JPH04183193A - Image processing device - Google Patents

Abstract

PURPOSE:To reduce data quantity maintaining the quality of original image by converting a brightness signal and color signal into predetermined representative values corresponding to a value of each color signal and by extending the difference between the representative value and the value obtained before conversion for successive unconverted picture elements. CONSTITUTION:Let the upper 4 bit data of a signal Y0 be Y1 and apply data compression to the data, and taking 4 representative points for I0 and Q0 to quantizing those points, and let these representative points be I1 and Q1, where I1 and Q1 be 2 bits each. Arrange Y1, I1, and Q1 into one-byte data and pack them for each picture element. Further, by an error detecting circuit 308, a quantized error of each picture element is calculated, by the extension matrix the error is two-dimensionally extended, and the error extended by a memory section 309 is accumulated. That is, data that is reduced by bit to Y1, I1, and Q1 is extended two-dimensionally for the error between the original image quality and the converted image quality, the extended amount is fed back, so when a two-dimensional area becomes large, the difference between the original image quality and the converted image quality is eliminated.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image processing device, and for example, to an image processing device that reduces the amount of color image data.

[Prior Art] Conventional color image data compression uses orthogonal transformation and encoding to reduce redundancy in images.

Compression was performed based on image correlation, etc.

[Problem to be solved by the invention] However, since conventional compression methods are based on orthogonal transformation of two-dimensional image data, correlations between three various components, and encoding methods, it is difficult to perform image editing. Sometimes, compressed data is restored once, a restored image is created, image editing processing such as affine transformation is performed, and the processed image data is compressed again.

This requires time for conversion and costs due to the complexity of the circuit, which poses practical difficulties.

Therefore, compressed data is handled only during transmission, and the above compression method is not applied to images that are intended for editing. That is, there is a drawback that image data compression suitable for image editing cannot be performed.

The present invention has been made in view of the above-mentioned drawbacks of the conventional example, and its purpose is to provide an image processing device that reduces the amount of data while maintaining the image quality of the original image.

[Means for Solving the Problems] In order to solve the above-mentioned problems and achieve the purpose, the image processing device according to the present invention converts the primary color signal of each pixel into a luminance signal and a color signal for processing. In the image processing device, a conversion means for converting the luminance signal and the color signal into predetermined representative values corresponding to the values of each signal, and detecting an error between each representative value converted by the conversion means and a value before conversion. and a diffusion means for diffusing the error detected by the detection means to subsequent pixels before the conversion by the conversion means is performed.

[Operation] According to this configuration, the converting means converts the luminance signal and the color signal into predetermined representative values corresponding to the values of each signal, and the detecting means converts each representative value converted by the converting means and the value before conversion. The diffusion means diffuses the error detected by the detection means to the subsequent pixel before the conversion is performed by the conversion means.

[Embodiments] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

<First Embodiment> FIG. 1 is a block diagram showing a first embodiment of an image processing apparatus according to the present invention, and FIGS. 2A and 2B are diagrams explaining human visual characteristics.

In FIG. 1, 101 is a color scanner that reads color image data and separates it into three different types.

Color image data R, G by photoelectric conversion element.

Then, the photoelectric conversion analog signal is digitally sampled to digitize the color image signal. The image signals to be pressed are R (red), G (green), and B.
The signal is output as three color components (blue), and three signals separated into colors using, for example, NTSC spectral characteristics are output. 102 is RGB
/YIQ converter, which is the Y of the three-color R, G, and B signals output from the color scanner 101. ,■. Conversion of the Qo multiplied signal, that is, the already known Yo
Conversion to (luminance signal) Io, Q, (color signal) is performed.

103 is a concentration preservation/YIQ reduction unit which stores Yo, 1.
,Q. The signal locally is color information.

