JP2608285B2 - Image processing device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus that encodes brightness information and color information representing a color image.

2. Description of the Related Art Conventionally, some color image coding apparatuses divide color image data into brightness information and color information, and code each of them.

[Problems to be Solved by the Invention] However, in the conventional coding method, when a high-resolution image is coded, if the coding is performed such that the amount of color information of the color image data is reduced, an edge is generated. Parts are not reproduced efficiently.

Conversely, when a halftone image is encoded, the encoding efficiency is reduced unless encoding is performed such that the amount of color information of the color image data is reduced.

Further, when decoding color image data, if the data amount of a data unit including coded data differs for each predetermined image unit, there is a problem that it is difficult to handle the data.

SUMMARY OF THE INVENTION The present invention has been proposed to eliminate the above-mentioned drawbacks of the conventional example, and its purpose is to decode encoded color image data when encoding image brightness information and color information. The purpose is to encode data units that are easy to handle for time and the like.

In particular, an object of the present invention is to perform efficient encoding using the type of the image to be encoded.

[Means for Solving the Problems] In order to solve the above problems, according to an image processing apparatus of the present invention, an image processing apparatus for processing a color image divided into a plurality of blocks, A feature amount indicating a distribution of pixel values of each block is extracted based on the color information, and each block is divided into a first block including a predetermined amount of edges in the block and the first block according to the feature amount. Classifying means for classifying the divided blocks into second blocks different from the blocks; coding means for coding brightness information and color information of the divided blocks; brightness information coded by the coding means; And an adding unit for adding classification information indicating a classification result by the classification unit to the color information, wherein the encoding unit converts the color information into a long code length for the image of the first block. With the sign of And the brightness information is encoded with a short code length, and for the image of the second block, the color information is encoded with a code having a short code length, and the brightness information is encoded with a long code length. Further, the adding unit converts the data unit including the brightness information and the color information encoded by the encoding unit of the divided block and the classification information added to the information into the data unit of the block by the classification unit. It is characterized in that a predetermined data amount is set regardless of the classification result.

[Embodiment] Hereinafter, two embodiments according to the present invention will be described with reference to the accompanying drawings in FIGS.

First Embodiment <Pixel Block> In this first embodiment, for simplicity of description, the size of a pixel block is 2 × 2 pixels. FIG. 2 is a diagram showing the division of the pixel block. In FIG. 2, the image is divided into four blocks (2a to 2d). 1 indicates a color pixel. The color pixel 1 has information of R (red), G (green), and B (blue), which are original stimuli of color, as a plurality of bits of information (hereinafter referred to as multi-valued information).
Here, for convenience of explanation, each of the R, G, and B signals is assumed to be 8 bits. Therefore, one color pixel has a total information amount of 24 bits. FIG. 3 shows the configuration of such a pixel in a block.

3, 3a to 3d represent one color pixel in FIG. Hereinafter, the color pixels 3a, 3b, 3c, 3 in this pixel block
Let the color image data of d be X ₁₁ , X ₁₂ , X ₂₁ , and X ₂₂ , respectively. X ₁₁ to X ₂₂ of each pixel is multivalue information of R, G, B as described above. Let the information of R of X _{11 be} R _11, and similarly, G for G and B
₁₁ , B _11, and the same applies to X ₁₂ , X ₂₁ , X ₂₂ , X ₁₁ = ｛R ₁₁ , G ₁₁ , B ₁₁ ｝, X ₁₂ = ｛R ₁₂ , G ₁₂ , B ₁₂ ｝, X _{21 = {R 21, G 21,} B 21}, X 22 = {R 22, G 22, B 22}, and made.

<Overview of Overall First Embodiment> FIG. 1 is an overall block diagram of an encoding device according to a first embodiment. This encoding apparatus encodes each block with four pixels as one block, and converts it into an encoded code as shown in FIG. 9A or 9B according to the characteristics of the block. In the encoding process, the RGB signal is converted by the ROM 4 into a Y (brightness) IQ (color difference) signal, and the converted YIQ signal is extracted by the arithmetic units 8, 9, and 10 for each signal. And, depending on its characteristics, two encodings by encoders 11 and 12 (encoder 11 encodes according to the format of FIG. 9A and encoder 12 encodes according to the format of FIG. 9B). One of these two encoded codes is output from the synthesizer 13.

FIG. 4 shows an outline of the encoding according to the first embodiment together with FIGS. 9A and 9B. In the figure, a decoding method is also shown to make it easier to understand the characteristics of the coding. m _Y indicates the average brightness in the block. σ _Y
Is edge information on the brightness in the block (in this embodiment, the standard deviation of Y _ij is used as edge information). N _ij is a value obtained by standardizing a change in brightness from the block center (having brightness m _Y ) in each pixel direction by the standard deviation σ _Y , that is, N _ij = (Y _ij −m _Y ) / σ _Y. Accordance connexion, by _{decoding, Y ij '= N ij ·} σ Y + m Y I I' = m I I Q '= m Q is obtained. The circuit of the embodiment shown in FIG. 1 is configured to realize the above encoding.

