Method and Apparatus for Processing Color Images Having High Maximum Saturation

The present invention relates to a color image processing method and an apparatus for processing a color image signal.

For example, in a conventional color video printer or the like, the Cy (cyan), M (magenta), and Ye (yellow) signals are mainly processed. FIG. 2 shows an example of the processes typically involved. The A/D converted R, G and B signals are LOG converted into the Cy, M, and Ye signals. Further, these signals are subjected to the masking correction and become Cy', M', and Ye' signals. These signals are input to a head driver and the heads are driven, so that a color image is printed. The masking correction is performed by the following matrix arithmetic operation. ##EQU1##

However, this method has a drawback that the good color reproduction is not obtained. Namely, if the logarithm conversion is simply performed, there is a case in which the concentration becomes a value above 3.0 because of the difference of the dynamic range of the input image signal and the difference of the reproducing range of the ink of the output. In such a case, the concentration obviously exceeds the maximum concentration of the print. On the other hand, a method of correcting the concentration range by the gamma conversion process is used. However, this method has a drawback that the saturation changes and the reproduced color image differs from the original image.

Even in a color display apparatus as well as the foregoing color printer, as shown in FIG. 13, a frame buffer is necessary for each image signal in order to display a color image. In this case, a problem occurs with respect to which kind of image signal is stored in the frame buffer. Namely, in the conventional example of FIG. 13, the R, G and B signals are converted by the matrix conversion into the luminance signal Y and two color difference signals (R-Y and B-Y). The Y, R-Y, and B-Y signals are stored and thereafter, the necessary image processes are performed. For example, in the case of displaying the color image by CRT device (not shown), those signals are again converted into the R, G and B signals and output. The storage of the color image signal in, e.g., the frame buffer or the like causes a problem. If the number of pixels cannot be reduced in the interest of saving the memory capacity, the bit number in the direction of depth of the color image signal cannot help decreasing. However, hitherto, for example, when the number of bits in the depth direction of each of two color difference signals is set to six, there is a problem such a difference occurs between the original image and the reproduced color image because of the decrease in number of bits. To avoid such problem, eight bits are needed to obtain good color reproduction. Due to such situation, since the number of memories cannot be reduced, the memory capacity and the cost increase.

On the other hand, hitherto, two kinds of methods have been generally used in order to obtain good color balance in the image.

(1) The color balance is adjusted before photographing.

(2) The photographed image is corrected.

Method (1) corresponds to "the white balance switch" of a video camera. A white paper or the like is photographed prior to starting the photographing operation and the white balance is set using the "white" image as a reference. Method (2) is widely used in the printing field and the like. However, it largely depends on the feeling and experience of the craftsman.

Therefore, in the case of method (2), hitherto, it is impossible to automatically set the color balance.

In the case of digitizing, hitherto, for example, the least significant bit is omitted or the data is merely compressed a regular interval. In the case of digitizing a regular interval, unless there is a limitation of the capacity of a ROM, the color reproducibility is improved more and more as the interval is made increasingly. However, there is a certain limitation of the ROM capacity as mentioned above.

When the saturation distribution of the color image signal from the actual natural image is examined, it will be understood that it is a rare case that the color image signals are uniformly distributed in color space, and in most cases, the color image signal falls within a region below the half value of the maximum saturation as shown in FIG. 20.

Since it is considered that the influence on the color reproducibility of the whole image by the color of a certain saturation corresponds to the number of pixels having the saturation, the influence of the pixels in the portion having a large saturation with a small distribution is relatively small. Therefore, in the case of constituting the masking ROM by the digitization of the equivalent interval, the portion corresponding to the high saturation in the ROM occupies a constant capacity although it is hardly used for the masking. Therefore, this portion is the vain portion (i.e., not efficiently used). Namely, in order to obtain the high color reproducibility, the portion of a relatively low saturation in which many pixels are distributed must be finely digitized. Thus, the capacity of the ROM increases and the vain degree (size of the vain portion) also increases.

It is an object of the present invention to provide a color image processing method and apparatus which can eliminate the foregoing drawbacks of the conventional techniques.

Another object of the invention is to provide a color image processing method and apparatus in which the memory capacity is reduced without deteriorating the color image quality.

Still another object of the invention is to provide a color image processing method and apparatus in which, regardless of the dynamic range of the input color image signal, good color reproducibility can be derived.

Still another object of the invention is to provide a color image processing method and apparatus in which the color image signal is output as an image signal which is close to the memory color which the human being feels preferable.