Control and reduce the brightness component so that it is the same as the brightness component, Y,
,I,,Q,outputs the signal. 104 is a 1-byte backing unit which stores Y, , 1 . , Q, back up the signal by 1 byte. 1
05 is a disk that stores the 1-byte backup created by the 1-byte backup unit 104. A 1-byte back decomposition unit 107 decomposes the density signal backed into 1 byte into Y', I', and Q' image signals. 108 is YIQ
/RGB conversion unit converts the Y', I', Q' image signals into R', G', B'' signals. 109 is a color printer, and the YIQ/RGB conversion unit 108 converts the R', I', Q' image signals into R', G', B'' signals.
', G', B' image signals are input to form a color image. Reference numeral 110 denotes a monochrome printer, which inputs the Y component, that is, the Y° image signal before conversion, from the YIQ/RGB conversion unit 108 to form a monochrome image. A CPU 106 controls general operations from the color scanner 101 to the color printer 109 or monochrome printer 110.

As shown in FIGS. 2A and 2B, human visual characteristics can be expressed by the number of gradations that the human eye can discern (vertical axis) and the image spatial frequency of the original image (horizontal axis).

The solid line curve is a table depicting the visual characteristics of the human eye based on the vertical and horizontal axes.

What can be concluded from both figures is that the lower the spatial frequency of an image, the more gradations per pixel are required (Figure 2A).
For high-resolution images, the number of gradations is small (Fig. 2B). For example, when the resolution is about 10 line pairs/mm, the number of gradations is about 2 gradations.

That is, it can be seen that a 20 pel printer can produce a sufficiently good image quality with a 20 pel spatial image if the binarization method is adapted to human vision.

Figure 2A shows a binary recording with a resolution of 12 da.
The image quality estimation curve for the human eye is shown when varying from t/mm to 60 dots/mm. In a binary recording printer, if the resolution is not about 48 to 60 dots/mm, the resolution will be below the characteristics of the human eye, and it is estimated that good image quality reproduction cannot be obtained. On the other hand, since it is difficult to manufacture a 60 dat/mm printer for technical reasons, there is an idea of increasing the gradation per pixel and suppressing the increase in resolution.

Figure 2B shows a resolution of 20 dots/m.
This is an example in which gradation is changed using m. What can be estimated from this is that at a resolution of 20 dots/mm, 2bi
The gradation of t to 3 bits, that is, the gradation of 9 to 4 gradations, satisfies the characteristics of the human eye, and the output image quality is sufficiently satisfactory for the human eye. estimated to be obtained.

Color printers currently in practical use are 400 dpi.
(About 16 pels, ie, 8 c/mm) There is a product that can perform multi-value recording, and the following explanation uses this as an example.

At 400 dpi, the brightness is about 4 bits (166 gradations satisfies the human visual characteristics, so the brightness is 4 bits.The color component is NTS By using this, data can be reduced.In other words, if the area is small, the visual characteristics will be less sensitive to color information, and only the brightness component will be left.Therefore, if the brightness component is only 4 bits, the color component will be smaller than that. It is estimated that a small number of bits is sufficient.

However, even if the number of color component bits for each pixel is reduced, the color components of the original image must be reproduced when the area becomes large.

Therefore, the output Y,,I of the RGB/YrQ converter 102.

, Qo uses a value calculated with high precision using a floating point or inch jar type, and the concentration preservation/YIQ reduction unit 103 calculates Y,
, 1. , Q, to reduce the number of bits.

Signal Y. Regarding, compress the data by compressing the upper 4 bits of data with Y. and Qo are quantized using four points as representative points, and the position information (quantized data) of the representative points is given to I,,Q,. Here, I, , and Q are each 2-bit values.

Above Y, 1. ,Q, into 1 byte data,8
Bit data binding 104 backs up each pixel. Therefore, the backed-up 1-byte data is formed of 4 bits for the Y component of the brightness signal, 2 bits per minute for the color component, and 2 bits for the Q minute, for a total of 8 bits. In this form, if one pixel and three colors are represented by one byte data, it is easy for a computer to handle (Affine transformation etc. can be processed in exactly the same way as conventional black and white data processing. Also, the memory of the disk 105 Even if the image is stored in the image, the memory capacity is reduced to 1/3, which is the same as that of a black and white image.

