JPH0970043A - Image coder, image coding method and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce a data quantity obtained as a result of image coding. SOLUTION: A texture area separate circuit 36 detects a texture area being a closed area where a degree of a change in a picture element is small. Then a characteristic amount calculation circuit 40 calculates an average value (hereinafter called activity) of absolute values of orthogonal transformation coefficients in each texture area, a quantization map generating circuit 41 decides the quantization step size for the orthogonal transformation coefficient in each texture area based on the activity. Then a quantization device 42 conducts quantization based on the quantization step size.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding device, an image encoding method and a recording medium. In particular, for example, by extracting a closed region in which the degree of change in image data is not large and by encoding the image data in the closed region in accordance with the feature amount, the encoded data The present invention relates to an image encoding device, an image encoding method, and a recording medium capable of reducing the amount.

[0002]

2. Description of the Related Art For example, when an image is transmitted or recorded on a recording medium, the image data has a high efficiency code in order to reduce the redundancy by utilizing the high correlation of the image. Be converted. As a conventional high-efficiency encoding method of image data, for example, predictive encoding, which handles an image on a pixel-by-pixel basis, or Discrete Cosine Tran
sform) or orthogonal transform coding using wavelet transform (one of subband coding).

A typical method of predictive coding is an intra-frame DPCM (Differential Pulse Code Modulatio).
According to this method, the difference between the original pixel and its neighboring pixels, which have already been encoded and locally decoded, is quantized and encoded. The predictive coding method is effective when the required compression rate is not so high as about 1/2 to 1/4,
It is difficult to further improve the compression rate.

On the other hand, according to the orthogonal transform coding, 1/10
The above high compression rate can be obtained, and with respect to, for example, DCT, which is a representative of orthogonal transform coding, a high-speed algorithm has been developed, hardware can be easily implemented, and the international standard (JPEG (JPEG Joint Photograph
ic Coding Experts Group), MPEG (Moving Pictur)
Since it has also been adopted by the e Experts Group)), at present, this DCT-based coding method is generally used.

In the image coding method using DCT, the entire image is divided into macroblocks, each macroblock is DCTed, and the resulting DCT coefficient is quantized. Then, when the DCT coefficient is quantized, the characteristic that the power of the low frequency component of the image is extremely large is used to reduce the quantization step size of the low frequency component of the DCT coefficient and to quantize the high frequency component. By increasing the step size, the information amount of the entire image is compressed.

However, in this case, there is a problem that block distortion and mosquito noise are generated by performing the quantization, and in particular, the block distortion due to the processing in units of macro blocks is encoded. It becomes remarkable when the rate is low. For this reason, when performing image encoding at an ultra-low bit rate, the above-described encoding method of performing DCT and quantizing an image obtained by dividing an image into macroblocks is not suitable.

Therefore, as a method for the purpose of image transmission and recording at an ultra-low bit rate, a method of preserving the contour line portion in the original image is proposed. According to this method, considering that human visual characteristics are particularly sensitive to the contour line of an object, by preserving the contour line portion in the original image, it is visually superior even at a low bit rate. Resiliency is realized.

In the image coding method in which the contour line portion in the original image is stored with emphasis, a range (pixels forming such a contour is defined as the contour line portion of the object in the image. Hereinafter, appropriately referred to as an edge area) is efficiently extracted, and the edge area and, for example, the pixel values of pixels adjacent to the edge area and the encoded texture image are output as encoded data.

That is, the pixel values of the edge area and the pixels adjacent to the edge area (for example, the pixels located in the direction normal to the edge strength of the edge area and adjacent to the edge area) According to this, it is possible to restore the global characteristic portion such as the contour or boundary of the object in the original image. Here, an image obtained by restoring the vicinity of the contour line of the object based on this edge region and, for example, some pixel values forming the original image, such as the pixel values of the pixels adjacent thereto, It is called a reconstructed image as appropriate.

However, in the reconstructed image, the degree of change (gradation) of the pixel value such as the pattern of the object (for example, subtle unevenness when there is a lawn or forest) is not so large (or , Small), that is, almost no information about the texture area is included. Therefore, only the pixel values of the edge area and the pixels adjacent thereto are restored on the decoding side in the same manner as the original image (close to the original image). It becomes difficult to obtain an image.

Therefore, in order to obtain a restored image (decoded image) similar to the original image (in order to reproduce fine changes in the original image), as described above, In addition to the pixel values of the edge region and the pixels adjacent thereto, a texture image, that is, a coded difference image between the original image and the reconstructed image is transmitted as coded data.

As a result, on the decoding side, the reconstructed image is obtained from the pixel values of the edge region and the pixels adjacent thereto, the texture image is decoded, and the texture image is added to the reconstructed image, so that the original image is obtained. It is possible to obtain a restored image of.

Here, FIG. 17 shows an example of the configuration of a conventional image coding apparatus (texture coding apparatus) for coding a texture image. Edge image buffer 3
An edge image is stored in 1. An edge image is a pixel value of a pixel forming one frame image,
For example, it is a binary image in which 0 is set for the edge region and 1 is set for other portions.

The edge image stored in the edge image buffer 31 is supplied to the edge area expansion circuit 34. The edge area expansion circuit 34 expands the edge image (edge area in the edge image). That is, the edge area expansion circuit 34 determines, in the edge image, from a pixel having a pixel value of 0 to a predetermined fixed range (for example, a range of 3 × 3 pixels centering on a pixel having a pixel value of 0). The pixel value of the pixel located in () is set to 0.

Here, the reason for expanding the edge image in this way is as follows. That is, in the pixel value of the pixel located in the vicinity of the edge area of the texture image output by the calculator 35, which will be described later, there may be a pixel value having a very large error from the original image. Therefore, in order to remove such a pixel from the texture image by masking it in the texture region separation circuit 36 described later,
Dilate the edge image. That is, the inflated edge image (hereinafter, appropriately referred to as an inflated edge image) is used in the texture region separation circuit 36 to mask pixels having a pixel value including a large error from the texture image.

The expanded edge image generated by the edge area expansion circuit 34 is supplied to the texture area separation circuit 36.

On the other hand, the reconstructed image frame buffer 32 stores the reconstructed image, and the original image frame buffer 33 stores the original image. The calculator 35 calculates the difference between the original image stored in the original image frame buffer 33 and the reconstructed image stored in the reconstructed image frame buffer 32, and the difference image, that is, the texture image, becomes the texture region. It is output to the separation circuit 36.

