JP2557537B2 - Device for setting density signal conversion characteristics - Google Patents

Description

The present invention relates to an apparatus for setting density signal conversion characteristics in an image processing apparatus such as a color scanner for plate making.
In particular, the present invention relates to a device useful for always automatically setting an appropriate conversion characteristic without depending on the skill level of an operator.

[Conventional technology]

In color scanners for platemaking, the density range of the original image is matched with the density range in the color separation circuit, and at the same time, gray balance correction is performed to restore the color cast to gray, and a floor to emphasize the subject of the original image. Processing such as tone correction and color correction is performed. In the conventional color scanner, the color separation condition setting for realizing these processes has been performed based on a manual operation that depends on the experience of the operator.

[Problems to be Solved by the Invention]

Therefore, the proper color separation condition setting can be performed only by a skilled operator, and even the expert person has a slightly different setting content for each operator. For this reason, it is difficult for an operator with a low degree of skill to set an appropriate color separation condition, and there is a problem that the process for setting the color separation condition cannot be made objective and the process can be automated.

[Object of the Invention]

The present invention is intended to overcome the above-mentioned problems in the prior art, and can always set an appropriate density signal conversion characteristic without depending on the skill level and individual difference of the operator, and is particularly suitable for automating the setting of the conversion characteristic. The purpose is to provide a device.

[Means for solving the problem]

According to the analysis by the inventor of the present invention, the important factors in setting the density signal conversion characteristics can be roughly classified into two.
The first is a factor relating to the optical properties of the image of the original image, that is, the density distribution, the color distribution, and the like, and this does not depend much on what is shown in the original image. The second is a factor that is determined by what is the subject of the original picture (that is, the main subject), and the human factor such as the intention of creating the original picture is important there. Even if either of these two factors is missing, it is not possible to set an appropriate conversion characteristic.

Therefore, according to the present invention, the density signal conversion characteristic for converting the density signal obtained by reading the image of the color original image into a signal suitable for processing in a predetermined image processing apparatus is set in the image processing apparatus. (A) reference rule storage means for storing a reference conversion characteristic determination rule for determining the reference conversion characteristic of the density signal of each color component in an arbitrary color image through statistical analysis of the density distribution in the image. And (b) correction rule storage means for storing a correction rule predetermined for each subject with respect to the reference conversion characteristic corresponding to the classification of the subject that can appear in any color image, and (c) any color Subject determination rule storage means for storing subject determination rules for determining the subject of the color image from the statistical distribution of color components of the image;
Means for determining a reference conversion characteristic of a density signal for each color component of the original image by applying the reference conversion characteristic determination rule to the original image when a color original image to be processed by the image processing apparatus is given; (E) means for applying the subject determination rule to a statistical distribution of density signals of the respective color components of the original image to determine a subject in the original image; and (f)
Means for selecting a correction rule according to the subject in the original image from the correction rules stored in the correction rule storage means; and (g) the reference conversion characteristic for the original image according to the selected correction rule. And means for setting the density signal conversion characteristic obtained by this correction in the image processing apparatus.

It should be noted that the term "density" in the present invention is a general term that represents not only the optical density but also the signal level obtained by photoelectrically reading the density, the Munsell value, and other quantities that express the density of image gradation. Is a term.

[Action]

The standard conversion characteristic determined in the intermediate stage of the present invention reflects the statistical property of the density distribution in the original image. Therefore, this reference conversion characteristic is determined in accordance with a factor unrelated to the intention of creating the original image.

This reference conversion characteristic is corrected according to the subject of the original image.
The determination of the subject of the original image is automatically performed according to a predetermined subject determination rule. Then, the correction rule for each subject is also predetermined, and the correction rule corresponding to the subject of the original image is selected from among them. Then, the reference conversion characteristic is corrected according to the selected correction rule, and thereby the conversion characteristic to be actually set is determined.

The density signal conversion characteristic determining apparatus according to the present invention does not require a step depending on the skill of the operator. Therefore, this method can set the conversion characteristic objectively and systematically, and is particularly suitable for automating the conversion characteristic setting. The present invention is applied to color and color images, but since the number of parameters for setting color separation conditions for color images is greater than that for gradation curve setting for monochrome images, the present invention systematizes the setting process. The significance is particularly great.

