g _m > threshold? " However, m <k. That is, the evaluation value of the past encoded block is used as the determination condition. Preferably, m = k-1. The process after step 38 is shown in FIG.
Is the same as. Note that the mth coded block and the kth coded block
It is preferable that the variation amount of the second coded block is approximately the same. This is because even if the kth coded block is flat, the mth coded block is not always flat. If the m-th coding block is complex, it is not necessary to quantize it in detail. Therefore, it is more preferable to use the determination condition of "g _m > threshold value 1 and (α _m / α _k ) <threshold value 2" in step 38 of FIG. However, the threshold 2
= (1 ± λ), λ << 1.

The third embodiment of the image coding method of the present invention will be described below. FIG. 5 shows a flow chart for explaining the third embodiment. This embodiment is an embodiment of an interframe predictive coding system as in the second embodiment, and a prediction signal is required for each coding block.

It is assumed that the optimum prediction signal block of the coding block has been detected from the reference image. This reference image is a past or future reproduced image in the display order. Hereinafter, this is referred to as a reproduction reference image. Also, the original image of the reproduction reference image is required.

First, in step 51, the first quantization width is determined. Next, the optimum prediction signal block is evaluated.
In step 52, the error amount ε _k is calculated from the pixel of the prediction signal block and the pixel at the same position as the pixel of the prediction signal block in the original image of the reproduction reference image, and the variation amount α _k of the coding block is calculated. An evaluation value g _k is _obtained using the evaluation function f (ε _k , α _k ) described in the first embodiment. That is, the prediction signal is input to the terminal 9 of FIG. 3, the original signal of the prediction signal is input to the terminal 10, the input of the variation calculator 12 shown in FIG. 3 is cut off from the terminal 10, and the coding block is input. The evaluation value g _k is _obtained .

Then, in step 53, the evaluation value g _k is compared with a predetermined threshold value. If the evaluation value does not exceed the threshold value, a difference signal block is obtained from the coding block and the prediction signal block (step 54), the difference signal block is DCT-transformed (step 55), and the difference signal block is quantized with the first quantization width. (Step 56). Finally, the quantized coefficient and other information are variable length or fixed length coded (step 61). If the evaluation value g _k exceeds a predetermined threshold value, the first quantization width determined in step 51 is multiplied by a predetermined coefficient to form a second quantization width (step 57). The predetermined coefficient may be a coefficient smaller than 1 or a coefficient larger than 1. A difference signal block is obtained from the encoded block and the prediction signal block (step 58), the difference signal block is DCT-transformed (step 59), and quantized with the second quantization width (step 60). Finally, encode the quantized coefficient and other information (step 61)
To do.

Although not shown in FIG. 5, a case may be considered in which a quantized transform coefficient is inversely quantized, inverse DCT transform is performed, and a prediction signal is added to a reproduced difference signal to restore a coded block into a reproduced block. To be This is because, in the case of the interframe predictive coding method, it is used as a reference image for coding the next image.

The fourth embodiment of the image coding method of the present invention will be described below. This embodiment is basically the same as the third embodiment. Only the parts different from the third embodiment will be described below. In the third embodiment, the evaluation value is obtained from the predicted signal and the original image of the predicted signal. That is, the prediction signal is input to the terminal 9 and the original signal of the prediction signal is input to the terminal 10 in FIG. However, in the fourth embodiment, the prediction signal is input to the terminal 9 of FIG.
Enter the coded block in. That is, the evaluation value is obtained from the prediction signal block and the coding block. Next, the evaluation value is compared with a predetermined threshold value, and the subsequent processing is the same as in the third embodiment.

Although the first to fourth embodiments have been described above, in the second to fourth embodiments, when the evaluation value exceeds a predetermined threshold value, the coding block is DCT-transformed instead of the differential signal block. Then, the coefficient can be quantized. This is because energy is concentrated on a specific coefficient in the conversion region in the flat portion. For example D
In the case of CT conversion, the energy of the flat part concentrates on the DC component and its periphery. In such a case, it is more efficient to encode only the DC component of the DCT coefficient than the difference signal.