<RGB → YIQ Conversion> Returning to FIG. 1, reference numeral 4 denotes a signal conversion read-only memory (hereinafter abbreviated as ROM) for converting RGB information into YIQ information with reference to a table. I and Q are luminance information Y and color difference information I and Q generally used as TV signals. This signal conversion ROM 4 uses R, G, B information as an input address and outputs Y, I, Q information as output data.
Therefore, the Y-information calculated from X ₁₁ Y _11, I information I _11, Q
Information is referred to as Q _11. Hereinafter, the same applies to X ₁₂ , X ₂₁ , and X ₂₂ .

Reference numerals 5, 6, and 7 denote registers for accumulating the Y, I, and Q information of one block converted by the signal conversion ROM 4 for each YIQ. That is, 5 is a Y information register, 6 is an I information register, 7 is Q
It is an information register. These registers 5, 6, 7 contain
_{X 11, X 12, X 21} , Y calculated from _{X 22 11 ~Y 22, I 11} ~I 22, Q 11
~Q ₂₂ information is incorporated into the order.

All information X ₁₁ to X ₂₂ as information for <feature extraction> block, after being stored in the register 5-7 is input to the arithmetic unit 8-10 connected to the register. The configurations of these arithmetic units 8 to 10 are as shown in FIG. Here, the arithmetic unit 8 has the circuit configuration shown in FIG. 5, and the arithmetic units 9 and 10 are equivalent to the configuration of the portion 34 indicated by a broken line including the average value calculator 22 in FIG.

The arithmetic unit 8 calculates the average value of the signal Y in the block.
m _Y , a standard deviation σ _Y indicating edge information about brightness,
Then, for each pixel direction, _Y of each pixel is calculated from the average value mY.
₁₁ brightness change of to Y ₂₂ calculates those normalized by the sigma _Y. Further, the calculator 9 and 10, the average value m _I in a block of each I information, calculates the average value m _Q in the block of Q information.

The arithmetic unit will be described with reference to FIG. 5 by taking as an example an arithmetic unit 8 for extracting a feature of Y information. Terminal 18 inputs the Y ₁₁ information, the terminal 19 is Y _12, the terminal 20 is Y _21, the terminal 21 inputs the Y ₂₂ information. The average value calculator 22 adds all of these pieces of information, and divides the sum by the number of input information (four). In this case, the lower 2 bits of the output of the average value calculator 22 are shifted to the lower side, thereby dividing by 4. Therefore, the output m _Y of the average calculator 22 is Becomes

As described above, the arithmetic units 9 and 10 use a circuit composed only of the portion surrounded by the broken line 34 in FIG. 5, and output the average value m _I in the block for I and Q from the output terminal 28 thereof. , m _Q is output.

Subsequently, the edge information amount regarding the brightness calculated by the calculator 8 will be described. In this first embodiment,
The standard deviation σ _Y in the block is used as the edge information amount regarding the brightness. To determine the standard deviation sigma _Y, subtractor 23, ROM24,26, and the average value calculator 25 are used. Σ _Y output from ROM 26 is Is calculated.

That is, subtractor brightness of the pixel in the 23 information Y ₁₁ to Y ₂₂ is input as minuend value, the average value m _Y in the block is input as a subtraction value. Accordingly, the subtractor 23 outputs a difference value S _ij from the average value m _Y of the brightness information Y _ij of each pixel. That is, S _ij = Y _ij −m _Y {i = 1,2, j = 1,2}. These difference values S _ij is input to the ROM 24. ROM24
Is a ROM for outputting the square of the input. The output of the ROM 24 is the square of the difference value S _ij of each pixel, that is, S _ij ² = (Y _ij −m _Y ) ² {i = 1,2, j = 1,2}. The square value of the difference value S _ij of each pixel is input to the average value calculator 25 and the output σ _Y ² is Which is the variance within the block. This result is RO
Input to M26. ROM 26 outputs the square root of the input. That is, Is output. This represents the standard deviation within the block. This is output from the terminal 29 as the edge information amount regarding brightness. The above ROMs 24 and 26 are LUT (Look Up Table)
A method ROM or the like may be used.

In the first embodiment, the value of σ _Y is used as a criterion for determining whether an edge exists in a block.

It will now be described a method for normalizing the Y information Y ₁₁ to Y _22. The output of the aforementioned subtracter 23 is the difference value S ₁₁ to S ₂₂ of the mean value m _Y and each Y ₁₁ to Y _21. At divider 27 divides these difference values in the standard deviation sigma _Y. That is, if this quotient is N _ij , It is. However, when the standard deviation σ _Y is “0”, N _ij is all zero. σ _Y is a standard deviation, that is, an average variation of each pixel. N obtained by dividing S _ij by this σ _Y
ij is considered to indicate the gradient of brightness from the center position of the block to each pixel direction as shown in FIG. S
_{Although ij is} also considered to be a gradient, by normalizing with σ _Y , N _ij is an amount indicating a gradient independent of the block size. Incidentally, N _ij In view of the fact YIQ is respectively 8 bits, and 9 bits long.

The standard value C ₁₁ in the Y _ij from the terminal 30, or less, from the terminal 31
Standard value C ₁₂ in the Y _12, the standard value C ₂₁ in the Y ₂₁ from the terminal 32, the terminal
From 33 outputs the standard value C ₂₂ in the Y _22.

<Encoding> The outputs m _Y , σ _Y , and N _ij of the arithmetic unit 8 are input to the encoder 11, and the outputs m _I , m _{Q of the} arithmetic units 9 and 10 and the σ _Y are encoded. Is input to the device 12 (see FIG. 1).