〔Example〕

A. Overall Configuration and Schematic Operation FIG. 2 is a block diagram of a plate-making color scanner to which an embodiment of the present invention is applied. A positive color original image film 1 is wound around an original image drum 2, and a pickup head 3 is provided so as to face the drum 2. Rotation of the drum 2 in the α direction and the pickup head 3
The main scanning and the sub-scanning of the original image 1 are achieved by the translation in the β direction, and the image of the original image 1 is photoelectrically read by the pickup head 3 scanning line by pixel for each pixel.

The image information of the original image is input to the input circuit 4 for each color component of blue (B), green (G) and red (R).
And are converted into digital color density signals D _B , D _G , and D _R for each of B, G, and R in the input circuit 4.

These signals D _B , D are used during pre-scan described later.
_G and D _R are stored in the frame memories 13B, 13G, and 13R for each pixel. The information processing device 14 is, for example, a microcomputer, which takes in these signals D _B , D _G , and D _R from the frame memories 13B, 13G, and 13R and generates a normalization curve for each color component. However, the "normalization curve" is a process for converting the density signal for each color component obtained by reading the original image so as to match the density range of the color separation device, that is, a conversion curve used for normalization conversion. Is. Although details of this processing will be described later, the CPU 15 and the memory 16 are used in this processing. Further, the console 17 is for manually inputting various commands and data to the information processing apparatus 14. Then, the normalized curve created by the information processing device 14 is loaded into the look-up table memories (LUT) 5B, 5G, 5R as numerical data for each color component.

Next, a main scan of the original image 1 is performed for image recording.
The signals D _B , D _G , and D _R obtained at this time are subjected to normalization conversion by the LUTs 5B, 5G, and 5R to become signals D _NB , D _NG , and D _NR , which are given to the color operation circuit 6 in the next stage. In the color calculation circuit 6, the signals D _NB , D _NG ,
Color calculation based on D _NR is performed, and each color plate signal for Y, M, C, and K is generated. The respective color plate signals are converted into Y, M, C and K halftone dot signals in the halftone dot converting circuit 7, combined in a time division manner by the selector 8 and given to the output circuit 9.

The recording head 10 has a built-in laser light source and ON / OFF-modulates a laser beam by a modulation signal provided from the output circuit 9. The photosensitive film 12 is wound around the recording drum 11, and due to the combination of the α rotation of the drum 11 and the translation of the recording head 10 in the β direction, the exposure of the photosensitive film 12 by the laser beam is performed in scanning line sequential and It is performed for each pixel. In the following, it is assumed that the Y, M, C and K color plate images are positively recorded on the photosensitive film 12.

B. Normalized Curve Generation Operation FIGS. 1A to 1C are flowcharts showing the operation of the above-described color scanner. Further, FIG. 3 is a diagram conceptually showing a process of generating a normalization curve as one of the color separation conditions, and S1, S101, ... In FIG. Corresponds to the step number in Figure 1C. In addition, white arrows indicate data usage relationships.

In the following, the process of normalizing curve generation will be divided step by step.

(B-1) Specification of Reference Density Value First, in step S1 in FIG. 1A, the entire copy target portion of the original image 1 is prescanned. Then, the color density signals D _B , D _G , D _R (FIG. 2) for each pixel are stored in the frame memory 13B,
Store in 13G, 13R. Hereinafter, the color density values designated by these signals D _B , D _G , and D _R will be represented by the same symbols D _B , D _G , and D _R.
When the size of the original image 1 is large, the color density values D _B , D
This storing operation of _G and D _R is executed while thinning out pixels.

In the next step S100, the highlight side average color density values D _HAB , D _HAG , D _HAR and the shadow side average color density values D _SAB , D _SAG , D _{SAR are} calculated from these color density values D _B , D _G , D _R. This is a subroutine for obtaining and. This subroutine is described in detail in Japanese Patent Application No. 63-312684 previously filed by the applicant of the present invention, but only an outline thereof will be described here.

In FIG. 1B showing the details of step S100, in the first step S101, the density values D _B , D obtained in step S1
The average density value for each pixel is calculated from _G and D _R : D _M = (D _B + D _G + D _R ) / 3. Further, this calculation is performed for all the pre-scanned pixels, and the class indicating the range of the average density value D _M is set on the horizontal axis and the number of pixels N _P is set on the vertical axis, and the average density value frequency histogram is obtained as shown in FIG. To create. In FIG. 4, the median of the classes is shown by C _Mi (i = 1 to n).