On the other hand, Y,1. R, G, B equivalent to original image data from Q signal
To restore the signal, the 1-byte back decomposition unit 107 converts Y
, I and Q components are separated, and for I and Q, the representative point data is returned from the representative point position information, and from that value R", G
', B' signals are calculated. Calculated R', G', B
' If you print 8 color images on the color printer 109 using ``, the same output as the original image will be obtained.

In addition, when using the black and white printer 110, YIQ/
By using only the Y' component of the RGB converter 108, it can be used as monochrome print output data of the monochrome printer 110.

The outline of the overall configuration of the present invention has been explained above, but next,
A method for reducing the number of bits while preserving density, which is the main part of the present invention, will be explained in detail.

FIG. 3A is a block diagram showing the configuration of the density storage/YIQ reduction unit 103 according to the first embodiment.

Here, density preservation is a method of locally controlling color information and brightness components to be the same, and reducing the amount of YIQ data in a manner that particularly suits human visual characteristics.

In FIG. 3A, 301 is the NTSC R from the input section.
GB3 primary color signal is shown, and 302 is RGB3 primary color signal 30
1 shows a color component conversion table for converting 1 into a YIQ signal.

305 is a quantization unit, which has 16 stages for the Y signal;
Threshold values are set in four stages for I and Q signals, and four levels are set for Y signal.
The I and Q signals are quantized to 2 bits. Examples are shown in the table below. Here, the threshold is -63, 1, 6
4 is used. For example, if the input ii signal is -50, the sign will be l and the representative value will be assigned -32. These codes are designated as Yl, If, and Ql by the quantization unit 305.
is output from. The Yl, It, and Ql signals that have been divided into
The error detection circuit 308 calculates the error of the V 3' + 1i+ qq multiplied signal before quantization, and calculates the error y, i.

q is stored in the memory unit 309. That is, Y, I,
The Q signal has a quantization error of 3'+1+ due to the reduction in the number of bits.
It is calculated for each Q, and is fed back to and added to the Y, r, and Q signals input sequentially later in the adder 304. Therefore, Y, I.

The variables corresponding to Q are yy+ ii+ qq, and these are expressed by the following formula (1). In other words, it becomes .

From this value, for the brightness component, yy (0≦yy≦2
The upper four bits of 55) are set as Yl, and for the optical signal among the color difference components, since ii satisfies -1160≦ii≦130, the codes according to the table below are selected. Since there are four types of codes, O to 3, 2 bits are sufficient.

Regarding the color difference component Q signal, qq is -104≦qq≦60
Therefore, the encoding ratio will be the same. Therefore, 2 bits are sufficient for the Q component as well.

306-1. ,Q, will send this code.

Reference numeral 307 is a representative value conversion circuit that converts the YIQ component into a representative value, and the Yl signal is the same 4-bit data, and the same 4 bits are added to the lower order, and the result is output to the error detection circuit 308 as a total of 8 bits. For example, yy is 37. (=55), it becomes Y 1 '' 3 s. Therefore, the output of the representative value conversion circuit 307 becomes 33 (=51).

On the other hand, here, the same processing is performed for ii and qq, so to explain about if, 1i=-20" F F
From E Cs, the sign is 1, so I+=1, the representative value is -32, and the output of the representative value conversion circuit 307 is -32=
FFEO, and enters the error detection circuit 308.

On the other hand, 308 is an error detection circuit that calculates the difference between y3'l l 11 qq and the output of the representative value conversion circuit 307. Letting the errors of each component be yerr, 1err, and qerr, and expressing it by the following equation (2), Become.

In the previous example for the Y component, yerr=55-51=+4
, 1err=-20-(-32)=+12.

FIG. 4 is a diagram showing an error diffusion table provided inside the error detection circuit 308. This error diffusion table is used to two-dimensionally diffuse the errors yerr, 1err, and qerr.

Diffusion error amount of brightness e r r o r [yl [X
] is expressed by the following equation (3), it becomes...(3). Similarly, the I and Q components are calculated according to the above formula.

A memory unit 309 stores e r r or r [yl [x] obtained by the above equation (3).