In the texture area separation circuit 36, the texture image from the calculator 35 is converted into the edge area expansion circuit 34.
By being masked with the dilated edge image from
To the pixel value of each pixel forming the texture image, the pixel value of the pixel forming the expanded edge image corresponding to the pixel (0
Or by multiplying by 1)), the dilated edge image is removed from the texture image. As a result, of the pixels forming the texture image, the pixels forming the expanded edge image having a pixel value of 0, that is, the pixels corresponding to the pixels forming the expanded edge are set to 0, and The pixel values of the texture images other than those are stored as they are.

FIG. 18 shows a texture image obtained by the texture area separation circuit 36. In the figure, the shaded portion indicates a portion that is not masked by the expanded edge image (hereinafter, appropriately referred to as a texture area), and the solid line surrounding it indicates the contour line of the texture area. (A contour line surrounding the outer periphery of the texture area) is shown. In addition, the white portion indicates a region where the pixel value of the pixels forming the dilated edge image is 0 (hereinafter, appropriately referred to as dilated edge region), and the dotted line surrounding the region is
The outline of the expanded edge area (the outline surrounding the outer periphery of the expanded edge area) is shown. Although the image frame is shown by a solid line, it is treated as a contour line of the expanded edge area.

As shown in the figure, the dilated edge region is a closed region, and the texture region separation circuit 36 obtains a texture region by masking the texture image with the dilated edge image obtained by dilating the edge image. Is also a closed area.

As described above, the texture image composed of the expanded edge area and the texture area, which are closed areas, is supplied to the macroblock generation circuit 51, where
It is divided into macroblocks (for example, blocks of 8 × 8 pixels) and output to the DCT circuit 52. DCT circuit 5
In 2, the macro block supplied from the macro block generation circuit 51 is DCT'ed, and thereby the DCT coefficient is obtained. The DCT coefficient is supplied to the quantizer 42, which quantizes the DCT coefficient with a predetermined quantization step size. The quantized data obtained as a result is supplied to a VLC (Variable Length Coding) circuit 46, where it is variable length coded, and then supplied to an output buffer 47 for storage. The texture image is encoded as described above.

[0022]

By the way, in the image coding apparatus shown in FIG. 17, since the texture image is divided into macroblocks in the macroblock generating circuit 51, in principle, block distortion still occurs. .
However, in this case, not the original image itself, but a texture image with a small degree of change in pixel value is coded,
Block distortion occurring in the decoded image can be reduced as compared with the case of encoding the original image itself.

However, even when the texture image is coded, the amount of data obtained as a result may increase (for example, when the checkered texture image is coded). In this case, FIG. Although not shown in the figure, the output data amount is fed back from the output buffer 47 to the quantizer 42, and the quantization step size in the quantizer 42 is increased in order to reduce the output data amount.

In this case, the quantizer 42 may not be able to allocate sufficient bits to the texture image, and as a result, block distortion may be significant.
Therefore, it is desired to further reduce the output data amount.

Furthermore, when encoding a texture image, the individual texture regions present in the texture image are generally independent, ie, for example, the surface portions of different objects, so that each texture region is usually The degree of change in the pixel value in the area varies depending on the texture area. In the past, however, the quantization step size of the quantizer 42 was set in units of texture images, so that the number of bits was too large. A large number of bits may be allocated to a texture area that does not need a texture, and vice versa, a texture area that requires a large number of bits may not have a large number of bits, and thus efficient coding may be performed. There was a problem that was not done.

The present invention has been made in view of such circumstances, and enables an image to be efficiently coded and further reduce the amount of data obtained as a result. To do.

[0027]

An image coding apparatus according to the present invention comprises a closed region extracting means for extracting a predetermined closed region from an image, and a feature amount detecting means for detecting a feature amount of image data in the closed region. , And encoding means for encoding the image data in the closed region in accordance with the characteristic amount thereof.

The image coding method of the present invention is characterized in that a predetermined closed area is extracted from an image and the image data in the closed area is coded in accordance with the feature amount.

The recording medium of the present invention is characterized in that a predetermined closed area is extracted from an image, and the image data in the closed area is encoded in accordance with the characteristic amount of the image data, and the data is recorded.

In the image coding apparatus of the present invention, the closed area extracting means extracts a predetermined closed area from the image, and the feature amount detecting means detects the feature amount of the image data in the closed area,
The encoding means is adapted to encode the image data in the closed area in correspondence with the feature amount.

In the image coding method of the present invention, a predetermined closed area is extracted from the image, and the image data in the closed area is
The encoding is performed according to the feature amount.

On the recording medium of the present invention, data in which a predetermined closed region is extracted from an image and the image data in the closed region is encoded in correspondence with the feature amount is recorded.

[0033]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below, but before that, in order to clarify the correspondence between each means of the invention described in the claims and the following embodiments. The features of the present invention are described as follows by adding a corresponding embodiment (however, an example) in parentheses after each means.

That is, the image coding apparatus according to the first aspect of the present invention is a closed area extracting means for extracting a predetermined closed area from an image (for example, the texture separation circuit 36 shown in FIG. 7).
And a characteristic amount detecting means for detecting the characteristic amount of the image data in the closed region (for example, the characteristic amount calculating circuit 40 shown in FIG. 7).
And encoding means for encoding the image data in the closed region in accordance with the feature amount (for example, the orthogonal transformation circuit 38, the small region deletion circuit 39, the quantization map creation circuit 4 shown in FIG. 7).
1, a quantizer 42, a high band elimination circuit 43, a coefficient elimination circuit 44, a coefficient rearrangement circuit 45, a VLC circuit 46, etc.).

An image coding apparatus according to a second aspect of the present invention is a contour line detecting means for detecting a contour line surrounding the closed area extracted by the closed area extracting means (for example, the contour line listing circuit 37 shown in FIG. 7 or the like). ) Is further provided, and the feature amount detecting means is
A feature is that the feature amount of the image data in the area surrounded by the contour line is detected, and the encoding means encodes the image data in the region surrounded by the contour line in accordance with the feature amount.

In the image coding device according to the third aspect, the coding means performs orthogonal transformation of the image data and outputs orthogonal transformation coefficients (for example, the orthogonal transformation circuit 38 shown in FIG. 7). , A quantizing means for quantizing the orthogonal transform coefficient with a predetermined quantization step size (for example, the quantizer 42 shown in FIG. 7), and a predetermined quantization when quantizing the orthogonal transform coefficient in the closed region. Quantization step size determining means for determining the step size corresponding to the feature amount (for example,
And a quantization map generating circuit 41 shown in FIG. 7).