Still another object of the invention is to provide a color image processing method and apparatus in which even in the case of a color correction processing table having less capacity, higher color reproducibility is maintained.

Still another object of the invention is to provide a color image processing method and apparatus which can automatically adjust the color balance.

The above and other objects and features of the present invention will become apparent from the following detailed description of the preferred embodiments, taken with reference to the accompanying drawings.

FIG. 1 is a block diagram of an embodiment when the present invention is applied to a color video printer;

FIG. 2 is a diagram for explaining a conventional example;

FIG. 3 is a diagram showing an example of a histogram of a luminance signal;

FIG. 4 is a characteristic diagram showing various conversion characteristics regarding the luminance signal;

FIG. 5 is a diagram for explaining the differences of the dynamic ranges of the image signals of the RGB system and the Ye, M, and Cy systems;

FIGS. 6 to 8 are characteristic diagrams showing the compression characteristics of the saturation;

FIGS. 9A to 9C are diagrams for explaining the sampling process of the pixels when the histogram is obtained;

FIGS. 10A and 10B are diagrams for explaining the case where a histogram with respect to the saturation is obtained and the compression is performed;

FIG. 11 is a block diagram showing another embodiment when considering the case where the memory area is reduced;

FIG. 12 is a circuit diagram showing an example of another construction of a color correction conversion table;

FIG. 13 is a block diagram showing a conventional construction;

FIG. 14 is a block diagram showing an image processing apparatus of another embodiment of the invention;

FIG. 15 is a diagram for explaining the digitization by an HC conversion ROM and a distribution of a skin color;

FIG. 16 is a diagram for explaining the correction to make the hue of the image signal in the skin-colored region approach the memory color as close as possible;

FIG. 17 is a detailed diagram of a masking circuit of an embodiment of the invention;

FIG. 18 is a conceptional diagram of the image processing according to an embodiment of the invention;

FIG. 19 is a block diagram of a color video printer according to the embodiment of the invention;

FIG. 20 is a frequency diagram of the pixels to the saturation of the natural image;

FIG. 21A is a diagram of a digitization space when simply digitizing at a regular interval;

FIG. 21B is a diagram showing a space of an ROM corresponding to the digitization of FIG. 21A;

FIG. 22A is a diagram of a digitization space according to an embodiment of the invention;

FIG. 22B is a diagram showing ROM space corresponding to the digitization of FIG. 22A;

FIG. 23 is a diagram showing a digitization space when the digitization is radially performed from an origin;

FIG. 24 is a block diagram showing a color image processing apparatus of an embodiment of the invention;

FIGS. 25A and 25B are color stereograms before and after the correction, for explaining the principle of the color balance correction in the embodiment of the invention;

FIG. 26 is a flowchart according to a control procedure for detecting the correction amount from the pixel of the maximum luminance;

FIG. 27A is a diagram for explaining another embodiment to detect the correction amount;

FIG. 27B is a flowchart according to a control procedure of the embodiment of FIG. 27A;

FIG. 28 is a flowchart for a control procedure in the case where the color balance correction is performed in a software manner; and

FIG. 29 is a circuit diagram of another embodiment of the color balance correction.

The preferred embodiments according to the invention will be described in detail hereinbelow with reference to the drawings.

As a color processing system for realizing the good color reproduction, an image processing system for processing by the luminance and color differences has been known. FIG. 1 shows an embodiment in which a constitution of a fundamental processing block was applied to a color video printer. After the R, G and B signals are A/D converted, they are stored in image memories 5, 6, and 7, respectively. These signals are matrix-operated upon by a matrix operation unit 8 and converted into a luminance signal Y and color difference signals R-Y and B-Y.

These luminance and color difference signals are subjected to predetermined converting processes, which will be explained hereinafter, by conversion tables 9 and 10 and converted into luminance (Y'), hue (H), and saturation (C). The luminance conversion table 9 is provided to obtain the luminance Y' by normalizing the luminance Y in order to match the dynamic range of the input image signal with the reproduction range of the output (e.g., ink). The normalization and compression are performed by the conversion from Y to Y'. The color difference conversion table 10 is provided to compress the dynamic range in the direction of the saturation C so as not to lose color reproducibility. In this embodiment, as shown in FIG. 1, both luminance conversion and the color difference conversion are performed. However, when considering the viewpoints of holding good color reproducibility and comprising the dynamic range, even if only the luminance conversion or color difference conversion (not both) is independently performed, the similar effect can be obtained as will be clearly understood from the description hereinafter.