Therefore, the y output is error [01[2] from the adder 3.
It is added to Y in 04. In this way, the error detection circuit 30
8, the quantization error of each pixel is calculated, and the error is two-dimensionally diffused using the diffusion matrix shown in FIG.
The errors diffused by 09 are cumulatively added.

Reference numeral 310 denotes a transmitter, which transmits (transmits) each data of Yl, Il, and Ql in a compressed form of 4.2.2 bits. In other words, R1 and Gl of human power.

Since the total 24-bit data (8 bits each) becomes 8 bits, it is compressed to 1/3. 311 is 1 above
This is a receiving unit that receives data compressed to 31.
The receiver receives data in the same format as that sent with 0.

312 is an expansion unit which converts the 4-bit data of Yl into 8-bit data Y° (upper 4 bits and lower 4 bits).
Since the IQ multiplied signal is sent with codes 0 to 3, it is returned to 2-bit data I', Q' using the representative values in the table on page 11, and pseudo-8bit data similar to the Y component is sent.
Convert it to a signal of t. Reference numeral 313 is a color component conversion table, which converts YIQ signals back to RGB signals. Therefore R'
The G'B° signal is an expanded image signal, and the receiving side uses this signal to print or display.

According to the above explanation, the data whose bits have been reduced to Yl, II, and Ql will two-dimensionally diffuse the error with the original image signal, and the amount will be fed back, so if the two-dimensional area becomes large, the difference from the original image will increase. It has the property of disappearing. In other words, the color difference signal has a large variation from the original image on a microscopic level, but when it comes to a large two-dimensional area, the signal difference from the original image becomes small.This is due to the visual characteristics of the human eye. It can be used as a method.

Further, FIG. 3B shows a modification of the circuit shown in FIG. 3A described above.

In this figure, each component is similar to that in FIG. 3A, except that errors are detected using r, g, and b signals before being converted into Y, I, and Q signals. Therefore, after the representative value conversion circuit 307, the 3'a+to+Qo signal is
A conversion circuit 308° for converting the q and bb multiplied signal color components is provided.

Further, FIG. 3A differs from FIG. 3A in that the weighting provided inside the error detection circuit 3°8 is provided at 308°.

By adopting the circuit configuration shown in FIG. 3B, r5g,
It is also possible to detect conversion errors when converting the b signal to Y, LQ signals, thereby further improving image quality.

Second Embodiment In the second embodiment, an example will be shown in which all the operations described in the first embodiment are realized by software. This is CPU10
6 or another CPU. In either case, it is assumed that a program according to the flowchart described below is stored in a ROM or the like. Of course, it is assumed that a RAM used as a work area for the above program is also provided.

Figures 5 (a), (b), and (c) are from RGB to YI of the original picture.
This is a flowchart explaining the operation of converting the Q signal to 8 bits, and FIGS. 6(a) and 6(b) are flowcharts explaining the operation of converting the 8-bit YIQ signal to RGB 24-bit data. In addition, here, FIG. 5(a),
The C language is used for the mathematical expressions in (b), (c) and FIGS. 6(a) and (b).

First, a method for compressing 24-bit color data of an original image into 8-bit data will be explained using FIGS. 5(a), (b), and (C).

First, the r, g, and b data of the original image are EY and EI.

When converting to EQ data and compressing, EY, El
, reduce the number of EQ bits. In this case, for secondary averaging, an error diffusion buffer memory is prepared to diffuse errors caused by bit reduction so that the original image and density are preserved, and the buffer memory is cleared at the time of initialization (step S1). Next, the original 24-bit data is read (
Step S2). This read data r+g+b is converted into EY, EI, and EQ (step S3).

Next, in order to obtain a positive value for the EY acid component, data outside the range of 0 to 255 is corrected to fall within the range (steps 84 to S8). After this, the EI acid component is divided into four regions, namely: ■EI<-64, ■-64≦EI<O1■
0≦EI≦64, ■64<Ef) and is encoded into data 0.1 and 2.3 corresponding to the judgment regions ■, ■, ■, and ■, respectively (step 89 ~Step 515). Similarly to the EQ acid component, there are four regions: ■EQ<-64, ■-64≦EQ≦0, ■O<EQ≦
64, ■ 64 < EQ) and the data 0, 1 corresponding to the determination areas ■, ■, ■, ■, respectively.
, 2.3 (step S16 to step 5)
22).

