JP2591736B2 - Contour / gradation separation / interpolation encoder - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a contour / tone separation / interpolation encoding apparatus capable of efficiently encoding an image signal and compressing the amount of information.

[Conventional technology]

2. Description of the Related Art Conventionally, a differential prediction encoding method has been widely used as an image signal encoding method.

In the difference prediction coding method, only the difference value of the gradation signal between adjacent pixels is quantized and transmitted by the encoder, and the received difference value is sequentially added to the reception-side decoder, and the gradation signal is added. Are synthesized.

Since such a difference prediction coding method can utilize the characteristic of an image signal that a correlation between adjacent pixels is high, compared with a method of directly quantizing a gradation signal of each pixel, image information is communicated. There is an advantage that a small amount of information is required. In other words, when the gradation signal of each pixel is directly quantized, a signal requiring a quantization level of 8 to 6 bits / pixel (256 to 64 levels / pixel) is required. Is 5 to 3 bits / pixel (32 to 8 bits) because the amplitude of the differential signal is less than about half the amplitude of the grayscale signal.
With the quantization level of (level / pixel), communication with almost the same image quality can be realized. For this reason, the amount of information can be reduced to about half as compared with direct quantization.

[Problems to be solved by the invention]

However, in the differential prediction encoding method, 2 bits /
If the quantization level is reduced to a pixel (4 levels / pixel) or less, the quantization error increases and the pixel quality deteriorates remarkably. Therefore, it is very difficult to compress information to 2 bits / pixel (4 levels / pixel) or less. Met.

Here, it is assumed that the amplitude of the difference signal is quantized at three levels / pixel. As shown in FIGS. 12 (a) and 12 (b), among the difference signals, the difference signal corresponding to the contour portion of the image has a relatively large amplitude, and the flat portion of the image (for example, a cheek portion in a human face image). Has a relatively small amplitude, it is very difficult to make the three levels appropriate quantization levels for either part. That is, using the quantization level as the threshold value of _GL , In the case of setting to three levels as described above, if the value is set to a value small enough to express the variation of the difference signal in the _GL flat portion, it cannot follow the rising of the steep difference signal in the contour portion. Also
If _GL is set to a value large enough to follow the steep rise of the contour portion, subtle variations in the flat portion cannot be expressed, and the image quality will be degraded.

An object of the present invention is that it is difficult to represent both a contour portion and a flat portion when the conventional difference prediction coding method reduces the quantization level of the gradation signal of each pixel to about three levels. And to provide an encoding apparatus which is excellent in the expression performance of both the contour part and the flat part even when the quantization level of each pixel is three levels.

[Means for solving the problem]