The luminance (Y'), hue (H), and saturation (C') are converted into the recording chrominance signal components of Cy (cyan), M (magenta), and Ye (yellow) by a color correction conversion table 11. Further, these components are subjected to the masking correction to correct the uneven color components of the ink and input to a head driver 12 and printed. The outline of the first embodiment is as explained above. The first embodiment will be further described in detail hereinbelow.

After the R, G and B signals were A/D converted, they are stored in the memories 5 to 7 and matrix-operated upon on the basis of the following matrix arithmetic operation of 3.times.3 matrix: ##EQU2##

The reason why the image signal is stored in the memory is mainly because the luminance conversion, which will be explained hereinafter, needs the image of one picture plane in order to examine the dynamic range in the image signal in one picture plane of the input. Therefore, in the embodiment of FIG. 1, the luminance conversion, which will be explained hereinafter, is performed after the R, G and B signals stored in the memories 5 to 7 are converted into the luminance signal Y and color difference signals R-Y and B-Y. If the high speed process is further needed, the memories 5 to 7 are provided at the post stage of the matrix operation unit 8. The input R, G and B signals are directly matrix-operated upon and converted into the luminance signal and color difference signals. These signals are stored in the memories 5 to 7.

The freeze (storage into the memories) of the image signals in the memories as mentioned above is performed at a timing when a freeze switch 21 on a console 20 is turned on.

The conversion of the luminance signal in the luminance conversion table 9 is performed to reduce the difference between the input dynamic range and the reproducible range of the output as mentioned above. However, when this difference is compensated, it is desirable that the input dynamic range not be narrowed anymore than necessary. Therefore, in this embodiment, the dynamic range of the luminance signal Y of the input image signal is examined, thereby selecting the conversion characteristic which is optimum to the input.

The dynamic range regarding the luminance in the image stored in the memory by the freeze switch 21 is detected by a CPU 2. Namely, the image data is sequentially read out of the memories 5 to 7 and a luminance histogram as shown in FIG. 3 is obtained, thereby detecting the dynamic range. The histogram is stored in a RAM 1. In this case, since it takes a long time if all pixels in the memory 5 are scanned, it is possible to thin out the pixels and to scan the pixels, for example, every five pixels in accordance with the sampling (picture plane of the luminance Y) as shown in FIG. 9A. Since the image data of one picture plane must be scanned within 1/30 second, it is necessary to use memories which can operate at a high speed as the ROM and as memories which constitute the A/D converter, conversion table, and the like (if such implementation is desired).

The dynamic range of the luminance is determined by a highlight point (H.sub.p) and a dark point (D.sub.p) in the histogram. It is defined that the highlight point (H.sub.p) denotes the luminance of the point which is 1% away from the bright luminance portion and the dark point (D.sub.p) represents the luminance of the point which is 1% away from the dark luminance portion.

In the luminance conversion table 9, it is possible to perform the conversion of a few kinds of gradation conversion characteristics as shown in FIG. 4. Either one of those conversion characteristics is selected in the following manner. Assuming that the highlight point (H.sub.p) and dark point (D.sub.p) in FIG. 3 have the values near H.sub.p1 and D.sub.p1 in FIG. 4, respectively, the conversion characteristic connecting H.sub.p1 and D.sub.p1 in FIG. 4 is the optimum conversion characteristic which does not lose the dynamic range of the input image. On the contrary, in the case of the characteristic connecting H.sub.p4 and D.sub.p1, the color reproducibility in the higher luminance portion will be lost. The luminance conversion table 9 having such a few kinds of characteristics is constituted by, e.g., a ROM or the like and the number of conversion tables can be increased or decreased in accordance with the capacity. It is sufficient to properly address the ROM on the basis of the values of the highlight point (H.sub.p) and dark point (D.sub.p) of the histogram.

In this manner, the dynamic range in the direction of the luminance is properly compressed (what is called the normalizing process is performed) and the accurate output image which cannot be obtained by the conventional unitary log conversion is reproduced. Namely, the gradation correction suitable for the output ink is performed. On the other hand, since the luminance converted luminance signal Y' is compressed, if it is necessary to store the image signal in the frame memory or the like, the memory capacity can be saved by use of the luminance signal Y' after the luminance conversion rather than the luminance signal Y after the matrix operation 8. As explained above, the invention has such an excellent feature.

On the other hand, if only the luminance is compressed without compressing the dynamic range in the direction of the saturation, which will be explained hereinafter, the saturation C is preferably held and good color reproduction is realized. Further, since the table conversion is used, an increase in circuit scale or the like can be prevented and the like.