In the image coding apparatus according to the fifth aspect, the coding means performs variable length coding on the orthogonal transform coefficient quantized by the quantizing means (for example, VLC shown in FIG. 7). Circuit 46) and the orthogonal transform coefficients quantized by the quantizing means are segmented for each frequency band and supplied to the variable length coding means (for example, the coefficient rearranging circuit 45 shown in FIG. 7). Etc.) and are further included.

In the image coding apparatus according to the sixth aspect, the coding means performs variable length coding on the orthogonal transform coefficient quantized by the quantizing means (for example, VLC shown in FIG. 7). Circuit 46) and the orthogonal transform coefficients quantized by the quantizing means are segmented for each closed region,
It is characterized by further comprising segmenting means (for example, the coefficient rearranging circuit 45 shown in FIG. 7) supplied to the variable length coding means.

In the image coding apparatus according to the seventh aspect, the coding means removes the high frequency band component removing means (for example, a component higher than a predetermined high frequency band) from the orthogonal transform coefficient (for example,
The high band elimination circuit 43 shown in FIG. 7) is further provided.

In the image coding apparatus according to the eighth aspect, the coding means removes the orthogonal transform coefficients in the closed area having an area equal to or smaller than a predetermined value from the small area closed area removing means (for example, the figure The small area deletion circuit 39 shown in FIG. 7) is further included.

In the image coding apparatus according to the ninth aspect, the coding means removes small absolute value coefficient removal means (for example, in FIG. 7) that the absolute value of the orthogonal transformation coefficient is less than a predetermined value. (A coefficient deleting circuit 44, etc. shown) is further included.

Of course, this description does not mean that each means is limited to the above.

FIG. 1 shows the configuration of an embodiment of an image coding apparatus to which the present invention is applied. In this image coding apparatus, 3C is a typical method of image coding at an ultra-low bit rate, which emphasizes preservation of the contour line of an object.
An image is encoded according to C (3 Component Coding). Note that 3CC is a method of dividing an original image into an edge image, local luminance (pixel values obtained by sampling the original image at a predetermined number of pixels, for example, in a grid pattern), and a texture image to perform encoding. is there.

The original image (image data) is supplied to the local luminance detection coding circuit 1, the edge detection coding circuit 2, and the arithmetic unit 4. The local luminance detection encoding circuit 1 detects local luminance (Local Luminance) as global information including luminance information from the original image, encodes it, and supplies it to the reconstruction circuit 3. Further, in the edge detection coding circuit 2, after the edge image is detected from the original image and coded, the reconstruction circuit 3 is also used.
Is supplied to.

The reconstruction circuit 3 decodes the local luminance from the local luminance detection coding circuit 1 and the edge image from the edge detection coding circuit 2 and generates a reconstructed image based on the decoding result. . This reconstructed image is supplied to the arithmetic unit 4. The arithmetic unit 4 obtains the difference between the original image and the reconstructed image from the reconstructing circuit 3, and thereby a texture image is generated and supplied to the entropy encoding circuit 5. In the entropy coding circuit 5, first, a texture area that is a closed area is extracted from the texture image, and the feature amount of the image data in each texture area is detected. Then, the image data in each texture region is encoded and output corresponding to the feature amount.

As described above, after the local luminance, the edge image, and the texture image are obtained from the original image and encoded, these data (hereinafter, appropriately referred to as 3CC data) are recorded on, for example, an optical disk or a magneto-optical disk. It is supplied and recorded on a recording medium 6 which is a disk, a magnetic disk, a tape or the like. Alternatively, it is transmitted via a predetermined transmission path.

FIG. 2 shows a configuration example of the edge detection coding circuit 2 of FIG. The original image (image data) (digital data) is supplied to the Sobel filters 11 and 12 and the filter 16. The Sobel filter 11 or 12 is, for example, as shown in FIG.
3 × 3 having tap coefficients as shown in FIG.
Value of the horizontal or vertical edge strength of the original image by multiplying and summing the tap coefficient and the original image data with the digital filter of Strength).

The horizontal edge strength or the vertical edge strength is
It is supplied to the computing unit 13 or 14, respectively. Calculator 1
In 3 or 14, respectively, the horizontal edge strength or the vertical edge strength is squared to obtain the horizontal edge strength power or the vertical edge strength power, and both are supplied to the calculator 15. The calculator 15 adds the horizontal edge strength power and the vertical edge strength power to calculate a value (hereinafter, simply referred to as edge strength) representing the edge strength of each pixel forming the original image. . This edge strength is supplied to the comparator 19.

The comparator 19 includes a threshold value determining circuit 1 which will be described later.
The threshold value T supplied from 8 and the edge strength of each pixel from the calculator 15 are compared, and the pixel whose edge strength is the threshold value T or more is detected. Then, the pixel value of the pixel is set to 0, and the pixel values of the other pixels are set to 1, so that the edge image is generated.

Here, the edge strength of each pixel obtained by the calculator 15 becomes large for a pixel whose degree of change in pixel value is locally large. That is, the edge strength is not limited to pixels having a large degree of change in pixel value, such as the contour of an object or a boundary between objects, and the pattern of the object, even when viewed locally or from a global perspective. In general, the degree of change in pixel value is small, but locally the degree of change in pixel value is large. Therefore, if the comparator 19 compares the edge strength of each pixel with a constant threshold value T, not only the contour of the object or the boundary between the objects but also the pattern portion of the object can be detected. Becomes

On the other hand, the edge image is an image drawn only by the contours of the object or boundaries between the objects. Therefore, an image drawn from pixels whose edge strength is equal to or more than a certain threshold T is the contour of the object or the object. It does not become an edge image because it includes not only boundaries between objects but also patterns of objects.

Therefore, in the edge detection coding circuit 2, the contours of the objects and the boundaries between the objects and the pattern of the objects are, for example,
An index value indicating a local change characteristic of the pixel value (gradation value),
It is designed to make a distinction by using an index value showing a global change characteristic. That is, the index value representing the local change characteristic of the pixel of interest includes, for example, edge strength, which is output from the calculator 15 as described above. On the other hand, the index value representing the global change characteristic of the target pixel includes, for example, a spatial frequency distribution around the target pixel, which is obtained by the mask area setting circuit 17 based on the output of the filter 16. Has been done.

That is, the filter 16 is a two-dimensional HPF.
(High Pass Filter), LPF (Low Pass Filter), or BPF (Band Pass Filger). The original image data is filtered to change the pixel value of each pixel forming the original image. A value (hereinafter, appropriately referred to as a pixel value change amount) corresponding to is calculated and output.

As the filter 16, for example, a 3 × 3 Sobel filter having tap coefficients as shown in FIG. 3 can be used. In this case, the filter 16 is HPF. However, as the filter 16 (the same applies to the Sobel filters 11 and 12), a filter having a tap coefficient other than the tap coefficient shown in FIG. 3 can be used, and the number of taps is not limited to 3 × 3. For example, it may be 5 × 5 or 7 × 7.