The present invention relates to an image from the image signal representing with a density gradation signals of the pixels and pixel input image, 2 preset to the difference of the gradation of an adjacent pixel threshold G _L and -G _L And +1, 0,
A contour information output unit that assigns -1 to generate ternary contour information of the image; a gradation information output unit that performs pixel thinning processing on the pixels to obtain gradation information of a coarse image; An encoder including a transmission unit that transmits the gradation information of the coarse image, a contour information buffer and a gradation information buffer that receive the encoded and transmitted contour information and the gradation information of the coarse image, respectively, A contour classification for supplying the contour information and the gradation information stored in the contour information buffer and the gradation buffer to an interpolation / synthesis unit, and classifying ternary contour information existing during each thinning interval (h). If there is no contour between the section and the two points to which the amplitude value of the gradation signal is applied, the two points (the n-th and n
Based on the amplitude value of the (+ h) th pixel, the amplitude values of the intermediate pixels (n + 1 to n + h-1) are linearly interpolated, and the amplitude value of the gradation signal exists between the two given points. value when a number sum n _E is positive ternary contour information of d ≧ n _E · G _L when ternary contour information representing an in contour contour information in the same direction all the d n A Gap value G _E = d / n _{E in the} contour portion and a Gap value G _N = 0 in the non-contour portion as a difference value between the amplitude values of the gradation signals of the nth and n + hth pixels, and d <n _E · a _{G L d> (2n E -h} ) G Gap values at Gap value G _E = G _L non contours in contour when the _{L G n = (n E ·} G L -d) / (h-n
_E ) However, the direction of the gap in the non-contour portion is set in the direction opposite to the corresponding ternary contour information, and d <n _E · _GL and d ≦ (2n
_E− h) In the case of _GL , two points (nth, n + h) given the amplitude value of the gradation signal
) Based on the amplitude value of the pixel (n)
+ 1-th to (n + h-1) -th amplitude values are linearly interpolated. When the sum n _E of the ternary contour information is the number of sub-elements and d ≦ n _E · _GL , the Gap value G _E = d in the contour portion / n _E Gap value in non-contour portion is set to G _N = 0, and when d> n _E · _GL and d <(2n _E + h) _GL , Gap value in contour portion G _E = _GL non-contour Gap value in section G _N = (d−n _E · _GL ) / (h + n)
_E ), where the direction of the gap in the non-contour portion is set in the direction opposite to the corresponding ternary contour information, and d> n _E · _GL and d ≧ (2n _E + h) _GL Sometimes, based on the amplitude values of the two points (n-th and n + h-th) pixels to which the amplitude values are given, intermediate pixels (n + 1-th to n-th)
The (n + h-1) th amplitude value is linearly interpolated. If the three-valued outline information existing between the two points to which the gradation signal is given and the three-valued outline information indicating the outline portion is in a different direction, the non- Set the Gap value G _N = 0 at the contour, and set n _P to represent the contour from the n-th to the (n + h) -th contour.
Among the values contour information, + direction of the sum of values, the n _M of the three values contour information representing the contour of the n-th to n + h th,
- as the sum of direction _{values, d ≧ (n P + n} M) · Gap value _{G P = (d-n M} · G L) in the edge portion of the + direction when the G _L / n _P - in the direction of the contour set Gap value _{G M = G L, d <} (n P + n M) · G Gap value G _P = G in the edge portion of the + direction when _{L L} - Gap value in the edge portion of the direction G _M = (d −n _P・_GL ) / n _M
And a Gap direction of a non-contour part corresponding to the ternary contour information E between adjacent pixels and the value of the ternary contour information, using the gradation information A (n) of the coarse image as an initial value. By adding or subtracting the obtained Gap value of the contour portion or the Gap value of the non-contour portion to A (n + h) between adjacent pixels with reference to the value of And a decoder including an interpolation / synthesis unit for interpolating the key information.

[Action]

The apparatus according to the present invention separates image information into ternary contour information representing a contour portion and gradation information representing a flat portion, and compresses the information using an optimal compression method for each information. Express both parts in quantities.

That is, the contour information representing the contour portion and the gradation information representing the flat portion have the following characteristic differences.

(I) Features of contour information representing a contour portion: Most of the essential semantic contents of image information are transmitted by the contour information. In other words, as an extreme example, a line drawing such as a binary portrait like a manga or animation can sufficiently convey the meaning content. For this reason, the image quality evaluation is better when the contour is clearly expressed.

As described above, the contour information requires high definition, but the expression of the halftone is considered unnecessary. Originally, it is suitable to handle the contour information as line drawing information, and information is effectively compressed by a coding method suitable for line images such as run-length coding.

(Ii) Features of gradation information representing a flat portion: The gradation information representing a gentle change in a flat portion is considered to be information for imparting naturalness and realism to contour information. The definition may be lower than the contour information, but in order to increase the naturalness, many quantization levels
It is necessary to increase the expression of the halftone. Therefore, there is a limit in reducing the quantization level.

The present invention focuses on the difference in characteristics between the contour information and the gradation information as described in (i) and (ii) above, and performs encoding and decoding according to the following procedure.

(A) Image information is ternary contour information quantized to three levels at high resolution, and at lower resolution,
It is separated into three or more multi-level quantized gradation information.