FIG. 6 shows a block diagram of the encoder 11 for brightness information.
The encoder 11 comprises a ROM 41, a ROM 42 and a selector 43. Terminals 37 to 40 receive the outputs σ _Y and N _ij from the computing unit 8 in FIG. 5, and the input terminal 36 similarly receives the edge information amount σ _Y. ROM41 and ROM42 are input standardized values respectively
N _{11 to} N ₂₂ are quantized in several steps according to their size. As described above, when the R, G, B are each 8 bits, Y, I, Q information also has information about each 8 bits and the normalized value N ₁₁ to N ₂₂ nine-bit (= 512 steps) Has the information amount of each. Therefore, each normalized value N _ij is equally divided and quantized into 4 bits (= 16 steps) in the ROM 41 and is equally divided and quantized into 3 bits (= 8 steps) in the ROM 42 to reduce the information amount. The reason why the degree of quantization is changed is that there is no problem even if the quantization is coarse when the block does not include a structure such as an edge. Therefore, the output from the ROM 41 is N ₁₁ ,
N ₁₂ , N ₂₁ , and N _{22 have} a 16-bit length in the order, and the output from the ROM 42 has a 12-bit length.

The output of the ROM 41 is sent to the selector 43. On the other hand, the 12 bits of the output of the ROM 42 further have 4-bit “00” as lower bits.
00 "is transmitted to the selector 43.
Outputs the outputs of the ROM 41 and the ROM 42 in accordance with the magnitude of the edge information amount σ _Y regarding the brightness. That is,
Assuming that T ₁ is a predetermined threshold value, N _ij : 4-bit quantization (σ _Y > T ₁ ) N _ij : 3-bit quantization (σ _Y > T ₁ ). Performing different quantization depending on the magnitude of σ _Y in this way is because when it is determined based on the brightness information Y that there is a large change in brightness in the block, information related to the brightness ( In particular, this is because the quantization density of N _ij ) is made fine. Conversely, when it is determined that there is no large change in brightness in the block, the information amount of the brightness information is made small, and as a result, This is because it is possible to increase the amount of color information.

The edge information amount σ _Y (7 bits) regarding the brightness is added to the most significant bit portion of the output of the selector 43 selected in this manner, and further below that, the average of the brightness information in the block is averaged. The value m _Y (8 bits) is added and output from the terminal 44 as brightness information (31 bits in total).

In FIG. 6, a symbol such as “MSB∨LSB” represents a shifter. For example, the shifter 63 converts the MSB side input data into MSB.
The input data on the LSB input side is arranged in order from the MSB side input data.

Next, FIG. 7 shows a block diagram of the encoder 12 (FIG. 1) for encoding the color information (I, Q).

Terminal 46 inputs the average value m _I in the block of I information outputted from the arithmetic unit 9 (Figure 1). Similarly, a terminal 47 inputs an average value m _Q in the block of Q information. One input of the selector 48 is obtained by combining m _I and m _Q in the order of I and Q.
Since m _I and m _Q are each 8 bits in size, the combined one is 16 bits. On the other hand, the other inputs of the selector 48 are shifted by the shifters 64 and 65 to the average values m _I and m _Q , respectively, in the lower two positions.
The bits deleted one by one are combined by the shifter 66 in the order of I and Q from the upper bits, and the shifter 67 adds 4-bit “0” to the upper bits. The selector 48 Yotsute selectively outputs these two inputs to the size of the edge information amount sigma _Y. That is, the average value
m _I and m _Q are quantized to 6 bits (if σ _Y > T ₁ ) and to 8 bits (if σ _Y > T ₁ ). If the magnitude of the edge information amount σ _Y is larger than the above-described predetermined threshold T ₁ (if σ _Y > T ₁ ), the latter input, that is, the higher order of the 6-bit average values m _I and m _Q 4 more
Select the one with the bit "0" added. Conversely, if the magnitude of σ _Y is smaller than the threshold T ₁ , the former input, that is, a combination of the 8-bit average values m _I and m _Q as they are, is selected. In other words, when it is determined based on the brightness information that a large change in brightness exists in the block, the information amount of the color information is reduced, and when it is determined that the block does not exist, the information amount of the color information is reduced. To increase.

The output of the selector 48 selected in this way is
This is one input of the synthesizer 13 shown in FIG.

<Synthesis> The outputs of the encoders 11 and 12 are input to the synthesizer 13. FIG. 8 shows the configuration of the synthesizer 13. In the figure, a terminal 50 inputs the encoded brightness information from the encoder 11. The terminal 51 receives the encoded color information from the encoder 12. The shifter 70 extracts the lower 4 bits from the brightness information, and the shifter 71 extracts the upper 4 bits from the color information. The OR circuit 52 calculates the logical sum of the 4 bits from the upper bits.

That is, when the magnitude of the edge information amount σ _Y regarding the brightness is larger than the threshold value T ₁ , significant information of the brightness information is contained in 4 bits of the LSB output of the shifter 70, and Since "0" is entered in the four bits of the MSB output, the output of the OR circuit 52 contains significant information (4
Bit length). Conversely, when σ _Y is smaller than the threshold value T ₁ , “0” is output to the LSB output of the shifter 70 and the shifter 71
Because significant information of color information is included in the MSB output of
The output of R circuit 52 is 4-bit significant information about the color.