The amount of change in pixel value obtained by the filter 16 is supplied to the mask area setting circuit 17. The mask area setting circuit 17 is supplied with the horizontal edge strength or the vertical edge strength from the Sobel filter 11 or 12, respectively, in addition to the amount of change in the pixel value. The area (window area) is set.

That is, the mask area setting circuit 17, as shown in FIG. 4, for example, shows the direction of the normal to the edge strength of the pixel of interest (the portion marked with diagonal lines in the upper left direction and the upper right direction in the figure) ( (The direction in which the rate of change in edge strength is greatest)
Is determined based on the horizontal edge strength or the vertical edge strength from the Sobel filter 11 or 12, respectively.
Then, the mask area setting circuit 17 sets, as a mask area, a one-dimensional area that is present within a predetermined distance from the pixel of interest and that is composed of pixels on the edge normal direction. Therefore, for example, when an edge exists in the original image as shown in black in FIG. 4, a region shaded in the upper left direction as shown in FIG. A mask area) and a hatched area in the upper right direction (hereinafter appropriately referred to as a second mask area) are set as the mask area.

Here, the reason why the mask area is set as described above is to obtain the change in the pixel value that greatly changes at the edge from the relationship of the pixels existing on both sides of the edge.

In the embodiment of FIG. 4, from the pixel of interest,
A one-dimensional area that is present within a distance corresponding to 5 pixels and that is composed of pixels on the normal direction of the edge strength is set as the mask area.

Further, in the embodiment of FIG. 4, a one-dimensional area is set as a mask area, but in addition, for example, it is composed of pixels existing around the edge strength normal line of the pixel of interest. It is also possible to set a two-dimensional area to be used as a mask area.

After that, the mask area setting circuit 17 obtains the average value M1 or M2 of the pixel value change amounts of the pixels existing in the first and second mask areas, respectively, and outputs it to the threshold value determining circuit 18. . Where the first or second
The average value M1 or M2 of the amount of change in the pixel value of each pixel present in each of the mask areas corresponds to the spatial frequency distribution in the global range centered on the pixel of interest. That is, when the degree of change in the pixel value of the pixel of interest is large in general, the average values M1 and M2 are both large.

Therefore, the average values M1 and M2 are basically large in the contours of the objects and boundaries between the objects, but are not large in the pattern of the objects. That is, according to the average values M1 and M2, it is possible to distinguish the contour of the object or the boundary between the objects and the pattern of the object. Therefore, in the threshold value setting circuit 18, the threshold value T to be applied to the target pixel by the comparator 19 is determined corresponding to the average values M1 and M2, and is output to the comparator 19.

That is, the threshold setting circuit 18 receives the average values M1 and M corresponding to the target pixel from the mask area setting circuit 17.
When 2 is received, a function (threshold function) T = F (M1, M2) having the average values M1 and M2 as arguments is calculated, and the value obtained as a result is output to the comparator 19 as a threshold value T.

The function F (M1, M2) is determined, for example, by simulation so that the optimum threshold value T for distinguishing the contour of the object or the boundary between the objects and the pattern of the object can be obtained. To be done. Here, FIG.
An example of a curved surface corresponding to the function T = F (M1, M2) in a three-dimensional space having 1, M2 and T as x, y and z axes is shown. This curved surface is expressed by the following equation, for example.

{M1- (aT-b)} {M2- (aT-b)} = c where a, b, and c are predetermined constants.

As described above, the comparator 19 detects pixels whose edge strength is equal to or higher than the threshold value T from the threshold value determining circuit 18. Therefore, the comparator 19 detects only the contour of the object and the boundary between the objects, and outputs the edge image.

The edge image output from the comparator 19 is
In an encoding circuit (not shown), for example, it is encoded into a code that can obtain the coordinates of a pixel having a pixel value of 0 (henceforth, a pixel forming an edge) and output.

In the edge detection coding circuit 2, the edge image is obtained by the edge separation method. That is, in the edge detection coding circuit 2, in order to extract the contours between the objects and the boundaries between the objects existing in the area where the luminance change is small, both the luminance signal and the color difference signal are subjected to the edge image processing as described above. Are required, and they are designed to be synthesized.

Next, FIG. 6 shows an example of the configuration of an image decoding apparatus for decoding the data encoded by the image encoding apparatus of FIG. The 3CC data recorded on the recording medium 6 (FIG. 1) is reproduced from there, and the local luminance, the edge image, or the texture image in the reproduced data is
It is supplied to the local luminance decoding circuit 21, the edge image decoding circuit 22, or the texture image decoding circuit 23, respectively. The local luminance decoding circuit 21 decodes the encoded local luminance and outputs the decoded local luminance to the edge image decoding circuit 22. The edge image decoding circuit 22 decodes the encoded edge image,
A reconstructed image is generated based on the decoded edge image and the local luminance from the local luminance decoding circuit 21. This reconstructed image is supplied to the arithmetic unit 24.

The arithmetic unit 24 includes the edge image decoding circuit 2
2, the texture image is supplied from the texture image decoding circuit 23 as well as the reconstructed image. That is, the texture image decoding circuit 23 decodes the encoded texture image and supplies the decoded texture image to the calculator 24. The arithmetic unit 24 decodes the original image by adding the reconstructed image and the texture image.

That is, the reconstructed image composed of the edge image and the local luminance contains information about global features such as outlines and boundaries of the object in the original image, but changes in pixel value. Since the information about the small portion is hardly included, the reconstructed image cannot reproduce, for example, subtle unevenness of a lawn or forest. Therefore, in the calculator 24, a texture image, which is information indicating a minute change in pixel value such as a pattern portion of an object, is added to the reconstructed image to reproduce subtle unevenness such as lawn or forest. A decoded image can be obtained.

Next, FIG. 7 shows an example of the structure of a texture coding apparatus for coding a texture image. In addition,
In the figure, the parts corresponding to the case in FIG.
The same reference numerals are given, and the description thereof will be omitted below as appropriate. The texture coding apparatus corresponds to the entropy coding circuit 5 in FIG. 1 (however, the computing unit 35 in FIG. 7 corresponds to the computing unit 4 in FIG. 1).

The edge area expansion circuit 34, the arithmetic unit 35, and the texture area separation circuit 36 perform the processing described with reference to FIG. 17, and as a result, the texture area separation circuit 36 produces the same processing as shown in FIG. As shown, the texture image output from the calculator 35 is masked with the expanded edge image output from the edge area expansion circuit 34, and the masked image is output. The texture image output from the texture region separation circuit 36 is supplied to the contour line listing circuit 37 and the orthogonal transformation circuit 38.