In the foregoing embodiment, an explanation has been made with respect to the example of the video printer having the frame memory. However, if a line memory is used, the invention can be also embodied without the frame memory.

As shown in FIG. 5, the dynamic range A in the direction of the saturation obtained with the R, G and B input signals is considerably wider than the reproduction range B in the saturation direction obtainable with the ink of the Cy, M, and Ye signals, so that the input image signals need to be compressed in the saturation direction. For this purpose, in this embodiment, after the color difference signals R-Y and B-Y are converted into the saturation signal C and hue signal H by the color difference conversion table 10, the compression in the saturation direction is performed, thereby obtaining the saturation signal C' from the saturation signal C. First, the color difference signals are converted into the hue and saturation signals in accordance with the following equations: ##EQU3## where, when R-Y=0, H=0. ##EQU4##

The results of the above arithmetic operations in the combinations of the respective values of R-Y and B-Y are stored in the conversion table 10. However, as mentioned above, the dynamic range in the saturation direction of the R, G and B input signals is wider than the reproduction range in the saturation direction of the inks of the Cy (cyan), M (magenta), and Ye (yellow) signals, so that even if the compression is performed, the color reproducibility in printing will not deteriorate excessively. The simplest compression characteristic is shown in FIG. 6. The portion of a high saturation is clipped on the basis of the fact that the ordinary images existing in the natural world are mostly concentrated in the low saturation region. In the diagram, C' is obtained by further converting the following saturation C by the conversion table 10 as follows: ##EQU5##

There occurs a problem that after all of the input saturation signals C larger than the output saturation signal C'.sub.max are compressed to C'.sub.max by the above conversion, the difference of the saturation in this portion cannot be expressed. In this case, it is also considered to use such a conversion characteristic as shown in FIG. 7 in order to prevent such a problem. However, when this characteristic is used, the saturations of all pixels decrease and good color reproduction cannot be derived. Therefore, it is considered to use a non-linear characteristic as shown in FIG. 8. With this characteristic, the saturations of most of the pixels do not decrease, namely, the saturation reproducibility is not lost. Moreover, the saturation signal C' can be formed without losing the difference of the saturation in the portion of the high saturation. This is because the ordinary natural images are concentrated in the region of a relatively low saturation.

The normalization which has been mentioned in the luminance conversion may be also used for the compression of the saturation in order to compress to the more proper dynamic range. Namely, a histogram of the saturation is obtained from the memories 6 and 7. It is assumed that the histogram is as shown in, e.g., FIG. 10A. Assuming that the maximum saturation derived similarly to the case of the luminance compression is C.sub.H and the minimum saturation is C.sub.L, the optimum compression/conversion characteristic is selected from a plurality of characteristics as shown in FIG. 10B on the basis of the values of those maximum and minimum saturations. In this way, the optimum color reproduction can be obtained. However, the selection of the proper conversion characteristic from the histogram is likely to cause the processing speed to be reduced; therefore, the high speed memory device or the like is needed. Examples of the sampling of the color difference signals to obtain the histograms at a high speed are shown in FIG. 9B (picture plane of the color difference R-Y) and FIG. 9C (picture plane of the color difference B-Y).

The resultant Y', H, and C' signals are input to the color correction conversion table 11 and converted into Cy', M', and Ye' signals. The values which are stored into the conversion table are calculated fundamentally by the conversion opposite to the foregoing conversion in accordance with the following equations:

The following matrix approximation is performed to correct the uneven color components of the ink: ##EQU6## where a.sub.11 to a.sub.33 are constants.

According to the foregoing embodiment, not only the difference between the saturations of the input and output is effectively compensated but also the maximum amount of the saturation signal is reduced. Therefore, even in the case of the same number of bits being assigned, the fine digitization can be performed and the efficiency rises. Namely, the storing efficiency by the compression increases.

Although the example of the video printer has been described in the embodiment, the invention can be also applied to other systems regarding the correction of the color image such as a constituting system in the printing and the like. On the other hand, in this embodiment, after the R, G and B signals are A/D converted, they are converted into the Y, R-Y, and B-Y signals by matrix operation. However, it is also possible to constitute the system in such a manner that the R, G and B signals are matrix converted into the Y, R-Y, and B-Y signals before the A/D conversion, and then these signals are A/D converted into digital signals and these digital signals are used.

The embodiment of FIG. 1 has not only a feature the that the color reproduction can be properly performed but also a feature such that the image signal can be compressed as mentioned above. Therefore, it will be understood that this also solves, the problem described above in reference to the conventional systems as occurring when the number of bits in the depth direction of each of two color difference signals is e.g., six, a difference occurs between the original image and the reproduced color image." In other words, even if memory is saved, the saved amount is compensated for by the compression and no problem occurs in the color reproduction because of the foregoing reason.