(B) The ternary contour information separated in (a) is subjected to information compression by a coding method suitable for line drawing type information compression, such as run-length coding and modified Huffman coding.

(C) For the gradation information separated by (a),
Compression is performed by a coding method suitable for information compression in a multi-tone format such as conventional difference prediction and linear prediction.

(D) The information compressed by (b) and (c) is
Transmit from the encoder to the other decoder.

(E) The decoder decodes each of the received compressed ternary contour information and compressed gradation information.

(F) The image signal is interpolated and synthesized based on the ternary contour information and the gradation signal obtained in (e).

Embodiment 1 FIG. 1 is a diagram for explaining a first embodiment of the present invention, wherein 1 is an encoder, 2 is a decoder, 3 is a communication path, and 4 is
An A / D conversion unit, 5 is an outline information output unit, 6 is a gradation information output unit, 7 is an interpolation / synthesis unit, and 8 is a D / A conversion unit.

In order to operate this, first, an input image signal is input to the A / D converter 4 and is quantized to a digital signal.

It is assumed that the digital signal has a gray scale of a bit ( ^2a level) for each pixel and the number of pixels is N × M. Next, this digital signal is output to the contour information output section 5 and the gradation information section output section 6.
Is input to

The contour information output unit 5 calculates a difference value between adjacent pixels with respect to a digital signal of an input image,
The difference value is converted into three values (three levels) under the threshold condition of the following expression and output.

Here, the threshold value _GL is set to a relatively large value so that the contour portion of the image can be extracted, ignoring the fluctuation of the flat portion of the image. As a result, the contour information of the number of pixels N × M is output at the three gradation levels (2 bits).

In the gradation information output unit 6, h in one direction (for example, X direction) with respect to the digital signal of the input image.
The number of pixels obtained by thinning out (subsampling) for each pixel Is further subjected to differential prediction encoding to set the gray scale of each pixel to b bits, that is, 2 ^b levels (b
<A). Here, the decimation process includes not only the decimation in the one-dimensional direction but also the two-dimensional direction (X, Y
Thinning can be performed in both directions). However,
In this case, the interpolation / synthesis processing in the interpolation / synthesis unit 7 described later also performs interpolation in the two-dimensional direction.

In this manner, the gradation d ^b level, the number of pixels (In the case of two-dimensional processing [H: X direction, h ': thinning interval in the Y direction]) is output.

The ternary contour information and the gradation information output as described above are transmitted to the decoder 2 via the communication path 3. Here, the total amount of information to be transmitted is given by the following equation.

That is, assuming that b = 3 bits and h = h '= 6, each pixel is compressed to an information amount of about 2.5 to 2.1 bits.

In the decoder 2, each of the ternary contour information and the gradation information is decoded from the received signal.
And synthesizes and outputs a digital image signal. The processing contents of the interpolation / synthesis unit 7 will be described later in a third embodiment.

The output digital image signal is decoded and output as an image signal via the D / A converter 8.

Embodiment 2 FIG. 2 is a diagram for explaining a second embodiment of the present invention, wherein 1 is an encoder, 2 is a decoder, 3 is a communication path, and 4 is
A / D conversion unit, 5 is an outline information output unit, 6 is a gradation information output unit, 7 is an interpolation / synthesis unit, 8 is a D / A conversion unit, 9 is a ternary outline extraction unit, and 10 is run-length encoding. , 11 is a thinning-out processing unit, 12 is a linear prediction encoding unit, 13 is a run-length decoding unit, and 14 is a linear prediction decoding unit.

In order to operate this, first, the input image signal is input to the A / D converter 4 and is quantized, the gradation of each pixel is a bit ( ^2a level), and the number of pixels is N × M. Digital signal.

Next, this digital signal is input to the contour information output unit 5 and the gradation information output unit 6.