In the shifter 72, the information (27 bits) excluding the lower 4 bits of the brightness information is set as the highest order, then the output (4 bits) of the OR circuit 52 is arranged, and the upper 4 bits of the color information are set at the lowest order. The removed information (12 bits) is arranged and output from the terminal 53, thereby completing the encoding.

FIGS. 9A and 9B show the arrangement of the 43 bits thus encoded. Figure 9A if the edge information amount sigma _Y is greater than thresholds T ₁ Katsuta, FIG. 9B is sigma _Y said thresholds T ₁
This is the arrangement of the bits of the code in the case of being smaller. or,
The numbers on the top of the figure indicate the bit positions.

<Decoding> The decoding of the code according to the first embodiment may be performed by reversing the above-described procedure. That is, 37 to 42 most significant bits of the input code to (sigma _Y) is taken out, as compared with the thresholds T ₁ of said, check whether those containing large changes in block brightness,
It is determined whether the code bit arrangement is as shown in FIG. 9A or FIG. 9B. Next, take out the 36-28 th bit as the mean value m _Y of Y information.

Further, N ₁₁ to N ₂₂ the Y information Y ₁₁ to Y ₂₂ of each pixel normalized
Is obtained by extracting four bits at a time when it is determined that there is a large change in brightness in the block (FIG. 9A), and is obtained by extracting three bits at a time when it is determined that there is no change (FIG. 9B). For the remaining bits, these are divided into two,
The average value m _I in the block of information and the lower order are obtained as the average value m _Q in the block of Q information.

Therefore, assuming that Y information, I information, and Q information to be reproduced are Y ₁₁ ′ to Y ₂₂ ′, I ₁₁ ′ to I ₂₂ ′, and Q _{11 to} Q ₂₂ ′ for each pixel, Y _ij ′ = N _ij × σ _Y + m _Y I _ij = m _I Q _ij = m _Q where i = 1,2, j = 1,2.

Furthermore, as in the case of encoding, R information reproduced using a ROM for converting YIQ information into RGB information with reference to a table is used.
GB information R ', G', B 'is obtained. In this manner, decoding may be performed by the reverse procedure of encoding.

<Modification of First Embodiment> In the first embodiment, the standard deviation value in the block is used as the edge information amount regarding the brightness. However, the difference between the maximum value and the minimum value of the pixel values in the block is also used. Can be used to simplify the circuit configuration. Further, in the embodiment, Y, I, Q information of TV system is used as an information form for encoding, but an information form for decomposing into other brightness information and color information,
For example, L ^* , a ^* , b ^* information in the CIE1976 (L ^* , a ^* , b ^* ) space may of course be used. Further, in the embodiment, thinning allocation of bits is used as an encoder for simplicity of description, but other encoding, for example, vector quantization encoding may be used.

<Effects of First Embodiment> According to the first embodiment, the information amount per block was 96 bits before encoding.
It achieves a compression ratio of 44.8%. Further, when brightness information and color information are independently subjected to fixed-length encoding without performing encoding according to the presence or absence of an edge, the brightness information 31
Bits, color information 16 bits, a total of 47 bits are required,
By performing encoding in accordance with the presence or absence of an edge, this can be further reduced by 4 bits, and encoding with less visual deterioration can be performed.

Thus, when a color image is divided into a plurality of pixel blocks, and the brightness information and the color information are decomposed in units of blocks, and a fixed-length encoding is performed as a whole, the brightness that is easily generated during the encoding is as follows: By changing the magnitudes of the information amounts of the brightness information and the color information based on the information of the edge information amount (for example, the standard deviation) with respect to, and performing variable coding, and performing fixed-length coding as a whole, Utilizing the visual characteristics of human masking, there is no visual deterioration,
Encoding that provides high efficiency can be performed.

: In addition, since the encoding is of fixed length in block units, when the image data is stored in an image memory or the like, it is possible to maintain the positional relationship of the pixels in the image. Therefore, the required amount of memory can be kept constant depending on the size of the image, and even when performing image processing, it is possible to easily extract information on pixels near the periphery.

Second Embodiment A second embodiment will be described with reference to FIG.

<Outline of Second Embodiment> The outline of the second embodiment will be described in comparison with the first embodiment.

: Resolution Information In the first embodiment, information on resolution is detected as an edge of brightness, and this edge information is also encoded and stored.
In the second embodiment, the color edge is detected from the color information itself. Color edges are detected for each of the two colors. Then, as shown in FIG. 11, the average value m _a1 (or m _b1 ) of the color information for each of the two regions divided by the color edge,
Output m _a2 (or m _b2 ). If no color edges what the mean value m _a or m _b of the entire block is quantized and encoded code.

: Edge detection In the first embodiment, the edge is detected from the standard deviation σ _Y in the block. In the second embodiment, the edge is detected from the difference between the maximum value and the minimum value of the color information.

： Color system YIQ in the first embodiment, whereas 1 in the second embodiment
CIE197, proposed as a uniform perceived color space at the 975 CIE Conference
^{6 (L *, a *,} b *) L * is an index of the space, ^{a *,} using ^{b *.} Here, the value of L ^* represents lightness, and the values of a ^* and b ^* represent perceived chromaticity.