FIG. 11 shows an embodiment which is constituted from the viewpoint of the saving of the memory. A large difference from the embodiment of FIG. 1 relates to memories 30 to 32. The memories 30 to 32 are, for example, the frame memories (frame buffers) to output an image. Hitherto, it has been a problem to reduce the memories in this portion as mentioned above. However, memory can be saved by the compression of the luminance signal or saturation signal in this embodiment.

There are a few methods of compressing the luminance signal as mentioned above. However, since the memory 5 is essentially necessary to obtain the histogram, the memory 5 will be unnecessary if the characteristics as shown in, e.g., FIGS. 6 and 7 are fixedly used. Because of the similar reason, the memories 6 and 7 are also unnecessary when such a compression that the histogram of the saturation is unnecessary is performed.

In this embodiment, the number of pixels has been set to the standard number (640.times.480) of the NTSC system. However, the number of pixels can be also reduced. In particular, if the color correction conversion table 11 in FIG. 1 or 11 is constituted as shown in FIG. 12, the color data can be remarkably reduced.

As described above, according to the foregoing embodiment, by storing the image signals as the formats of luminance, hue, and saturation, the compression can be performed, so that the capacity of the memory section can be decreased without deteriorating the image quality.

In addition, since the conversion is performed in accordance with the dynamic range of the input color image signal, in particular, the luminance signal, even in the case of the input color image signal of any dynamic range, good color reproduction can be derived.

An explanation will now be made with respect to an embodiment such that in the color correcting process, when the hue is close to the hue of the stored color of a predetermined color, the hue is output as a value which is closer to the hue of the stored color of the predetermined color.

FIG. 14 is a block diagram of a color video printer of the embodiment. This embodiment includes a masking circuit 114 which receives the color difference signals R-Y and B-Y and an HC conversion ROM 113 disposed at the front stage thereof as main components. As the outline of the whole operation, the input signals of Y, R-Y, and B-Y are A/D converted and stored into image memories (110, 111, 112). A CPU 115 takes out the Y, R-Y, and B-Y signals of the pixel to be printed from the image memories (110, 111, 112). The color difference signals R-Y and B-Y taken out are converted into the hue signal H and saturation signal C by the HC conversion ROM 113. Data such that the color difference signals R-Y and B-Y are used as address inputs and converted into the hue signal H and saturation signal C is stored in the HC conversion ROM 113. The Y, H, and C signals are transmitted through the masking circuit 114 and converted into the Cy' (cyan), M'(magenta), and Ye' (yellow) signals and further D/A converted and printed.

The HC conversion ROM 113 converts the color difference signals R-Y' and B-Y' into the luminance signal Y and saturation signal C. In this case, such a digitization as shown in FIG. 15 is executed when converting into the H (hue) and C (saturation) signals. Namely, the digitization is performed in a manner such that the plane of the R-Y and B-Y signals is divided into the regions surrounded by circles and straight lines and the values in the regions are represented by points ".". In this case, ##EQU7## where, when R-Y=0, H=0, and ##EQU8## The amplitude from a reference axis corresponds to a value the hue H and the radius of a circle corresponds to a value the saturation C. In FIG. 15, H of the output shows the case of four bits and C shows the case of three bits. The hatched portion 100 shows the region in which the colors near the skin color are distributed.

In the diagram, h.sub.0 denotes a hue of the preferable skin color of the stored color, and h.sub.1 and h.sub.2 represent ranges of the hues which ought to be sensed as the skin color by the digitization of FIG. 15. The digitization as shown in FIG. 15 is nothing but the process such that the values of hue (H) are taken every equal angle and the values of saturation (C) are taken at regular intervals and these values are set representative values. Therefore, no consideration is made to the visual sensation of the human being in the case of such a digitization.

The skin color to which the human being pay the largest attention is reproduced by use of the masking circuit 114 having such a constitution as shown in FIG. 17. The masking circuit 114 comprises: an HC decoder 120 for converting the H and C signals to the corrected H' and C' signals as shown in FIG. 16; a converter 121 to reversely convert the Y', H', and C' signals into the Y', R-Y', and B-Y' signals; an RGB converter 122 to further convert the Y', R-Y', and B-Y' signals into the R, G, and B signals; a converter 123 to convert the R, G, and B signals into the Cy (cyan), M (magenta), and Ye (yellow) signals; a masking matrix circuit 124 to perform the ordinary masking processes; and the like.