In step S102, for each class of the average density value frequency histogram, each color component B, G, R of each pixel included in the histogram
The density values D _B , D _G , and D _{R for each} are cumulatively added. This process is performed independently for each class. Further, after performing this calculation, each class value D _Mi (i = 1 to n) is set on the horizontal axis, and the cumulative density values ΣD _B , ΣD _G , ΣD _R corresponding to the pixels included in each class are set on the vertical axis. , A cumulative density value histogram is created for each color component. FIG. 5 shows an example of this cumulative density value histogram, and such a cumulative density value histogram is obtained for each of X = B, G, and R.

As an example, class value D _Mi = 1.0, class width 0.1 (0.95
≦ D _M <1.05) will be described. The number of pixels included in this class is set to 3 for convenience, and the density data of each pixel is pixel 1: D _B = 1.10, D _G = 0.90 D _R = 0.95 (D _M ≈0.98) Pixel 2: D _B ＝ 1.00, D _G = 1.10 D _R = 0.90 (D _M ≈1.00) Pixel 3: D _B = 1.00, D _G = 0.95 D _R = 0.95 (D _M ≈0.97)

When the cumulative density value histogram shown in FIG. 5 is X = B, the cumulative density value ΣD _B in that class (0.95 ≦ D _M <1.05) is ΣD _B = 1.10 + 1.00 + 1.00 = 3.10. Similarly, the other cumulative density values ΣD _G and ΣD _R are also ΣD _G = 0.90 + 1.10 + 0.95 = 2.95 ΣD _R = 0.95 + 0.90 + 0.95 = 2.80.

Such processing is performed for each class, and each color component B, G,
Complete each cumulative density histogram for each R. (Third
See also the block of step S102 in the figure. Incidentally, in FIG. 3, each histogram is approximately drawn by a curve. Further, the processes of steps S101 and S102 described above are performed in parallel. ) In step S103, from the average density value frequency histogram shown in FIG. 4, the relative value RN (%) of the pixels cumulatively added from the lower density is plotted on the horizontal axis of the class value D _Mi (i = 1 to n). The cumulative relative frequency histogram as shown in FIG. 6 is created on the vertical axis. The histogram has a shape that changes from 0% to 100% within the range of the minimum and maximum generated density values D _Mmin and D _Mmax . However, in FIG. 6, this cumulative relative frequency histogram is approximated by a curve on the assumption that the class width is sufficiently small.

In the next step S104, a predetermined cumulative density appearance rate
RN _H and RN _S are applied to the cumulative relative frequency histogram shown in FIG. 6 to obtain the temporary highlight average density value D _MH and shadow average density value D _MS , respectively. The values of the cumulative density appearance rates RN _H and RN _S are values obtained in advance by analysis of a large number of sample original images so as to give statistically optimum highlight points and shadow points, and for example, 1
%, About 98%.

In step S105, in the cumulative density value histogram for each color represented in FIG. 5, as shown by hatching in FIG. 7, for the highlight side, the temporary highlight average density value D _MH or less. Area (D _Mmin ≤ D _M ≤ D _MH ), and for the shadow side, pay attention to the area (D _MS ≤ D _M ≤ D _Mmax ) above the provisional shadow average density value D _MS , and accumulate within that range. The pixel values of the density values ΣD _B , ΣD _G , and ΣD _R are averaged for each color component.

That is, if the pixel average on the highlight side is written as <...> _H and the pixel average on the shadow side is written as <...> _S , the highlight side average color density value D obtained by this pixel average is written.
_HAB , D _HAG , D _HAR and shadow side average color density value D _SAB , D _SAG , D
_SAR is D _HAX = <ΣD _X > _H (X = B, G, R) (1) D _SAX = <ΣD _X > _S (X = B, G, R) (2)

For example, X in FIG. 7 is B, and the temporary highlight density
When D _MH is D _M5 , D _HAB = (D _M1 + D _M2 + D _M3 + D _M4 + D _M5 ) / (N _P1 + N _P2 + N _P3 + N _P4 + N _P5 ) ... (3) However, N _Pi (i = 1 to 5) is the average density value D _M
Is the number of pixels belonging to the class D _Mi , and is obtained from the histogram in FIG.

In this way, the average color density values D _HAX and D _SAX on the highlight side and the shadow side for each color component shown in the center of FIG.
(X = B, G, R) is obtained. With the above, the subroutine corresponding to step S100 is completed, and the process returns to the main routine of FIG. 1A.

(B-2) Calculation of reference value In the next step S2 in the main routine, B, G, R
The highlight reference value D _H0 and the shadow reference value D _S0 common to each color component of are calculated according to the following equations (4) and (5).