: Pixel block 4 pixels / block shown in FIG. 3 are used. However, for convenience of explanation, in the second embodiment, signals related to the pixels X ₁₁ , X ₁₂ , X ₂₁ , and X ₂₂ are given suffixes 1, 2, 3, and 4 to the signal names, respectively.

Based on the above features, the second embodiment will be described with reference to FIG. 10 and subsequent figures.

The conversion unit 208 inputs R information in the order of R ₁ , R ₂ , R ₃ , R 4 from the terminal 205, inputs information in the order of G ₁ , G ₂ , G ₃ , G ₄ from the terminal 206, , B information is input in the order of B ₁ , B ₂ , B ₃ , B ₄ .
The conversion unit 208 further separates brightness information L ^* and color information a ^* , b ^* for each pixel from the input R, G, B information and outputs them. 209, 210, 211 are L ^* , a ^* , b in block units
^This register temporarily stores the value of ^* . These registers are serial input → parallel output.

<Encoding of Brightness L ^* > Encoding of brightness information, that is, L * information will be described with reference to FIG. First, the operation of 277 to 282 in the circuit of FIG. 12 will be briefly described. The average calculator 277 is
As in Y11 to Y22 input in FIG. 5 in the first embodiment, L * information L _p (p = 1 to 4) for four pixels is input, the sum of these input values is calculated, and the L * information Average value without block
Output m _L.

Subsequently, the subtractor 278, the S _{p (p} = 1~4) minus the block mean m _L from the information L _p of each pixel obtained as S _p = _{L p} -m L. ROM279, average value calculator 280, ROM281, Is obtained, which represents the standard deviation value within the block.

Finally, normalization of L * information L _p (p = 1 to 4) will be described. The encoder 282 calculates and outputs the standardized code values C _{L1 to} C _L14 in accordance with C _Lp (p = 1 to 4) = (L _p −m _L ) / σ _L as in the first embodiment. .

The output of the average value calculator 277 has an 8-bit information amount as an average value m _L of L ^* information of four pixels. Therefore,
The shifter 300 receives the average value m _L as 8-bit information, shifts it by 3 bits, quantizes it into 5-bit information, and adds 2 bits of “0” to the lower order. Input to one input. The other input of the selector 287 is an average value m _L. The selector 287 uses these two inputs as the logical sum of the a ^* edge signal (S _a described later) and the b ^* edge signal (S _b ) output from the OR circuit 214 (FIG. 10) input from the terminal 289. Selection is made by a certain select signal SEL. That is, SEL = S _a + S _b, the output of the selector 287: 5 bits m _L (SEL = 1 ... color edge there) or 8-bit m _L (SEL = 0 ... Mu color edges).

Similarly, the standard deviation value σ _L in the block output from the ROM 281 also has a 7-bit information amount.
1 quantizes this σ _L to 4 bits and adds 3 bits “0” to the lower bits. That is, the output of the selector 288: 4 bits σ _{L (SEL} = 1 ... color edge there) or 7 bits σ _{L (SEL} = 0 ... Mu color edge), and outputs the sigma _L from the terminal 291.

<Edge Detection> FIG. 13 is a configuration diagram of the determination units 212 and 213. Judgment unit 212,2
13 is equivalent to the circuit configuration itself,
Will be described.

The determination unit 212 checks whether the color of an edge in a block from a ^* of the information in the block, and sends it as an edge signal S _a for a ^*. As a guide for checking for the presence or absence of a color edge, the determination unit 212 uses the color plane a ^* b ^*
For ^{the a *} axis in the plane, using a difference between the maximum value and the minimum _{value, E a = Max (a 1} a 2 a 3 a 4) -Min (a 1 a 2 a 3 a 4). Ea, which is _a measure of the presence or absence of the color edge, is defined as the color edge amount. The maximum / minimum unit 228 generates a maximum value Max (a ₁ , a ₂ , a ₃ , a
_4), and outputs the minimum value _{Min (a 1, a 2,} a 3, a 4), the subtracter 229 outputs the edge amount E _a. The comparator 230 compares the threshold T4 previously _{determined, S a = 0 (T4>} Ea) ... a * without color edge _{S a = 1 (T4 ≦ Ea} ) ... a * color edge there to become color The edge signal _Sa is output.

Similarly, the determination unit 213 determines whether or not there is a color edge in the block based on b ^* information in the block, E _b = Max (b ₁ b ₂ b ₃ b ₄ ) −Min (b ₁ b ₂ b ₃ b) _{4) S b = 0 (T5} > Eb) ... b * color edge without _{S b = 1 (T5 ≦ Eb} ) ... b * output color edge signal S _b to be there color edges. SEL = S _a + of the S _b are as described above. Thus, SEL is set to "1" when either one of the S _a and S _b is "1", one of the S _a or S _b even in this case also be a "0" Note that you get This relates to the color encoding result of the second embodiment described later (see FIG. 17).

<Color gradation information> block average value calculator 215 and 216 the average value m _a or m _b of a ^* or b ^* information in the block. That is, Is calculated and output.

<Color Edge Distribution Detection> FIG. 14 shows the configuration of the binarization units 218 and 219. Binarization unit 21
8 and 219 are equivalent to each other in terms of circuit, and therefore, the binarization unit 218 that binarizes the a ^* information will be described as an example. This binarization unit expresses the distribution of the two divided areas by binary codes C _{a1 to} C _a4 when a block has a color edge and the block is bisected by the color edge. . That is, for example, if _Ca1 is “1”, it indicates that the color a ^{* of the} pixel X1 belongs to the area of the larger color value in the block.