The contour information output unit 5 includes a ternary contour extraction unit 9 and a run length encoding unit 10. First, the ternary contour extraction unit 9 calculates a difference value between adjacent pixels based on the input digital signal. Further, ternary contour information of the number N × M of pixels of the image is output according to the expression (1) of the first embodiment.

After that, the run-length encoding unit 10 performs run-length encoding on the ternary contour information.

That is, the lengths (run lengths) in the X direction and the Y direction for each of the three values of +1, 0, and -1 are represented by variable-length codes, and optimization according to the output frequency is performed according to the Huffman coding method. According to this method, ternary (2 bits) information of the number of pixels N × M can be compressed to a compression ratio α = 1/4 to 1/10.

That is, the contour information can be compressed to about 0.5 to 0.2 bits per pixel.

The gradation information output unit 6 includes a thinning-out processing unit 11 and a linear predictive encoding unit 12, and performs a process similar to that of the first embodiment to apply a one-dimensional direction (for example, X Direction)), thinning out h pixels
Number of pixels obtained Is further subjected to linear predictive encoding to compress the gray scale of each pixel to b bits, that is, ^2b levels (b <a). Here, similar to the first embodiment, the thinning processing can be performed not only in one-dimensional direction but also in two dimensions.

The contour information and the gradation information output as described above are transmitted to the decoder 2 via the communication path 3. Where the total amount of information transmitted is given by: Here, α is an average compression rate by run-length coding. That is, if b = 3 bits, h = h ′ = 6, α = 1/6, each pixel is compressed to an information amount of 0.8 to 0.4 bits.

In the decoder 2, the received ternary contour information is decoded in the run-length decoding unit 13, and the gradation information is decoded in the linear prediction decoding unit 14.

Further, the decoded ternary contour information and gradation information are input to the interpolation / synthesis unit 7 to synthesize and output a digital image signal. Then, it is decoded and output as an image signal via the D / A converter 8.

[Embodiment 3] Fig. 3 is a view for explaining an embodiment of the interpolation / synthesis unit 7 of the present invention, wherein 15 is an outline information buffer, 16 is a gradation information buffer, 17 is an outline classification unit, and 18 is a difference. Absolute value calculator (below
Gap calculation unit), 19 is an image signal synthesis unit.

In order to operate this, first, the received ternary contour information and gradation information are stored in the contour information buffer 15 and the gradation information buffer 16, respectively.

Next, the contour classifying unit 17 classifies the case where a contour is generated, and calculates the Gap value of each contour part (the absolute value of the amplitude difference between the grayscale signals of adjacent pixels in the contour part) according to the classification. On the basis of this, the image signal synthesizing section 19 synthesizes and outputs a digital signal.

Details of the above mechanism will be described below. Further, in realizing the embodiment shown in FIG. 3, it is easy to realize each unit by firmware using a microprocessor or the like instead of realizing each unit by a hardware circuit.

[3.1] In the following description, abbreviations are defined as follows, as appropriate. FIG. 4 shows an example of the relationship between the abbreviations.

(1) K: number of each pixel K = 0, 1, 2,..., N,..., N + h,... (2) J: difference position for representing the difference between the gradation signals of each pixel The position of the difference between the gradation signals of the (n-1) th pixel and the nth pixel is referred to as the nth difference position.

J = 1,2, ..., n, ..., n + h, ... (3) A: Amplitude of gradation signal of each pixel A (N): Amplitude of gradation signal of nth pixel (4) E : Outline information of each difference position Indicates the magnitude of the difference of the gradation signal between the pixels on both sides of each difference position.

The contour information of each difference position is given by one of the following three values.

Also, in this method, From the two types of information, K = n + 1, n + 2,..., N + h−1
The amplitude A (n + 1), A (n +
2),..., A (n + h−1) are estimated / interpolated and decoded.

The following abbreviations are used to describe this decoding scheme.