The contour line listing circuit 37 detects a contour line surrounding the texture area forming the texture image from the texture area separation circuit 36. That is, the contour line listing circuit 37 first sets the pixel value of a pixel having a pixel value other than 0 to 1. As a result, as shown in FIG. 9, the pixel values of the pixels forming the texture area or the expanded edge area forming the texture image are 1 or 0, respectively.
Becomes

Further, the contour line list forming circuit 37 detects a portion where the pixel value changes from 0 to 1 and a portion where the pixel value changes from 1 to 0 among the pixels forming the texture image, and among the portions, A pixel whose pixel value is 1 is detected. Thus, for example, from the texture image shown in FIG. 8, a contour line (solid line portion) surrounding the outer circumference of the texture region shown in FIG. 8 and a contour line (dotted line portion) surrounding the outer circumference of the expanded edge region (this contour line Can also be regarded as a contour line surrounding the inner circumference of the texture area).

The contour line list forming circuit 37 performs a list forming process after detecting the contour lines. That is, the contour line listing circuit 37 sets a certain contour pixel (pixel on the contour line) as a target point, and stores, for example, the coordinates of the pixel as the target point.
Then, the contour pixel adjacent to the pixel as the attention point, which is not yet the attention point, is newly set as the attention point, and its coordinate value is stored. Repeat until. After that, the contour line list forming circuit 37 sets any contour pixel which is not listed (coordinates are not stored) (thus, not yet set as a target pixel) as a target point, and the above-described processing is performed in the list. Repeat until there are no unconstituted contour pixels.
As a result, the contour line listing circuit 37 constructs a list (hereinafter, appropriately referred to as a contour line list) in which the coordinates of pixels forming all the contour lines existing in the texture image are listed.

According to the above list forming processing, when attention is paid to a certain contour line, a certain contour pixel on the contour line is used as a starting point and adjacent contour pixels are traced. The direction in which the contour pixels are traced (hereinafter, appropriately referred to as the list direction) is set to be as constant as possible. By doing so, it is possible to prevent the contour pixel once set as the point of interest from being set as the point of interest again, so that an incorrect contour line is not detected.

Further, when tracing the contour pixels, the two adjacent pixels to the point of interest in the direction orthogonal to the list direction.
Each of the two pixels always belongs to a certain area. That is, since the contour line is a line that divides the texture region that is a closed region and the dilated edge region, if the contour line is traced so that it can be written in a single stroke, at a certain point, the list If there is a texture area (expanded edge area) or expanded edge area (texture area) on the right or left side with respect to the list direction, respectively, at other times, on the right or left side with respect to the list direction, respectively. There will be a texture area (expanded edge area) or an expanded edge area (texture area). Therefore, with respect to the point of interest, pixels adjacent to the list direction in the right or left direction are
By always tracing the contour pixels so as to belong to a certain area, it is possible to prevent the detection of an incorrect contour line.

The contour line list obtained as described above includes the small area deletion circuit 39 and the feature amount calculation circuit 4 which will be described later.
0, the quantization map creating circuit 41, the coefficient deleting circuit 44, and the coefficient rearranging circuit 45. In these blocks, the texture area included in the texture image is recognized in correspondence with the contour line list, and the processing is performed for each recognized texture area.

The small area deleting circuit 39, the characteristic amount calculating circuit 40, the quantization map creating circuit 41, and the coefficient deleting circuit 4
4, and the coefficient rearranging circuit 45 recognizes the texture area as follows, for example.

That is, the contour line list forming circuit 37 is adapted to refer to the pixel values of the pixels on the right and left sides in the list forming direction when tracing the contour pixels forming the contour line. Thus, the area on the right side or the left side in the listing direction is recognized as a texture area or a dilated edge area. Then, the contour line listing circuit 37 outputs the information (hereinafter, appropriately referred to as region information) in which the region on the right side or the left side in the listing direction is recognized as described above together with the contour line list. ing.

Then, the small area deleting circuit 39, the feature amount calculating circuit 40, the quantization map creating circuit 41, and the coefficient deleting circuit 4
4, and the coefficient rearranging circuit 45 first reconstructs the contour line based on the contour line list, and further, based on the region information, among the regions surrounded by the reconstructed contour line,
Arbitrary pixels forming the texture area are detected.
Then, the pixels within a predetermined range centered on that pixel are filled, and further, the pixels within a predetermined range centered on each pixel on the outline of the filled region are filled. The region is configured so that it does not exceed the reconstructed contour line, and when there are no more pixels that can be filled, the region formed by the filled pixels is recognized as a texture region.

On the other hand, in the orthogonal transformation circuit 38, the texture image from the texture area separation circuit 36 is orthogonally transformed. That is, in the present embodiment, in the orthogonal transformation circuit 38,
The entire texture image is wavelet-transformed using, for example, a 5-3 filter, and the orthogonal transformation coefficient obtained as a result is output to the small area deletion circuit 39.

Here, the reason why the wavelet transform is adopted as the orthogonal transform performed in the orthogonal transform circuit 38 is that the position information (coordinates) of the pixels, which is not saved by the DCT, is saved by the wavelet transform. That is, for example, when the texture image shown in FIG. 8 is wavelet transformed using the 5-3 filter, as shown in FIG. 10, for example, the low frequency component (Low) from the upper left to the lower right, This is because it is possible to obtain an array of texture images composed of the middle frequency component (Medium) or the high frequency component (High).

It should be noted that the orthogonal transform circuit 38 may perform the orthogonal transform on each texture region forming the texture image, instead of performing the orthogonal transform on the entire texture image. is there.

In the small area deletion circuit 39, the orthogonal transformation circuit 3
Among the orthogonal transform coefficients from 8, the ones in the texture area whose area is equal to or less than a predetermined value are removed (deleted). That is, in the small area deletion circuit 39, the texture area existing in the texture image is recognized as described above,
For example, the number of pixels forming each texture area is detected. Then, in the small area deletion circuit 39, the orthogonal transformation circuit 3
Of the orthogonal transform coefficients from 8, those in the texture area having 20 or less pixels are set to 0.

An image decoded based on an orthogonal transform coefficient in a texture area having an area of a predetermined value or less, that is, a texture area having a small area is not visually significant, and therefore such an orthogonal transform coefficient is used. Even if set to 0, the image quality of the decoded image is not so badly affected. On the other hand, by setting the orthogonal transform coefficient in the pixels constituting the small area texture region to 0, so-called 0 runs increase in the input of the VLC circuit 46 in the subsequent stage, so the encoded data output from the VLC circuit 46 is increased. The data amount (output data amount) of can be reduced. That is, the amount of output data can be reduced by minimizing the deterioration of the decoded image.