Referring to FIG. 14, the comparator 238 includes a _{1 to} a ₄
By comparing the respective and block average value m _a of, when the comparison result as B _a1 .about.B _a4, against i = 1 to 4, (when _{ai> m a) B ai =} 1, or 0 (ai <m _a ) is output. These B _{a1 to} B _a4 are 2-input AND gates 239 to 24
Entered in 2 respectively. The other input of the AND gate is a ^*
The edge signal _Sa is input. The logical product of B _ai and S _a of each pixel is obtained. That is, when S _a = 1, that is, when a color edge exists in the block, the AND gates 239 to 242 convert the binarized values B _{a1 to} B _a4 of the information of each pixel into the binarized information of the a ^* information.
Output as C _{a1 to} C _a4 . Conversely, when S _a = 0, that is, when no color edge exists in the block, “0” is output as C _{a1 to} C _a4 . These C _{a1 to} C _a4 are used to calculate the average color in two regions divided by the color edge.

<Extraction of Intra-region Color Information> FIG. 15 is a configuration diagram of the averaging units 220 and 221. Average part 220,2
Since 21 has the same configuration, the averaging unit 220 will be described. The inputs to the averaging unit 220 are the above-mentioned binarized information C _{a1 to} C _a4 and the a ^* information _a1 to _a4 of each pixel. Enter Counter 255 is C
The number of "1" s in _a1 to _Ca4 is counted. Counter 265 is C _a1
Count the number of "0" of ~ _Ca4 .

Counter 255, AND gate 256-259, average value calculator 260
Is when there is a color edge (S _a = 1) in the block, sieved only pixel a ^* information is larger than the average value m _a, calculate the average m _a1. This part corresponds to _ma1 in FIG. When the color edge is not in the block, Ca
Since i is “0”, the average value _ma1 is “0”. Average value m
_a1 has an information amount of 8 bits. Meanwhile, counter 265, AND
Circuit 266-269, an average value calculator 270 outputs the average value m _a2 of a ^* information of the pixel a ^* information is smaller than m _a.

Thus, the averaging unit 220 outputs m _a1 (8 bits) and m _a2 (8 bits). The color of the edge when not in the block, the mean value m _a1 is "0", the mean value m _a2 is equal to the average m _a of the entire block a ^* information.

The average unit 221 also outputs the m _b1, m _b2, likewise, when the color of the edge is not in the block, the mean value m _b1 is "0",
Average m _b2 is equal to the average m _b of the entire block b ^* information.

The selector 222 will be described. The selector 222 converts the outputs C _{a1 to} C _a4 and C _{b1 to} C _b4 of the binarization units 218 and ₂₁₉ into edge signals S
_a, is Yotsute selected combination of S _b. The output C ₁ -C ₄ is as follows.

When S _a = S _b = 0, C _{1 to} C ₄ = C _{b1 to} C _b4 (however, from the results of the above-described binarizer of C _{b1 to} C _b4 , all “0”
When S _a = 0 and S _b = 1, C _{1 to} C ₄ = C _{b1 to} C _{b4 When} S _a = 1 and S _b = 0, C _{1 to} C ₄ = C _{a1 to} Ca ₄ S _a = S _b = 1, C ₁ -C ₄ = C _a1 -C _a4 (However, since the two-color distribution in the block of a ^* and b ^* does not differ greatly, it is considered that the distribution follows the distribution of a ^*. Is).

From the above, the selector 222 looks at only the value of S _a and
_{1 to} C ₄ may be selected from C _{b1 to} C _b4 or C _{a1 to} C _a4 .
See the table in FIG.

Obtained in <Synthesis> _{above, SEL, m L, σ L} , C L1 ~C L4,
S _a , S _b , m _a1 , m _a2 , m _b1 , m _b2 , and C _{1 to} C ₄ are input to the combiner 223.

As described above, the output of the selector 287 (FIG. 12) is 5 bits m _L (SEL = 1... With color edge) and 8 bits m _L (SEL = 0... Without color edge). Is 4-bit σ _L (SEL = 1... With color edge) and 7-bit σ _L (SEL = 0... Without color edge). Therefore, first, the synthesizer 223 outputs m _L , σ according to the SEL value.
The bit length of _L is selected and shift-combined (see FIG. 16). C _{L1 to} C _L4 are output from the synthesizer 223 as they are.

For the synthesis of m _a1 , m _a2 , m _b1 , m _b2 , the synthesizer 223
Synthesis is performed as shown in FIG. 17 according to the values of S _a and S _b . Here, m _a1 , m _a2 , m _b1 , and m _b2 are originally 8 bits. For example, when there is an edge at a ^* (S _a = 1), the upper 4 bits of m _a1 and m _a2 are output. Is done. Alternatively, for example, when b ^* has no edge ( _Sb = 0), _mb2 is output as _mb (8 bits) as it is.

In this manner, the information and the information, that is, the color information and the brightness information are each a code whose information amount changes depending on the presence or absence of the color edge. As is clear from FIGS. 16A and 16B, the ratio occupied by the brightness information and the color information in the code changes depending on the presence or absence of the color edge in the block.