(5) d = A (n + h) -A (n): difference value of the amplitude of the gradation signal between the n-th pixel and the n + h-th pixel : Sum of ternary contour information at difference positions from nth to n + hth : the sum of values in the + direction (contour portion in the + direction) of the ternary contour information at the difference positions from the nth to the (n + h) th position : Sum of values in the negative direction (outline portion in the negative direction) of the outline information of the difference positions from the nth to the (n + h) th positions. (9) G _E , G _N , G _P , G _M : Tone of adjacent pixels The absolute value of the difference between the signal amplitudes. In the following description, this is called a Gap value.

G _E is the Gap value in the contour part G _N is the Gap value in the non-contour part G _P is the Gap value in the positive contour part G _M is the Gap value in the negative contour part [3.2] Decoding method by interpolation / synthesis , The following two types of information are provided to the decoder.

From these information, K = n + 1, n + 2,..., N + h−1
The amplitude A (n +
1),..., A (n + h−1) are interpolated and decoded.

[3.3] Classification of ternary contour information between two points given the amplitude value of the gradation signal Three-valued contour information E (n + 1), (n + 2),.
.., E (n + h-1) can be classified into the following three types.

(A) When there is no contour between two points (B) When all the contour information existing between two points is in the same direction (C) When there is contour information in two directions of + and-between two points For each of the three cases, the amplitude values of the tone signals of the n + 1 to n + h-1th pixels from the contour information are interpolated by the following methods.

[3.3-A] When there is no contour between two points given the amplitude value of the gradation signal FIG. 5 shows an example of interpolation in this case. For this case, two pixels (n-th and n-th) given the amplitude value
Based on the amplitude values A (n) and A (n + h) of the (h) th pixel, the amplitudes of the intermediate pixels (h + 1 pixels from the (n + 1) th to the (n + h-1) th) are represented by A (n) and A (n). (N +
h) is given as a value to be linearly interpolated. That is, it is given by the following equation.

Here, j = 1, 2,..., H-1 [3.3-B] When the ternary contour information existing between the two points to which the gradation signal is given is all in only the same direction, From the nth to n +
Interpolation is performed on the pixels up to the h-th pixel.

(Condition 1) The same Gap value G _{E for} all contour parts
give.

(Condition 2) For the non-contour parts, all have the same Gap value G
Give _N.

(Condition 3) Gap, value G _E contoured portion is set to a value of more than G _L (G
The minimum value of _E is _GL ).

(Condition 4) The Gap value G _N of the non-contour part is a value less than _GL (G
The maximum value of _N is _GL ) and is as small as possible.

Based on the above condition, the gradation signal A of the n-th pixel
When performing interpolation between (n) and the gradation signal A (n + h) of the (n + h) th pixel, it is determined whether or not there are Gap values G _E and G _N satisfying the above conditions by comparing d and n _E with each other. According to the relationship between them, they can be classified as shown in Table 1. That is, Table 1 shows the relationship between the Gap values G _E and G _N when the contour information existing between two points is only in the same direction. Note that the shaded portion in the table indicates the case where no portion exists. D classified in Table 1
6 and 7 show examples of the relationship between _nE and nE.

Based on the relationship between d and n _E classified in Table 1, the Gap value G _E of the contour portion and the Gap value G _N of the non-contour portion are given by the values shown in Table 2.

That is, Table 2 shows the relationship between the Gap values G _E and G _N of the contour portion and the non-contour portion when the contour information existing between the two points is only in the same direction. Generally, the amplitude of the gradation signal of each pixel is interpolated by the following equation.

However, Here, the values of G _E and G _N are given according to Table 2.

However, for the case of _{d> n E · G L and} d ≧ (2n E + h) G L and n E <0 d <n E · G L and d ≧ (2n E -h) G L and n E> 0 Since the conditions 1 to 4 cannot be satisfied, the gradation signal of each pixel is given by linear interpolation between two points, and is interpolated by the expression (2).

j = 1, 2,..., h-1 [3.3-C] When the ternary contour information existing between the two points to which the gradation signal is given is in two directions (+ and-directions) Based on this, interpolation is performed for pixels between the n-th pixel and the (n + h) -th pixel.