The HC decoder 120 will be first described. When it is assumed that the HC conversion ROM 113 outputs the H signal of h bits and the C signal of k bits, the hue and saturation to the region other than the skin-colored region 100 are

where C.sub.max denotes the maximum saturation. The HC decoder 120 does not perform any process in such a region 100.

On the other hand, assuming that the HC conversion ROM 113 generated the image signals included in the skin-colored region 100 to the HC decoder 120 (as the signals of h bits and k bits), the HC decoder 120 outputs the hue signal as h' which is closer to h.sub.0 as shown in FIG. 16 when h lines between h.sub.1 and h.sub.2. Further, when also considering the saturation in the skin-colored region 100, only the output of the HC conversion ROM 113 in the range where h falls between h.sub.1 and h.sub.2 and 2.sup.k .ltoreq.C.sub.0 is corrected. The HC decoder 120 to convert the input as shown in FIG. 15 into the output as shown in FIG. 16 can be easily constituted as a ROM.

In this manner, by setting the hue of the image signal which is considered to be the skin color into the value near the hue of "the preferable skin color", the skin-colored region which is generally distributed can be collected to the region of "the preferable skin color" and the reproducibility of the skin color can be improved.

The resultant Y, H, and C signals are input to the converter 121 and converted in the following manner:

Further, the following R, G, and B signals are output from the RGB converter 122:

The following conversion is performed by the converter 123:

Then, the following matrix conversion is performed by the masking matrix circuit 124 to correct the uneven color components of the ink:

where a.sub.11 to a.sub.33 are constants.

The HC conversion ROM 113 and HC decoder 120 can be also constituted by one ROM. Although the foregoing embodiment has been described with respect to the video printer, in the case cf outputting to a CRT device or the like, a more preferable skin color can be reproduced if the output of the RGB converter 122 is displayed by the CRT. Although the embodiment has been described with regard to the skin color, the invention can be also applied to other specified color, e.g., sky color or the like.

As described above, according to the present embodiment, by approaching the present hue in a predetermined range in the image signals to the hue of a desired stored color, the color which the human being will deem preferable can be reproduced. In particular, the skin color can be effectively reproduced.

A non-linear digitizing method of the color image signals will now be described.

FIG. 18 is a diagram showing a fundamental constitutional concept of an embodiment in which the luminance signal Y and color difference signals R-Y and B-Y are input as color image signals and masking processed. When the luminance signal Y and color difference signals R-Y and B-Y are used as the color image signals, the luminance Y is unnecessary for the saturation. Therefore, the luminance signal Y is not input to an address conversion table 201. The address conversion table 201 consists of, e.g., a ROM or the like and the color difference signals are input as addresses in the ROM. Data whose low-saturation portion has been finely digitized and a high-saturation has been roughly digitized is stored in advance in the ROM. The ROM has the functions to perform the digitization and the conversion of the address space. The address conversion table 201 outputs the color image signals R-Y' and B-Y' which are the converted addresses and whose values were changed. In this case, even when the signals each consisting of six bits are input to the address conversion table 201 and the signals each consisting of five bits are output therefrom, the digitization is not simply performed at regular intervals as in the conventional apparatus. Therefore, the accuracy does not simply deteriorate to 1/4.

FIG. 19 is a block diagram of a color video printer to which the invention is applied. This printer comprises: a masking ROM 214 which receives the Y, R-Y, and B-Y signals; and an address conversion ROM 213 arranged at the front stage thereof. As the outline of the whole operation, the Y, R-Y, and B-Y input signals are A/D converted and stored into image memories (210, 211, 212). A CPU 215 reads cut the Y, R-Y, and B-Y signals of the pixel to be printed from the image memories 210 to 212. The R-Y and B-Y signals are transmitted through the address conversion ROM 213 and digitized into the signals each consisting of less bits. Next, the Y, R-Y, and B-Y signals are transmitted through the masking ROM 214 and converted into the Cy (cyan), M (magenta), and Ye (yellow) signals and further D/A converted and printed.

When the digitization is performed at regular intervals as in the conventional apparatus, the content of the address conversion ROM 213 (or address conversion table 201) is as shown in FIG. 21A. The digitization (i.e., the address conversion) of the (color difference) image signals denotes that the plane which is expressed by R-Y and B-Y is represented by points ".multidot.". At this time, the cross point of the R-Y and B-Y axes has the saturation 0 and the saturation increases as one moves outside, away to the outside from this point.