D _H0 = FH (D _HAB , D _HAG , D _HAR ) (4) D _S0 = FS (D _SAB , D _SAG , D _SAR ) (5) However, the functions FH, FS are FH (D _HAB , D _HAG , D _HAR ) = r _H・ MIN (D _HAB , D _HAG , D _HAR ) + (1-r _H ) D _HF (6) FS (D _SAB , D _SAG , D _SAR ) = r _S・ MAX ( D _SAB , D _SAG , D _SAR ) + (1-r _S ) D _SF … (7), MIN (…): Minimum value selection operation, MAX (…): Maximum value selection operation, D _HF , D _SF : Standard standard highlight density value and shadow density value, r _H , r _S : 0 <r _H <1,0 <r _S <1, a preselected constant. That is, the highlight reference value D _H0 is D _HAX (X =
B, G, R) is the weighted average of the minimum value and the standard value D _HF , and the shadow reference value D _S0 is the maximum value of D _SAX (X = B, G, R) and the standard value D
_It is a weighted average with _SF . The functions FH and FS are D _HAX (X
= B, G, R) and D _SAX (X = B, G, R).

These reference values D _H0 and D _S0 are common to each color component, but gray balance correction according to the color cast amount of the original image 1 for each color component can be applied to these. If this correction amount is d _SX , d _HX (X = B, G, R), the highlight value reference density value D _HX0 and the shadow side reference density value D for each color component
_SX0 is given by the following equations (8) and (9).

D _HX0 = D _H0 + d _HX (X = B, G, R) (8) D _SX0 = D _S0 + d _SX (X = B, G, R) (9) The correction amount d _HX , d _SX is, for example, , Is calculated using the following equations (10) and (11).

d _HX = K _H · GH (D _HAX −D _H0 ) (X = B, G, R) (10) d _SX = K _S · GS (D _SAX − D _S0 ) (X = B, G, R) (11) However, K _H and K _S are positive constants that have been experimentally determined in advance, and the function GH and GS is, for example, GH (D _HAX −D _H0 ) = (D _HAX −D _H0 ) / [1 + {(D _Hmax −D _Hmin ) / A _H } ^m ] (12) GS (D _SAX −D _S0 ) = D _SAX −D _S0 (13) However, D _Hmax = MAX (D _HAB , D _HAG , D _HAR ) ... (14) D _Hmin = MIN (D _HAB , D _HAG , D _HAR ) ... (15) A _H : Preselected positive constant, m : A preselected positive constant, eg "3" ,.

These reference density values D _HX0 and D _SX0 are the quantities that are the basis of normalization curve generation. That is, the highlight side reference value and the shadow side reference value of the predetermined normalized density range are respectively set to D _NHX and D _NSX (see FIG. 9, X = B,
G, R), the curve NC _X0 passing through two points on the normalized conversion coordinate plane shown in FIG. 9: HL _X0 = (D _HX0 , D _NHX ) SD _X0 = (D _SX0 , D _NSX ). Is the reference normalization curve (reference conversion characteristic). Then, the normalization curve NC _{X that} is actually used is obtained by applying the later-described correction to this reference normalization curve NC _X0 . In addition, for the purpose of facilitating understanding,
Although the reference normalization curve is specifically shown in the figure,
Actually, it is _sufficient to hold only the coordinate values of the above two points HL _X0 and SD _X0 as the data expressing the reference normalization curve.
It is not necessary to create a look-up table representing the reference normalization curve at this point.

(B-3) Judgment of the subject of the original image In step S200 next to FIG. 1A, it is determined what the subject appears in the color original image 1. However, the subject of the original picture includes, for example, skins, silver tableware, and white clothes. In the following, various subjects that can be assumed in advance are represented by the symbol T _i (i = 1 to k) (k is an integer of 2 or more). Therefore, for example, T ₁ = skin, T ₂ = silver tableware, T ₃ = white clothes, and which of the subject T _i (i = 1 to k) appears in the original image 1 is determined. This determination is automatically performed, and the routine executed by the CPU 15 at that time is shown in FIG. 1C. The routine of FIG. 1C is described in detail in Japanese Patent Application No. 63-201646 filed earlier by the applicant of the present invention, and its outline will be described below.