[0052]

As is apparent from the above description, the image coding method of the present invention uses the human visual characteristics to perform weighting so as to enhance the distortion in the flat region, and to distort the distortion in the high spatial frequency region. Distortions that are easily detected visually are detected by weighting so that the detected area is further detected.Furthermore, the detected area is quantized more finely, or the original signal is encoded instead of the differential signal. Thereby, the distortion appearing in the hidden flat area can be significantly reduced as compared with the conventional method.

[Brief description of drawings]

FIG. 1 is a flowchart for explaining a first embodiment of an image coding method according to the present invention.

FIG. 2 is a pattern diagram showing that a hidden background (diamond) gradually appears.

FIG. 3 is a block diagram for obtaining an evaluation value in the embodiment of the present invention.

FIG. 4 is a flowchart for explaining a third embodiment of the image coding method according to the present invention.

FIG. 5 is a flowchart for explaining a fourth embodiment of the image encoding method of the present invention.

[Explanation of symbols]

 11 error amount calculator 12 fluctuation amount calculator 13 evaluation value calculator

Claims (12)

[Claims] 1. An original image to be encoded is input, the original image is divided into a plurality of encoding regions, the encoding region is converted into an encoding conversion region by a predetermined conversion method, and the encoding is performed. The coefficient of the coded transform region is quantized by the first quantization width, dequantized, and inversely transformed to restore the coding region to the playback region, and the error amount ε _k is calculated from the playback region and the coding region. determined, the determined variation amount alpha _k of the encoding region, and increases with increasing the epsilon _k, the evaluation function f (ε _{k, α k)} as to decrease with increase of the alpha _k obtains the evaluation value g _k with When the evaluation value g _k exceeds a predetermined threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width, and the coefficient of the coding conversion area is set to the second quantization width. An image coding method characterized by quantizing with a quantization width. 2. An original image to be encoded is input, the original image is divided into a plurality of encoding areas, the encoding area is converted into an encoding conversion area by a predetermined conversion method, and the code is converted. The coefficient of the quantized transform region is quantized with a first quantization width and inversely quantized to form a quantized transform region, and an error amount ε _k is obtained from the quantized transform region and the coding transform region. of seeking variation alpha _k of region, and increases with increasing the epsilon _k, the evaluation function that decreases with increasing _{α k f (ε k, α} k) obtains the evaluation value g _k in said evaluation value When g _k exceeds a predetermined threshold value, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width, and the coefficient of the coding conversion region is set to the second quantization width. An image coding method characterized by quantizing. 3. When the evaluation value g _m (m <k) of the past coding area instead of the evaluation value g _k of the current coding area exceeds a predetermined threshold value, the first quantization width is set to a predetermined value. A coefficient is multiplied to form a second quantization width, the current coding area is converted into a coding conversion area by a predetermined conversion method, and the coefficient of the coding conversion area is converted into the second quantization width. The image coding method according to claim 1 or 2, wherein the image coding method is quantization. 4. An original image to be encoded and a reference image are input, the original image is divided into a plurality of encoding regions, and a prediction signal region of the encoding region is detected from the reference image. , A difference signal region is obtained from the coding region and the prediction signal region, the difference signal region is converted into a difference signal conversion region by a predetermined conversion method, and the coefficient of the difference signal conversion region is set to a first value.
Quantize with a quantization width of, dequantize, inverse transform to reproduce the difference signal region, add the predicted signal region data to the reproduced difference signal region data, the encoding region into a reproduction region. Restoration, the error amount ε _k is obtained from the reproduction area and the coding area, the variation amount α _k of the coding area is calculated, and increases with the increase of ε _k and decreases with the increase of the α _k. Do the evaluation function f (ε _k, α _k) obtains the evaluation value g _k with, if the evaluation value g _k exceeds a predetermined threshold value, the second quantum is multiplied by a predetermined coefficient to the first quantization width An image coding method, comprising forming a quantization width, and quantizing the coefficient of the differential signal conversion region with the second quantization width. 5. An original image to be encoded and a reference image are input, the original image is divided into a plurality of encoding regions, and a prediction signal region of the encoding region is detected from the reference image. , A difference signal region is obtained from the coding region and the prediction signal region, the difference signal region is converted into a difference signal conversion region by a predetermined conversion method, and the coefficient of the difference signal conversion region is set to a first value.