(Condition 1) For contour parts in the same direction, the same Gap
Give a value.

(Condition 2) + direction Gap value G _P or the outline of the - direction of the smaller of either the Gap value of Gap value G _M contoured portion and G _L. (The minimum Gap value of the contour part is _GL ).

(Condition 3) The Gap value of the non-contour portion is set to zero (= 0).

Based on the above condition, the gradation signal A of the n-th pixel
When performing interpolation between (n) and the gradation signal A (n + h) of the (n + h) th pixel, the Gap value _GP of the contour portion in the + direction and −
Which of the smaller the Gap value G _M of the contour part in the direction becomes the smaller value (that is, G _L ) depends on the relationship between d, n _P and n _M.
It can be classified as shown in Table 3. That is, Table 3 shows the magnitude relationship between the Gap values G _E and G _N of the contour portions in the + direction and the − direction when the contour portions in both the + and − directions exist between the two points.
Note that the shaded portion in the table indicates the case where no portion exists. Relationship between d and n _P , n _M classified in Table 3 8 to 11 are shown in FIGS.

Based on the relationship between d and n _P , n _M classified in Table 3,
+ Direction and the - direction of the contour of the Gap values G _P, G _M is given by the values shown in Table 4. That is, Table 4 shows +,-between two points.
+ Direction and the case where both of the contour is present - illustrates the direction of the contour of the Gap values G _P, the value of G _M. Note that the Gap value for the non-contour portion is 0.

For the n-th to n + h-th pixels, the Gap value of the gradation signal (the absolute value of the difference value of the gradation signal of the adjacent pixel) for the difference position between each pixel is defined by the contour information (+1, Interpolation is performed by giving the values shown in Table 4 according to (0, -1).

That is, generally, the amplitude of the gradation signal of each pixel is interpolated by the following equation.

However, Here, G _P, the value of G _M gives accordance Table 4.

j = 1, 2,..., h−1 As can be seen from the results described above, by using the present invention, it is possible to compress the information amount to 1 bit / pixel or less without deteriorating the image quality. Image communication in a narrower band than before can be realized.

〔The invention's effect〕

As described above, the apparatus according to the present invention separates image information into ternary contour information indicating a contour and gradation information indicating a flat part, and compresses the information using an optimal compression method for each information. As a result, the image signal can be compressed to an information amount of 1 bit or less per pixel, so that there is an advantage that moving picture communication at a low bit rate in a narrow band using an existing telephone line can be realized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment of the present invention, FIG. 2 is a block diagram showing a configuration of a second embodiment of the present invention, and FIG. 3 is a third embodiment of the present invention. FIG. 4 to FIG. 12 are explanatory diagrams of a mechanism for performing interpolation / synthesis from ternary contour information and gradation information according to the present invention. In the figure, 1 is an encoder, 2 is a decoder, 3 is a communication path, and 4 is A / D.
Conversion unit, 5 is an outline information output unit, 6 is a gradation information output unit, 7
Denotes an interpolation / synthesis unit, 8 denotes a D / A conversion unit, 9 denotes a ternary contour extraction unit, 10 denotes a run-length encoding unit, 11 denotes a thinning processing unit, and 12 denotes a thinning processing unit.
Is a linear prediction encoding unit, 13 is a run length decoding unit, 14 is a linear prediction decoding unit, 15 is an outline information buffer, 16 is a gradation information buffer, 17 is an outline classification unit, 18 is a Gap calculation unit, and 19 is An image signal synthesizing unit.