FIG. 20 shows a histogram of the saturations of the pixels of the color image which ordinarily exists in the natural world. In such a natural image, the number of pixels having the saturations above a certain saturation value (C.sub.n) rapidly decreases. Therefore, in the case of the digitization space of regular intervals shown in FIG. 21A, only the space of the masking ROM shown in FIG. 21B is obtained. In the ROM space of FIG. 21B, the portions of the saturations above the saturation value C.sub.n relatively become vain as mentioned before.

A method according to the embodiment is shown in FIGS. 22A and 22B. It is the fundamental concept of the digitization of the embodiment that the pixel distribution according to the saturation is plotted in the color difference (R-Y, B-Y) space, and the digitization interval is changed at a long interval of a few stages in accordance with the distribution density. FIG. 22A shows a digitization space corresponding to FIG. 21A in the embodiment. The digitization interval is set to two stages using the saturation value C.sub.n as a reference. The portion of a low saturation is digitized at a short interval. The region of a high saturation is digitized at a long interval. FIG. 22B shows an ROM space in the address conversion ROM 213. The sampling point in FIG. 22A corresponds to the sampling point in FIG. 22B in a one-to-one corresponding manner. Thus, as shown in FIG. 22B, the area in the portion surrounded by C.sub.n and -C.sub.n is widened as compared with that in FIG. 21A. The fine masking process is performed to the image signal of a low saturation and the color reproducibility is improved by only the amount of increased area. The reason why the intervals in FIG. 22B are equal is because, in general, the addresses must be input into the ROM at regular intervals.

The address conversion ROM 213 will be further described with reference to FIG. 22A. In the diagram, the region obtained when the maximum saturations of R-Y and B-Y were set to C.sub.max is shown by a straight line. The region of the saturation C.sub.n which is half of C.sub.max is indicated by a broken line. Assuming that the number of bits of each of the outputs R-Y and B-Y of the address conversion ROM 213 is n, the number of sample points to be digitized is m (=2.sup.n .times.2.sup.n). When it is assumed that K points among the m points are contained in the region of a low saturation in the area surrounded by the broken line, the number of sampling points to be included in one side of the area in the broken line is .sqroot.K. Therefore, when the region between -C.sub.n and +C.sub.n is divided by .sqroot.K at regular intervals, it is sufficient to assign the intervals in the vertical and horizontal directions at every other 2.multidot.C.sub.n /.sqroot.K. In the region other than the area in the broken line, the number of sample points is n.sup.2 -K and the area is 3.multidot.C.sub.max.sup.2. Therefore, it is sufficient to arrange in both of the vertical and horizontal directions at every other .sqroot.3.multidot.C.sub.max /.multidot.m-k.

Although the digitization in the square region has been performed in FIG. 22A, in the case of the saturation, there is also considered a method whereby the region of a low saturation is set to a circle and a larger number of bits are assigned to this circular region as compared with that in the other regions as shown in FIG. 23 by paying due attention to the fact that the saturation increases like a circle from the cross point of the R-Y and B-Y axes.

As described above, the address conversion ROM which receives the Y, R-Y, and B-Y signals is arranged at the front stage of the masking ROM, and a larger number of bits are assigned to the region of a low saturation, so that the natural image can be extremely efficiently digitized.

Although the masking process has been described as an example in the foregoing embodiment, the color correcting process is not limited to the masking process. The input color image signals are not limited to the luminance signal and color difference signals but may be the signals of the RGB system or other color expression system. When the distribution of the pixels is complicated, the number of digitization stages is not limited to two but may be set to a large number of stages. In this case, the fine color reproduction can be realized with less memory capacity.

As described above, according to the embodiment, an attention is paid to the deviation of the pixel distribution to the saturations of the input image signals, the fine color correcting process is performed to the image of a low saturation consisting of a large number of pixels, and the rough color correcting process is executed to the image of a high saturation consisting of less number of pixels. Therefore, the memory capacity of the color correction processing table can be efficiently used.

An embodiment in which the color balance is automatically corrected will now be described.

The correction of the color balance in this embodiment is performed by paying an attention to the highlight point (the pixel of the highest luminance) in the image. Namely, in the case of the image whose colors were well balanced, the protability such that the color difference of the highlight point is "0", namely, "white" is high. Therefore, the color balance correction in the color image processing apparatus in this embodiment will be summarized as follows.

(1) Since the color difference amount of the highlight point pixels in the image obtained is considered to be the "deviation" of the color balance, this color, difference amount is set to the value .DELTA.E to correct the color balance.