It should be noted that the output data amount of the VLC circuit 46 is fed back to the small area deletion circuit 39, and the small area deletion circuit 39 calculates the orthogonal transform coefficient corresponding to this output data amount. The area of the texture area to be deleted is changed. That is, when the output data amount is small, that is, when the output data amount increases, the small area deletion circuit 39 reduces the area of the texture area to which the orthogonal transform coefficient is deleted, This improves the quality of the decoded image. In addition, the small area deletion circuit 39
When the output data amount is large, that is, when the output data amount needs to be reduced, the area of the texture region where the orthogonal transform coefficient is to be deleted is increased, thereby reducing the output data amount. It is done like this.

The orthogonal transformation coefficient is supplied to the feature amount calculation circuit 40 and the quantizer 42 after the small area deletion circuit 39 has made the value in the small area texture area zero.

The feature amount calculation circuit 40 detects (calculates) the feature amount of the image in each texture region in order to determine the quantization step size in the quantizer 42. That is, the feature amount calculation circuit 40 recognizes the texture region existing in the texture image as described above, and as the feature amount of the image in each texture region, for example, the absolute value of the orthogonal transformation coefficient in each texture region is used. The average value of is calculated. This average value is supplied to the quantization map creation circuit 41.

In order to determine the quantization step size, the feature amount of the image in the texture region calculated by the feature amount calculating circuit 40 is limited to the average absolute value of the orthogonal transform coefficients in the texture region. However, the average value of the squared values of the orthogonal transform coefficients may be used as long as it represents the degree of change of the pixel value in the texture region. Here, hereinafter, as appropriate, what represents the degree of change in pixel value in the texture region, such as the average value of the absolute values of the orthogonal transform coefficients in the texture region,
It is called activity.

In the quantization map creating circuit 41, the quantization step size when quantizing the orthogonal transform coefficient in the texture area is determined in accordance with the activity from the feature quantity calculating circuit 40. That is, the quantization map creating circuit 41 recognizes the texture area existing in the texture image based on the output of the contour line listing circuit 37. Then, as a general rule, the quantization map creation circuit 41 sets a larger or smaller quantization step size for a texture area in which the activity from the feature amount calculation circuit 40 is larger or smaller. In addition,
In the quantization map creation circuit 41, an infinite quantization step size (for example, the quantizer 42 outputs the texture area for which the activity is equal to or lower than a predetermined lower limit value and the texture area whose activity is equal to or higher than a predetermined upper limit value. The quantization step size corresponding to the maximum quantization value) is set.

Specifically, in the feature quantity calculating circuit 40, as an activity, for example, as described above,
When the average value of the absolute values of the orthogonal transform coefficients in the texture region is calculated, the quantization step creating circuit 41 determines each texture according to the correspondence between the activity and the quantization step size as shown in FIG. 11, for example. The quantization step size is set in the area. In the example of FIG. 11, the infinite quantization step size is set for the texture area of 20 or less and the texture area of 70 or more, and the activity is set to the texture area of other values. Has a quantization step size set to a value proportional to its activity.

The feature amount calculation circuit 40 sets a quantization step size for each texture area, and then a map describing the quantization step size corresponding to each pixel forming the texture image (hereinafter, appropriately referred to as a quantization map). ) Is output to the quantizer 42. In the quantizer 42,
According to the quantization step size described in the quantization map from the feature amount calculation circuit 40, the small area deletion circuit 39
The orthogonal transform coefficients from are quantized.

The reason why the quantization step size having a value proportional to the activity of the texture area is set is as follows. That is, in the texture area where the activity is small, the degree of change in the pixel value in the texture area is small. Therefore, in order to reflect such a change in the pixel value in the quantized value, the quantization step size is reduced. It is necessary. On the other hand, in the texture area where the activity is large, the degree of change in the pixel value in the texture area is large, and such a change in the pixel value is reflected in the quantized value even if the quantization step size is increased. This is because that.

Note that it is necessary to set an infinite quantization step size for a texture area whose activity is equal to or lower than a predetermined lower limit value or a texture area whose activity is equal to or higher than a predetermined upper limit value. This is because the change in pixel value is extremely small or large, and therefore it is not necessary to code as a texture area, so that the quantization value output from the quantizer 42 is set to 0.

As described above, for each texture area,
Since the optimum quantization step size is set on the basis of the activity, the quantizer 42 can allocate an appropriate number of bits to each texture region, and as a result, the texture image can be efficiently coded. Can be done on a regular basis. In addition, since the infinite quantization step size is set for the texture area where the activity is equal to or lower than the predetermined lower limit value and the texture area where the activity is equal to or higher than the predetermined upper limit value, the quantized value becomes 0. As a result, 0 runs increase at the input of the VLC circuit 46, and the amount of output data can be reduced.

The quantization map creating circuit 41 has V
The output data amount of the LC circuit 46 is fed back, and in the quantization map creation circuit 41, the upper limit value and the lower limit value of the activity that makes the quantization step size infinity correspond to the output data amount. It is designed to change. That is, when the output data amount is small, that is, when there is no problem even if the output data amount increases, the quantization map creation circuit 41 sets the upper limit value of the activity that makes the quantization step size infinite (see FIG. 11). In the embodiment, it is set to 70) or the lower limit value (which is set to 20 in the embodiment of FIG. 11) is increased or decreased, respectively, thereby improving the image quality of the decoded image. In addition, the quantization map creation circuit 41
When the output data amount is large, that is, when the output data amount needs to be reduced, the upper limit value or the lower limit value of the activity for which the quantization step size is infinite is made smaller or larger, respectively. It is designed to reduce the amount of data.

Further, the quantization map creating circuit 41 has
It is possible to make the quantization step size have a weight corresponding to the frequency band of the orthogonal transform coefficient. Here, FIG. 12 shows an example of weights to be added to the quantization step size corresponding to the frequency band of the orthogonal transform coefficient.

In FIG. 12A, the quantization step size is weighted corresponding to, for example, the visual importance of the frequency band of the orthogonal transform coefficient. That is, now in FIG.
As shown in FIG. 12A, if numbers are given to the frequency bands after the wavelet transform, in the embodiment of FIG. 12A, weights of 1 are assigned to frequency bands 1 to 4 and frequency band 5 is assigned. ^Has a weight of (0.8 + a) / (2 × 2 ^1/2 ),
The frequency band 6 has a weight of (0.7 + a) / (2 × 2 ^1/2 ) and the frequency band 7 has a weight of a / (2 × 2 ^1/2 ).
The frequency band 8 is weighted with 0.8 + a, the frequency band 9 is weighted with 0.7 + a, and the frequency band 10 is weighted with a. However, a is a real number larger than 0 (a is usually a value of 1 or more).