<Decoding> Finally, the decoding of the second embodiment will be briefly described. No.
After receiving the encoding as shown in FIGS. 16A and 16B, the SEL of the most significant bit is referred to.

First, when SEL is “0”, decoding is performed assuming that there is no color edge in the block. Assuming that L ^* information after decoding of each pixel is L ₁ ′, L ₂ ′, L ₃ ′, L ₄ ′ for each corresponding pixel,
It is expressed by the following equation.

L _i ′ = M _L + σ _L × C _{1i Further} , a ^* ′ and b ^* ′ after decoding are represented by the following equations.

_{^{a * '= m a b *}} ' = m b of each pixel thereof decoded L ^{* ',} a ^*', when coding the b ^{* ',} performs inverse transformation of the transformation performed by the converter 208, decoding The obtained R, G, B information R ', G', B 'is obtained.

If SEL is "1", decoding is performed as if there is a color edge in the block.

Similarly to SEL = 0, L ₁ ′ to L ₄ ′ are L _i ′ = m _L + σ _L × C _1i .

Regarding a ^* information and b ^* information, reference numerals in FIG.
It is determined according to the example of FIG. 17 from the combination of the S _a bit and the S _b bit. That is, when _{S a = 0, S b =} 1, when _{a i '= m a b i} ' = m b1 × C i + m b2 × C i S a = 1, S b = 0, a i '= m _a1 × C _i + m _a2 × C _i b _i ′ = m _{b When} S _a = 1 and S _b = 1, a _i ′ = m _a1 × C _i + m _a2 × C _i b _i ′ = m _b1 × C _i is a + m _b2 × C _i.

L ^* , a ^* , b ^* information L ^* ,
a ^* , b ^* are R, G, B decoded in the same manner as when SEL = 0.
Information R ', G', B 'is obtained.

<Modification of Second Embodiment> In the second embodiment, the encoding of the brightness information is realized by the average value, the standard deviation, and the binary code in the block. Of course, a difference value or the like may be used. Further, the encoding of the brightness information may be performed by another encoding method, for example, vector quantization.

In the second embodiment, the brightness, Y, I, Q information is used as color information, but an information system that separates brightness information and color information,
For example, the L ^* , a ^* , b ^* space defined in CIE1976, or the Y, I, Q space generally used in TV or the like may be used.

The size of the pixel block is 2 × 2 in this embodiment, but may be 2 × 2 or more. It is also possible to replace a part of each component block or circuit in the embodiment with a predetermined ROM.

In the embodiment, the plurality of colors are two, but if the size of the block is large, the number of colors is three or four. Therefore, the number of colors may be multi-valued by the number of colors considered to be included in the block depending on the size of the block, and the color information of each color may be increased at the same time.

In the second embodiment, the color distribution information is included in the code separately from the brightness resolution information. However, it is possible to substitute the color distribution information with the brightness resolution.

Further, the comparison with the threshold value for judging the presence / absence of a color edge is performed using a ^* , b ^*.
Separately but having conducted, a ^*, information that integrates information of b ^*, for example, a ^* of the maximum in the block, a minimum a _max, a _min, b ^* of the maximum, minimum b _max, from b _min {(a _{max -a min) + (b max} -b min)}, or, is also considered to use such ^{_{{(a max -a min) 2}} + (b max -b min) 2} 1/2.

<Effects of Second Embodiment> As described above, a color image is divided into blocks each including a plurality of pixels, and is decomposed into brightness information and color information in block units. Edge information from color information, and based on this information, switch the code length when brightness information and color information are individually variable-length coded, and change the resolution to color information as necessary. By providing information, deterioration such as color bleeding can be suppressed, and high-efficiency color image signals suitable for human visual characteristics can be encoded.

In addition, since fixed-length encoding is performed in block units, there is an effect that the positional relationship of pixels in an image can be maintained when the pixels are stored in an image memory. Therefore, the required amount of image memory can be uniquely determined by the size of the image. In addition, even when performing image processing on this color image, it is possible to easily obtain information on peripheral neighboring pixels,
In addition, since processing can be performed in units of blocks, there is an effect that image processing can be speeded up.

[Effects Common to First Embodiment and Second Embodiment] The encoding apparatuses disclosed in the first embodiment and the second embodiment can be summarized as follows: A-1: First Embodiment to Second Embodiment In common, a fixed-length codeword indicating that an edge has been detected in a block is generated, and the brightness information and the color information each have a predetermined bit length of a different length depending on the presence or absence of an edge. A configuration for encoding into codes is disclosed.

With this configuration, the brightness information (Y or L ^* ) and the color information (IQ or a ^* b ^* ) are converted according to the degree of the edge in FIG. 9A, FIG. 16A, or FIG. 9B or FIG. 16B. In this way, the bit length (quantization density) is encoded to be short (coarse). For this reason, the compression efficiency is increased, and at the same time, a compressed image matching the human visual characteristics can be obtained.

Fixed length code word which indicates that the edge is detected, for example, in the first embodiment has a fixed length of sigma _Y, in the second embodiment corresponds is SEL. This is because in the second embodiment, σ _Y is encoded (quantized) into two lengths in accordance with the presence of an edge, so that a SEL is required.