Then, processing is performed to synthesize the EY, E, I, and EQ acid components and combine them into 8 bits (step S23, step 52).
4) Furthermore, errors 3' eyr HL srr yq and rr are calculated from the original image data EY, EI, EQ and the bit-reduced data (Yl, 11.Ql shown in the figure) (step 525). Error Y obtained by this calculation
, rr + ' rr + qsrr is 21
Bone-dimensionally diffused (3×5 pixels) and stored in an error diffusion buffer memory (step 826).

The process in step S26 is applied to all pixels in the X direction,
When 1942 minutes are completed (step 527), the data stored in the error diffusion buffer memory is shifted by one line (step 528).

Next, the third line of the error diffusion table is cleared (step 529), and the data (buft) compiled into 8 bits in steps S23 and S24 is stored in a file in RA rr (step 530).

All the operations described above end at the same time as the processing of all lines ends (step 531).

Next, Figures 6(a) and (b) show how to restore the 8-bit compressed color data to the original 24-bit color data.
Explain using.

First, one line of 8-bit data is read from a file in the RAM (step 561). EY, EI, and EQ acid components are extracted from the read-compared 8-bit data (step 562). First, the EI acid component and EQ acid component are returned to the representative value of the original El or EQ value (-96, = 32° 32 or 96) by the code 0, 1° 2 or 3 ( Steps 863 to 5672).

Then, Y, I, and Q are converted to r, g, and b (
Step 573), the converted data of r+g+b is set to a range of 0 to 255 (step S6).

As a result, the restore right to be put into said file)
is determined (step 378), and the restoration r for one line is determined (step 378).
g,b data are stored in a file (step 580
).

In this way, the present operation ends when all lines have been processed by the above process (step 581).

In this way, even when density preservation and YIQ reduction are performed using software, the amount of data can be reduced while maintaining the image quality of the original image.

In the above-described embodiment, the compressed data is stored on a disk, but it may also be communicated by facsimile over a public line or the like.

Color printers include laser beam printers, inkjet printers, and dot printers.

Such as a thermal transfer printer.

Furthermore, if the receiving (reproducing) side does not have a color printer 109, the black and white image is reproduced by the Ori printer 110 using only the luminance signal Y'' without converting it to an RGB signal on the receiving (reproducing) side. be able to.

[Effects of the Invention] As explained above, according to the present invention, there is an effect that the amount of data can be reduced while maintaining the image quality of the original image.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the first embodiment of the image processing device according to the present invention, FIGS. 2A and 2B are diagrams explaining human visual characteristics, and FIG. 3A is the density according to the first embodiment. 3B is a block diagram showing a modification of the circuit shown in FIG. 3A, and FIG. 4 is an error diffusion table provided inside the error detection circuit 308. Figures 5 (a), (b), and (c) show the RGB to YI of the original picture.
A flowchart explaining the operation of converting the Q signal to 8 bits.
3 is a flowchart illustrating an operation of converting into 24-bit data. In the figure, 101...color scanner, 102...RG
B/Y IQ conversion unit, 103...Concentration storage/YIQ
Reduction part, 1-04...1 byte pack part, 105...
・Disk, 106...CPU, 107...1 byte back decomposition unit, 108...YIQ/RGB conversion unit,
109... Color printer, 110... Black and white printer, 302... Conversion table, 305... Quantization section, 307... Representative value conversion circuit, 308... Error detection circuit, 309... Memory section, 310... Transmission section, 3
12... Decompression unit, 313... Conversion table.

Claims (1)

[Scope of Claims] An image processing device that converts and processes a primary color signal of each pixel into a luminance signal and a color signal, the luminance signal and the color signal being converted into a predetermined representative value corresponding to the value of each signal. a converting means for detecting an error between each representative value converted by the converting means and a value before conversion; and detecting means for detecting an error between each representative value converted by the converting means and a value before conversion; An image processing device comprising: a diffusion means for diffusing errors.