First, in step S201 in FIG. 1C, one pixel is extracted as a sample pixel from the image data stored in the frame memories 13B, 13G, 13R. Then, the color density values D _B , D _G , D _R are substituted into the respective predetermined functions: f _i (D _B , D _G , D _R ) (i = 1 to k) (16) . However, the function f _i is a function given to the i-th theme, and when the typical color densities of the colors characteristic of the theme are B _i0 , G _i0 , and R _i0 , f _i (D _B , D _G , D _R ) = C _i11 (D _B −B _i0 ) ² + C _i22 (D _G −G _i0 ) ² + C _i33 (D _R −R _i0 ) ² + 2C _i12 (D _B −B _i0 ) (D _{G -G i0) + 2C i23 (D} G -G i0) (D R -R i0) + 2C i31 (D R -R i0) (D B -B i0) ... (17) defined by. C _i11 , C _i12 , ..., C _i31 are weighting constants, and have predetermined values.

The values of the function f _i thus calculated are i = 1 to k
Is compared with a predetermined threshold L _i ² for each of When f _i (D _B , D _G , D _R ) ≦ L _i ² (18), the i-th counter among the k counters (not shown) provided in the information processing device 14 Increments the count value n _i of “1” by “1” (steps S202 to S2
04). However, the initial value of n _i is “0”.

In the relation expressed by the equation (18), a point Q (FIG. 8) having (D _B , D _G , D _R ) as a three-dimensional coordinate is a point P _i = (B _i0 , G _i0 in the BGR color space. , R _i0 ) ... (19) means that it exists inside the ellipsoid LC _i . However, the size of the ellipsoid LC _i is determined according to the threshold L _i ² .

Therefore, for example, for the subject T ₁ (skin), the center point P _i
If is set as a point corresponding to a typical skin color, the count value
Based on the size of n ₁ , it is possible to know whether the original image 1 has many skin colors, in other words, whether the subject of the original image 1 is a skin object.

Therefore, after sampling is performed over all the pixels in the frame memories 13B, 13G, and 13R, and the above-described calculation and comparison are executed for these pixels, the ratio n _i / N of the count value n _{i to} the total number N of samplings is predetermined. Threshold TH _i
It is determined whether or not it is larger (steps S205, S206).

When n _i / N> TH _i (20) holds for the i-th subject T _i , it is determined that the subject T _i is included in the original image 1 (step S207). On the contrary, when there is no i satisfying the expression (20), the original picture 1 is the subject assumed in advance.
It is determined that the image does not include any of T _i (i = 1 to k). Then, after obtaining such a determination result,
Return to the main routine in the figure.

(B-4) Correction According to _Subject The correction amount ΔD _HX (X = B, G, R) of the highlight reference density value D _HX0 for each subject T _i is stored in advance in the memory 16 of FIG. Has been done. However, the value of ΔD _HX is predetermined, and for example, for subject T ₁ (skin), ΔD _HB = ΔD _HG = 0, ΔD _HR = 0.02 (21) and subject T ₂ ( For silver tableware, ΔD _HB = ΔD _HG = ΔD _HR = 0.05 (22).

Then, in step S3 of FIG. 1A, the above step S200
The correction amount ΔD _HX (X = B, G, R) corresponding to the subject determined in (1) is selected and read from the memory 16. This correction amount ΔD _HX is added to the highlight reference density value D _HX0 in step S4, whereby the corrected highlight-side density value D _HX can be obtained as in equation (23).

D _HX = D _HX0 + ΔD _HX (X = B, G, R) (23) Further, in this embodiment, the shadow density value D _SX is used as the shadow side reference density value D _SX = D _SX0 (X = B , G, R)… (24), but if necessary, shadow density value D _SX
Alternatively, the intermediate density value can be corrected according to the subject. In addition, when the subject is not found in the original picture 1, the addition of the expression (23) is not performed.

When the highlight density value D _HX and the shadow density value D _SX are determined in this way, 2 on the coordinate plane of FIG.
Point: HL _X = (D _HX , D _NHX ) ... (25) SD _X = (D _SX , D _NSX ) ... (26) Normalized curve NC _X that passes through X = B, G, R To decide. The normalized curve NC _X may be a straight line, or a predetermined non-linear function (standard curve) should be prepared and the standard curve should be deformed so as to pass through the two points HL _X and SD _X. The normalized curve NC _X may be obtained by In each case, the normalized curve NC finally obtained
_X (density signal conversion characteristic) is a curve (or straight line) obtained by shifting the highlight points of the reference conversion characteristic NC _X0 according to the subject.

When the normalized curves NC _X (X = B, G, R) are specified in this way, the numerical data expressing these curves NC _X is stored in the CPU 15
Created by, and loaded into LUT5B, 5G, 5R respectively (step S5).