Quantized with a quantization width of, and inversely quantized to form a quantized difference signal conversion region, and an error amount ε _k is obtained from the quantized difference signal conversion region and the difference signal conversion region, and the variation of the coding region determine the amount alpha _k, increases with the epsilon _k increase, the alpha _k evaluation function f (ε _{k, α k)} as decreases with increasing determined evaluation values g _k in the evaluation value g _k is predetermined When the threshold value is exceeded, the first quantization width is multiplied by a predetermined coefficient to form a second quantization width, and the coefficient in the difference signal conversion region is quantized with the second quantization width. An image encoding method characterized by: 6. When the evaluation value g _m (m <k) of the past coding area exceeds the predetermined threshold value instead of the evaluation value g _k of the current coding area, the first quantization width is set to a predetermined value. 5. The method according to claim 4, wherein the coefficient is multiplied to form a second quantization width, and the coefficient in the difference signal transform area of the current coding area is quantized with the second quantization width. 5. The image coding method according to item 5. 7. An original image to be encoded, a reference image, and an original image of the reference image are input, the original image to be encoded is divided into a plurality of encoding regions, and the reference image is divided. From, the prediction signal region of the coding region is detected, and the error amount ε _k is calculated from the pixel of the prediction signal region and the pixel value at the same position as the pixel of the prediction signal region in the original image of the reference image. determined, the determined variation amount alpha _k of the encoding region, and increases with increasing the epsilon _k, the evaluation function f (ε _{k, α k)} as to decrease with increase of the alpha _k obtains the evaluation value g _k with If the evaluation value g _k does not exceed a predetermined threshold value, a difference signal area is obtained from the coding area and the prediction signal area,
When the difference signal region is converted into a difference signal conversion region by a predetermined conversion method, the coefficient of the difference signal conversion region is quantized with a first quantization width, and the evaluation value g _k exceeds the predetermined threshold value, An image code characterized by multiplying the first quantization width by a predetermined coefficient to form a second quantization width, and quantizing the coefficient of the differential signal conversion region with the second quantization width. Method. 8. An original image to be encoded and a reference image are input, the original image is divided into a plurality of encoding regions, and a prediction signal region of the encoding region is detected from the reference image. , An error amount ε _k is obtained from the coding region and the prediction signal region, a variation amount α _k of the coding region is obtained, and increases as the ε _k increases, and decreases as the α _k increases. the evaluation function f (ε _k, α _k) obtains the evaluation value g _k with, if the evaluation value g _k does not exceed the predetermined threshold value, determines a difference signal area from said encoding region and the predicted signal region,
When the difference signal region is converted into a difference signal conversion region by a predetermined conversion method, the coefficient of the difference signal conversion region is quantized with a first quantization width, and the evaluation value g _k exceeds the predetermined threshold value, An image code characterized by multiplying the first quantization width by a predetermined coefficient to form a second quantization width, and quantizing the coefficient of the differential signal conversion region with the second quantization width. Method. 9. The method according to claim 1, wherein when the evaluation value exceeds a predetermined threshold value, the first quantization width is multiplied by a coefficient smaller than 1 to form the second quantization width. The image coding method according to any one of 1. 10. When the evaluation value exceeds a predetermined threshold value, the coding area is converted into a coding conversion area by a predetermined conversion method instead of the difference signal area, and the coefficient of the coding conversion area is quantized. The image coding method according to any one of claims 4 to 8, wherein: 11. The image coding method according to claim 1, wherein the evaluation function f (ε _k , α _k ) is proportional to ε _k / α _k . 12. The image coding method according to claim 1, wherein the variation amount α _k is normalized by an average value of the variation amounts of all the reference areas.