Claims (2)

(57) [Claims] An image signal is input and an image signal is represented by using a pixel and a density gradation signal of the pixel. The difference between the gradation of adjacent pixels and two preset threshold values _GL and -G are obtained. by comparing the _L respectively in accordance with their size + 1,0, -
1; a contour information output unit for generating ternary contour information of the image; a gradation information output unit for performing pixel thinning processing on the pixels to obtain gradation information of a coarse image; An encoder having a transmission unit that transmits the gradation information of the contour information buffer and a gradation information buffer that respectively receive the encoded transmitted contour information and the gradation information of the coarse image; A contour classifying unit that supplies the contour information and the gradation information stored in the information buffer and the gradation buffer to an interpolation / synthesis unit, and classifies ternary contour information existing during each thinning interval (h); If there is no contour portion between the two points to which the amplitude value of the gradation signal is given, the two points (the nth and n +
Based on the amplitude value of the (h-th) pixel, the amplitude values of the intermediate pixels (n + 1-th to (n + h-1) -th) are linearly interpolated, and the amplitude value of the gradation signal exists between two given points. value when a number sum n _E is positive ternary contour information of d ≧ n _E · G _L when ternary contour information representing an in contour contour information in the same direction all the d n A Gap value G _E = d / n _{E in the} contour portion and a Gap value G _N = 0 in the non-contour portion as a difference value between the amplitude values of the gradation signals of the nth and n + hth pixels, and d <n _E · a _{G L d> (2n E -h} ) G Gap value _{G n = (n E · G} L -d) in Gap value G _E = G _L non contours in contour when the _L / (h-
n _E ) However, the direction of the gap of the non-contour portion is set in the direction opposite to the corresponding ternary contour information, and d <n _E · _GL and d ≦ (2n _E
-H) In the case of _GL , based on the amplitude values of the two points (n-th and n + h-th) pixels to which the amplitude value of the gradation signal is given, an intermediate pixel (n +
The first to (n + h-1) th amplitude values are linearly interpolated. When the sum n _E of the ternary contour information is a negative number and d ≦ n _E · _GL , the Gap value G _E = d in the contour portion / n _E Gap value in non-contour portion is set to G _N = 0, and when d> n _E · _GL and d <(2n _E + h) _GL , Gap value in contour portion G _E = _GL non-contour Gap value G _N = (d−n _E · _GL ) / (h +
n _E ), where the direction of the gap in the non-contour portion is set in the direction opposite to the corresponding ternary contour information, and d> n _E · _GL and d ≧ (2n _E + h) _GL In the case of, based on the amplitude values of the two points (n-th and n + h-th) pixels to which the amplitude values are given, intermediate pixels (n + 1 to n)
+ H-1) is interpolated linearly. If the ternary contour information existing between the two points to which the gradation signal is given and the ternary contour information indicating the contour portion is in a different direction, the non- Set the Gap value G _N = 0 at the contour, and set n _P to represent the contour from the n-th to the (n + h) -th contour.
In the value contour information, the sum of the values in the + direction, n _M, is the ternary contour information representing the contour portions from nth to n + hth.
- as the sum of direction _{values, d ≧ (n P + n} M) · Gap value _{G P = (d-n M} · G L) in the edge portion of the + direction when the G _L / n _P - in the direction of the contour set Gap value _{G M = G L, d <} (n P + n M) · G Gap value G _P = G in the edge portion of the + direction when _{L L} - Gap value in the edge portion of the direction G _M = (d −n _P · _GL ) / n _M Gap calculation unit, and the value of ternary contour information E and ternary contour information between adjacent pixels with coarse image gradation information A (n) as initial values With reference to the value of the Gap direction of the non-contour portion corresponding to the above, the addition or subtraction of the obtained Gap value of the contour portion or the Gap value of the non-contour portion is performed up to A (n + h) between adjacent pixels. And a decoder having an interpolation / synthesis unit for interpolating the gradation information of the decimated pixels. 2. An encoder having means for variable-length encoding the length of each level value in the number of pixels with respect to ternary contour information by run-length encoding, compressing and transmitting the encoded data, and run-length encoding by means of run-length encoding. 2. A decoder according to claim 1, further comprising means for decoding the compressed received ternary contour information by run-length decoding. An interpolation coding device;