If the total length of the encoded code corresponding to one block of color image data (43 bits in the first embodiment, 36 bits in the second embodiment) is fixed regardless of the presence or absence of an edge, as described above. When storing in a simple image memory, etc.,
There are effects such as the positional relationship between pixels in an image can be maintained.

A-2: Edge detection includes detecting a brightness edge from brightness information (first embodiment) and detecting a color edge from color information (second embodiment).

A-3: The brightness portion of the coded code has a standard deviation (σ _Y ), an average brightness (m _Y ), and a gradient (N _ij or C
_ij ) is encoded (quantized) to a predetermined length, and the color portion is a color average value (m _I , m _Q , m _a , m _b ) or a color gradient (E _I , E _Q , m _a1 , m
_a2 , _mb1 , _mb2 ) are encoded (quantized) to a predetermined length.

B-1: In the second embodiment, a fixed-length codeword indicating that an edge is detected in a block is generated, and when an edge is detected, brightness information (Y, L ^* ) and color information are generated. Mean values m _I , m _Q , m _a , m _b ) and color gradients (E _I , E _Q , m _a1 , m _a2 , m _b1 ,
m _b2 ) is encoded into a code word of a predetermined bit length having a different length depending on the presence or absence of an edge.

With this configuration, information on edges is efficiently compressed and stored, and the decoded image also matches human visual characteristics.

C-1: Common to the first and second embodiments, a fixed-length codeword indicating that an edge has been detected in a block is generated,
A configuration is disclosed in which information indicating a distribution configuration in a block of color image data and an average value of the color image data in each distribution region are encoded into codewords having predetermined bit lengths of different lengths depending on the presence or absence of an edge. Have been.

With this configuration, information about edges is efficiently compressed and stored, and the decoded image also matches human visual characteristics.

C-2: In the second embodiment, block mean value m _a color information, calculates the m _b, the color distribution in the block from the difference between the average value and the color information values of the pixels C _az -C _a4 , C _{b1 to} C _b4 are detected, and the average value m _a1 , m _a2 , m _b1 , m _b2 of the color information of each area corresponding to this color distribution is detected.
Is calculated.

[Effects of the Invention] As described above, according to the present invention, a data unit including brightness information, color information, and classification information added to these information is set to a predetermined amount of data, so that the When encoding the brightness information and the color information, the encoded color image data can be encoded into a data unit that is easy to handle, for example, when decoding the encoded color image data.

In particular, in order to perform efficient encoding using the type of the image to be encoded, for a block including an edge, color information is encoded with a long code length code and brightness information is encoded with a short code. For blocks that do not contain edges, color information is coded with a code with a short code length and brightness information is coded with a long code length, so efficiency is reduced as much as possible to minimize image degradation. Encoding can be performed.

[Brief description of the drawings]

1 is an overall configuration diagram of an encoding circuit according to a first embodiment, FIGS. 2 and 3 are diagrams illustrating pixel blocks used in the first and second embodiments, and FIG. 4 is a first embodiment. FIG. 5 is a block diagram showing the configuration of an arithmetic unit according to the first embodiment; FIG. 6 is a block diagram showing the configuration of an encoder 11 according to the first embodiment; 7 is a block diagram showing the configuration of the encoder 12 of the first embodiment, FIG. 8 is a block diagram showing the configuration of the synthesizer 13 of the first embodiment, and FIGS. 9A and 9B are diagrams of the first embodiment. FIG. 10 is a block diagram showing the configuration of an encoded code format, FIG. 10 is a block diagram showing the entire configuration of the second embodiment, FIG. 11 is a diagram illustrating how blocks are divided by edges in the second embodiment, FIG. 12 is a block diagram showing the configuration of the L ^* encoding unit 217 of the second embodiment, and FIG. 13 is a block diagram of the second embodiment. FIG. 14 is a block diagram showing a configuration of the binarizing unit 218 of the second embodiment, FIG. 15 is a block diagram showing a configuration of the averaging unit 220 of the second embodiment, 16A and 16B are block diagrams showing the configuration of the format of an encoded code in the second embodiment. FIG. 17 is an example of the encoded format of color information according to the presence of a color edge in the second embodiment. FIG. In the figure, 8,9,10 ... Calculator, 11,12 ... Encoder, 13 ... Synthesizer, 2
12,213 …… Judgment unit, 215,216 …… Block averager, 217…
.. L ^* encoding unit, 220,221... Average unit, 222.

Claims (1)

(57) [Claims] 1. An image processing apparatus for processing a color image divided into a plurality of blocks, comprising extracting a feature amount indicating a distribution of pixel values of each block based on color information of the divided blocks. In accordance with the feature amount, each block is divided into a first block including a predetermined amount of edges in the block and a second block different from the first block.
Classifying means for classifying the divided blocks into blocks; coding means for coding brightness information and color information of the divided blocks; and brightness information and color information coded by the coding means. And an adding unit for adding classification information indicating a classification result by the classification unit, wherein the encoding unit encodes color information with a code having a long code length for the image of the first block. And the brightness information is coded with a short code length, and for the image of the second block, the color information is coded with a code having a short code length and the brightness information is coded with a long code length. The adding unit converts a data unit including the brightness information and the color information of the divided block encoded by the encoding unit and the classification information added to the information into a classification result of the block by the classification unit. To An image processing apparatus characterized in that a predetermined data amount is set regardless of the data amount.