Further, in the embodiment shown in FIG. 12B, a uniform weight a is given to all frequency bands 1 to 10.

Returning to FIG. 7, the quantized value obtained as a result of the quantization in the quantizer 42 is supplied to the high band elimination circuit 43. In the high band elimination circuit 43, one of the quantized values from the quantizer 42 that corresponds to an orthogonal transform coefficient in a predetermined high frequency band or more (hereinafter, appropriately referred to as a high frequency component).
Are removed (deleted). That is, the high band elimination circuit 43
From the quantized value, a high frequency component (quantized value) in, for example, frequency bands 8 to 10 as a high frequency band among the frequency bands 1 to 10 after the wavelet transform shown in FIG. Set the value to 0.

As a result, in the VLC circuit 47, only the quantized values in the frequency bands 1 to 7 are practically the targets of encoding, as shown in FIG. The amount of data can be reduced.

Here, the high frequency component corresponds to a fine change in the original image, and even if it is set to 0, it does not affect the global change so much. Therefore, even if the high frequency component is set to 0, there is almost no influence on the image quality of the decoded image obtained thereby.

The output data amount of the VLC circuit 46 is fed back to the high band elimination circuit 43, and the high band elimination circuit 43 deletes the quantization corresponding to this output data amount. It is designed to change the frequency band of values. That is, the high band elimination circuit 43
When the amount of output data is small, the frequency band of the quantized value to be deleted is increased to improve the quality of the decoded image. In addition, the high band elimination circuit 43
When the amount of output data is large, the frequency band of the quantized value to be deleted is lowered, whereby the amount of output data is reduced.

The quantized value in which the high frequency component is set to 0 in the high band elimination circuit 43 is supplied to the coefficient elimination circuit 44. In the coefficient deleting circuit 44, of the quantized values from the high band deleting circuit 43, those whose absolute value is a predetermined value or less are removed (deleted). That is, the coefficient deleting circuit 44 recognizes the texture region of the texture image based on the output of the contour line listing circuit 37, and among the quantized values in each texture region, does not significantly affect the image quality of the decoded image. , Which has a small absolute value is detected, and its quantized value is set to zero.

More specifically, the coefficient deleting circuit 44 uses, for example, a texture region (in the figure, as shown in FIG. 14A).
When a shadowed portion) is detected, the quantized value whose absolute value is 5 or less is set to 0. As a result, the quantized value output from the coefficient deleting circuit 44 is as shown in FIG.

In this case also, the input of the VLC circuit 46 is
Since the number of 0 runs increases, the output data amount can be reduced with almost no influence on the image quality of the decoded image.

The coefficient deletion circuit 44 is adapted to feed back the output data amount of the VLC circuit 46, and the coefficient deletion circuit 44 corresponds to this output data amount and outputs the quantization value to be deleted. It is designed to change the absolute value. That is, when the output data amount is small, the coefficient deleting circuit 44 reduces the absolute value of the quantized value to be deleted, thereby improving the image quality of the decoded image. Further, the coefficient deleting circuit 44 increases the absolute value range of the quantized value to be deleted when the output data amount is large, thereby decreasing the output data amount.

In the coefficient deleting circuit 44, the texture area is recognized and then the quantized value having a small absolute value is removed. However, this quantized value removal does not recognize the texture area. It is also possible to do this.

The quantized value whose absolute value is set to 0 is supplied to the coefficient rearranging circuit 45. The coefficient rearranging circuit 45 rearranges the quantized values from the coefficient deleting circuit 44 in a predetermined order. That is, the coefficient rearrangement circuit 4
5 recognizes the texture region of the texture image based on the output of the contour line listing circuit 37, and segments the quantized values for each texture region together. In this case, the texture image is a texture area Area1, A
15A, each texture area Area1, Area2, as shown in FIG.
.. are all arranged in the order of frequency bands 1 to 10 shown in FIG.

Alternatively, the coefficient rearrangement circuit 45
Divides the quantized value into groups for each frequency band. In this case, the quantized values are grouped regardless of which texture region they are in. That is, in this case, the quantized values are, for example, as shown in FIG. 15B, those in the frequency bands 1 to 4 as the low frequency band (Low), and the frequency bands 5 to 7 as the medium band (Medium). In the high frequency band (Hi
gh) in the frequency bands 8-10.

For texture images, the quantized values of orthogonal transform coefficients in the same texture region or the same frequency band are often almost the same. Therefore, by summing up such quantized values, V
The correlation of the input of the LC circuit 46 becomes high, and as a result, the output data amount can be reduced.

The quantized values rearranged by the coefficient rearranging circuit 45 are supplied to a VLC circuit 46, where they are subjected to variable length coding processing such as Huffman coding and run length coding, and then output. It is output via the buffer 47.

The image coding apparatus of the present invention is an image transmission apparatus that performs image transmission using a transmission medium having a limited transmission capacity, or records an image on a recording medium, such as a VTR.
It is applicable to devices such as (video tape recorder) and others.

Further, in the present embodiment, the wavelet transform is performed in the orthogonal transform circuit 38. However, the orthogonal transform circuit 38 is allowed to perform an orthogonal transform such as DCT other than the wavelet transform. Is possible.

In this case, it is necessary to divide the texture image into, for example, macroblocks, so that in principle block distortion occurs. However, as described above, the entropy coding circuit 5 of FIG. According to this, it is possible to reduce the amount of output data, which allows the quantizer 42 to allocate an appropriate number of bits. As a result, it is possible to prevent block distortion that greatly affects viewing of the decoded image.

However, in this case, for example, the texture image as shown in FIG. 8 is divided into macroblocks as shown in FIG. (A shaded portion in FIG. 16) is a processing target in blocks after the orthogonal transform circuit 38. However, in the blocks after the orthogonal transformation circuit 38, only the texture region needs to be processed, and in this case, the range to be processed unnecessarily expands. Therefore, from the viewpoint of performing efficient processing, it is preferable that the orthogonal transform circuit 38 be made to perform wavelet transform by DCT.

[0118]

As described above, according to the image coding apparatus and the image coding method of the present invention, a predetermined closed region is extracted from the image, and the image data in the closed region corresponds to the feature amount. Since the data is encoded, the encoding can be performed efficiently, and the data amount obtained as a result of the encoding can be reduced.