Then, the above-described main scan is performed in step S6,
Density signal for each color component obtained by reading original image 1 again
D _B , D _G , D _R are normalized by these normalization curves NC _X (X = B, G, R) to become signals D _NB , D _NG , D _NR . Then, after undergoing processing such as color calculation and halftone dot conversion, it is provided for halftone dot image recording.

In the above process, it is important that no step depending on the skill of the operator is required. That is, the numerical values of various parameters may be determined once, and it is not necessary for the operator to determine the values for each original image. Then, each process is systematically executed and has high objectivity.

C. Modified Example [1] When there are a plurality of subjects in the original image, the correction amount used for the actual correction may be obtained by performing a weighted average of the correction amounts for the respective subjects. For example, if the original picture includes skin and silver tableware and both of them are subjects, the correction amount ΔD _HX (X = B, G, R) is _{calculated as} ΔD _HX = e _X · ΔD _1HX + (1 −e _X ) ・ ΔD _2HX … (27) However, ΔD _1HX , ΔD _2HX (X = B,
G, R) are correction amounts (Equations (21) and (22)) that are predetermined for skin and silver tableware, respectively,
e _X is a weighting factor selected within the range of 0 <e _X <1.
The weighting factor e _X may be automatically determined according to the area ratio occupied by each subject, or may be given by manual input by the operator.

[2] For the subject discrimination, automatic discrimination and visual discrimination may be used together. That is, it is possible to entrust the automatic discrimination for the subject whose characteristic color is easy to be specified and to perform the visual discrimination for the other subjects.

[3] The present invention is applicable not only to the plate-making scanner but also to various image processing apparatuses.

〔The invention's effect〕

As described above, according to the present invention, the standard conversion characteristic determined from the statistical property of the density distribution in the original image is corrected according to the automatically determined subject, and the correction amount is different for each subject. Since it is determined in advance, it is possible to always set an appropriate density signal conversion characteristic without depending on the skill level of the operator and individual differences. Since this conversion characteristic determination process is systematic and is based on information that can be objectively grasped, it is a device particularly suitable for automating the conversion characteristic setting. Therefore, the conversion characteristic can be set more quickly than in the case of manual operation.

[Brief description of drawings]

1A to 1C are flowcharts showing the operation of an embodiment of the present invention, FIG. 2 is a block diagram of a plate-making color scanner to which the embodiment is applied, and FIG. 3 is a data processing procedure in the embodiment. Conceptually, FIG. 4 is a diagram showing an average density value frequency histogram, FIG. 5 is a diagram showing a cumulative density value histogram for each color component, and FIG. 6 is a diagram showing a cumulative relative frequency histogram. FIG. 7 is an explanatory diagram of a process of calculating an average color density value from a part of the histogram of FIG. 6, FIG. 8 is a diagram showing an ellipsoid in the BGR color space, and FIG. It is a graph which shows the example of the normalization curve generated. D _X (X = B, G, R) …… Color density signal before normalization, D _NX (X = B, G, R) …… Color density signal after normalization, D _HX0 (X = B, G) , R) …… Highlight side reference density value before correction, D _HX (X = B, G, R) …… Highlight density value, ΔD _HX (X = B, G, R) …… Depending on the subject Correction amount, D _SX (X = B, G, R) …… Shadow density value, CN _X …… Normalization curve

Claims (1)

(57) [Claims] 1. A density signal conversion characteristic for converting a density signal obtained by reading an image of a color original image into a signal suitable for processing in a predetermined image processing apparatus, for setting in the image processing apparatus. An apparatus comprising: (a) reference rule storage means for storing a reference conversion characteristic determination rule for determining the reference conversion characteristic of the density signal of each color component in an arbitrary color image through statistical analysis of the density distribution in the image; (B) correction rule storage means for storing a correction rule predetermined for each subject with respect to the reference conversion characteristic corresponding to the classification of the subject that can appear in any color image, and (c) any color image. Subject determination rule storing means for storing a subject determination rule for determining the subject of the color image from the statistical distribution of the color components of (d) processing in the image processing device Means for applying the reference conversion characteristic determination rule to the original image to determine the reference conversion characteristic of the density signal for each color component of the original image when a color original image to be rendered is given, (e) the subject determination rule Means for determining the subject in the original image by applying to the statistical distribution of the density signal of each color component of the original image, and (f) from the correction rules stored in the correction rule storage means, Means for selecting a correction rule according to the subject in the original image, and (g) correcting the reference conversion characteristic of the original image according to the selected correction rule, and converting the density signal conversion characteristic obtained by this correction into the image. An apparatus for setting a density signal conversion characteristic, comprising: means for setting the processing apparatus.