Further, in the recording medium of the present invention, data in which a predetermined closed region is extracted from an image and the image data in the closed region is encoded in correspondence with the feature amount is recorded. More images can be recorded. Also, the same image as the conventional one can be recorded with a small capacity.

[Brief description of drawings]

FIG. 1 is a block diagram illustrating a configuration of an embodiment of an image encoding device to which the present invention has been applied.

FIG. 2 is a block diagram showing a configuration example of an edge detection coding circuit 2 in FIG.

FIG. 3 is a diagram showing an example of tap coefficients of Sobel filters 11 and 12 in FIG.

4 is a diagram for explaining first and second mask regions set by a mask region setting circuit 17 in FIG.

5 is used in the threshold value determining circuit 18 of FIG.
It is a figure which shows the function which gives the threshold value T of edge strength.

6 is a block diagram showing a configuration example of an image decoding device that decodes an image encoded by the image encoding device of FIG. 1.

7 is a block diagram showing a configuration example of an entropy coding circuit 5 in FIG.

8 is a diagram showing an example of a texture image output from the texture region separation circuit 36 of FIG.

9 is a diagram for explaining the operation of the contour line listing circuit 37 of FIG. 7. FIG.

10 is a diagram showing an output example of the orthogonal transform circuit 38 of FIG. 7. FIG.

11 is a diagram showing a correspondence relationship between an activity and a quantization step size used in the quantization map creation circuit 41 of FIG.

12 is a diagram showing an example of weights given to a quantization step size in the quantization map creation circuit 41 of FIG.

13 is a diagram for explaining the operation of the high band elimination circuit 43 of FIG. 7. FIG.

14 is a diagram for explaining the operation of the coefficient deleting circuit 44 of FIG.

15 is a diagram for explaining the operation of the coefficient rearrangement circuit 45 of FIG.

16 is a diagram showing a range of macroblocks to be processed in subsequent blocks when DCT is performed by the orthogonal transform circuit 38 of FIG. 7.

FIG. 17 is a block diagram showing a configuration of an example of a conventional image encoding device (texture encoding device) for encoding a texture image.

FIG. 18 is a diagram showing an example of a texture image.

[Explanation of symbols]

 1 Local Luminance Detection Encoding Circuit 2 Edge Detection Encoding Circuit 3 Reconstruction Circuit 4 Operation Unit 5 Entropy Encoding Circuit 31 Edge Image Buffer 32 Reconstruction Image Frame Buffer 33 Original Image Frame Buffer 34 Edge Area Expansion Circuit 35 Operation Unit 36 Texture Region separation circuit 37 Contour line listing circuit 38 Orthogonal transformation circuit 39 Small region deletion circuit 40 Feature amount calculation circuit 41 Quantization map creation circuit 42 Quantizer 43 High band deletion circuit 44 Coefficient deletion circuit 45 Coefficient rearrangement circuit 46 VLC circuit 47 output buffer

Claims (17)

[Claims] 1. A closed region extraction unit that extracts a predetermined closed region from an image, a feature amount detection unit that detects a feature amount of image data in the closed region, and image data in the closed region. An image coding apparatus, comprising: a coding unit that codes in accordance with a quantity. 2. A contour line detecting unit for detecting a contour line surrounding the closed region extracted by the closed region extracting unit, wherein the feature amount detecting unit is image data in an area surrounded by the contour line. The image encoding according to claim 1, wherein the encoding means encodes the image data in the area surrounded by the contour line in correspondence with the characteristic amount. apparatus. 3. The encoding means includes an orthogonal transformation means that orthogonally transforms the image data and outputs orthogonal transformation coefficients, and a quantization means that quantizes the orthogonal transformation coefficients with a predetermined quantization step size. And a quantization step size determining means for determining the predetermined quantization step size when quantizing the orthogonal transform coefficient of the closed region in correspondence with the feature amount. The image encoding device according to 1. 4. The quantization step size determining means comprises:
The image coding apparatus according to claim 3, wherein a weight corresponding to a frequency band of the orthogonal transform coefficient is added to the predetermined quantization step size. 5. The coding means includes a variable length coding means for variable length coding the orthogonal transform coefficient quantized by the quantizing means, and the orthogonal transform coefficient quantized by the quantizing means. 4. The image coding apparatus according to claim 3, further comprising: segmenting means for segmenting each of the frequency bands into each of the frequency bands and supplying the segment to the variable length coding means. 6. The variable coding means for variable-length coding the orthogonal transform coefficient quantized by the quantizing means, and the orthogonal transform coefficient quantized by the quantizing means. The image coding apparatus according to claim 3, further comprising: segmenting means for segmenting each of the closed regions and supplying the segmented segment to the variable length coding means. 7. The image code according to claim 3, wherein the encoding unit further includes a high frequency band component removing unit that removes a component of a predetermined high frequency band or more from the orthogonal transform coefficient. Device. 8. The encoding means further comprises a small area closed area removing means for removing, of the orthogonal transform coefficients, those in the closed area having an area of a predetermined value or less. The image encoding device according to 1. 9. The encoding means further comprises a small absolute value coefficient removing means for removing those of the orthogonal transform coefficients whose absolute value is equal to or less than a predetermined value. Image coding device. 10. The image coding apparatus according to claim 1, wherein the feature amount detecting means detects a degree of change of image data in the closed region as the feature amount. 11. The feature amount detecting means detects a degree of change of image data in the closed region as the feature amount, and the quantization step size determining means determines a degree of change of the image data to a predetermined value. 4. The image coding apparatus according to claim 3, wherein the quantization step size in the closed region equal to or less than the lower limit value of, and in the closed region equal to or greater than a predetermined upper limit value is infinite. 12. The quantizing step size determining means determines the predetermined quantizing step size corresponding to the output data amount of the quantizing means in addition to the characteristic amount. Item 3. The image encoding device according to Item 3. 13. The image coding apparatus according to claim 7, wherein the high frequency band component removing means determines the predetermined high frequency band in accordance with an output data amount of the quantizing means. 14. The image coding apparatus according to claim 8, wherein the small area closed region removing means determines the predetermined value in correspondence with the output data amount of the quantizing means. 15. The image coding apparatus according to claim 9, wherein the small absolute value coefficient removing means determines the predetermined value in correspondence with the output data amount of the quantizing means. 16. An image coding method, wherein a predetermined closed region is extracted from an image, and the image data in the closed region is coded in accordance with its feature amount. 17. A recording medium, characterized in that a predetermined closed region is extracted from an image, and image data in the closed region is coded in correspondence with the characteristic amount of the image data.