(2) With respect to all pixels, the .DELTA.E.times.luminance/maximum luminance is subtracted from the color difference amount of each pixel and the resultant difference is used as the color image signal after the correction, thereby adjusting the color balance.

FIG. 25A shows a color stereogram of the image whose color balance is deviated. FIG. 25B shows a color stereogram of the image after the color balance was corrected. The color differences (R-Y, B-Y) of the pixel having the maximum luminance (Y) are considered to represent the "deviation" of the color balance. Therefore, when the color correction is performed by regarding the color difference amounts as the correction amounts .DELTA.E (.DELTA.R.sub.Y, .DELTA.B.sub.Y), the corrected image as shown in FIG. 25B is obtained. When the correction amounts are calculated, the correction amounts corresponding to the color differences of an arbitrary pixel in the image are set to the values such that .DELTA.E was proportionally distributed by the luminance/maximum luminance (Y/Y.sub.max) as shown in FIG. 25A. In the following embodiment, two examples of the method detecting .DELTA.E (.DELTA.R.sub.Y, .DELTA.B.sub.Y) will be explained. (Outline of the apparatus in the embodiment)

FIG. 24 is a block diagram of a color image processing apparatus in the embodiment. The outline of the image processing apparatus shown in FIG. 24 is as follows.

The A/D converted color image signal Y and color difference signals R-Y and B-Y are stored in an image memory 302. A CPU 308 reads out the data from the image memory 302 and detects the correction amounts .DELTA.E by either the method shown in FIG. 26 or the method shown in FIGS. 27A and 27B. The correction amounts (.DELTA.R.sub.Y, .DELTA.B.sub.Y) are respectively input to correction ROMs 303 and 304. The luminance signal Y and color difference signals R-Y and B-Y are also input to the correction ROMs 303 and 304, respectively. The outputs of the ROMs 303 and 304 and the luminance signal Y are input to a masking circuit 305 and converted into the Cy, M, and Ye signals. Then, these signals are D/A converted and printed by a head driver 307. The correction of the color balance by the correction ROM is shown in FIG. 24 and will be explained in detail hereinafter. An explanation will be also made hereinafter with respect to an embodiment such that the correction of the color balance to each of the pixels constituting the image is executed by the CPU 308 in a software manner. The program for a procedure as shown in FIG. 26 and the like is stored in a RAM 309.

In the method of determining .DELTA.E shown in FIG. 26, the color difference signals of the pixel having the maximum luminance (Y.sub.max) are set to .DELTA.E. First, in step S1, the pixels of the highlight points are found out by comparing the luminance Y of each pixel in the image memory. In step S2, the color difference amounts R-Y and B-Y of the pixels are regarded as the "deviation" of the color balance and set to the correction amounts .DELTA.E (.DELTA.R.sub.Y, .DELTA.B.sub.Y).

FIG. 27A shows the concept of another method of obtaining .DELTA.E. FIG. 27B shows a procedure for this method. According to the foregoing method of obtaining .DELTA.E, the values for .DELTA.E are obtained from one point of the maximum luminance. Therefore, there is a fear that the correction amount might become unstable. To prevent this problem, about ten sample points of the pixels having the highest luminance are collected and the average of the color difference amounts of these pixels is calculated to obtain the correction amounts, thereby stably and effectively correcting the color balance.

FIG. 27B is a flowchart for the process to determine the correction amount .DELTA.R.sub.Y and .DELTA.B.sub.Y. First, ten highlight points are searched from the image in step S10. However, in this case, it takes a long time to search ten highlight points from all of the pixels. Therefore, as shown in FIG. 27A, when the input image consists of 640.times.480 pixels, the pixels are sampled every other four pixels in the vertical and horizontal directions in the region of (121 to 520).times.(86 to 295) inside of this image and ten highlight points are searched. Even by such a thinned-out sampling, almost the same result as that in the case of searching them from all of the pixels can also be derived.

When the color difference amounts R-Y and B-Y of ten highlight points are (R-Y).sub.1, (R-Y).sub.2, . . . , (R-Y).sub.10 and (B-Y).sub.1, (B-Y).sub.2, . . . , (B-Y).sub.10, .DELTA.E (.DELTA.R.sub.Y, .DELTA.B.sub.Y), become

Thus, the more effective correction amounts can be obtained as compared with the case of deciding the correction amounts from one highlight point.

The maximum correction amounts are derived in this manner. An embodiment in the case where the CPU 308 performs the correction of the color balance in a software manner will now be explained. In the correction of the color balance, the correction amounts .DELTA.R.sub.Y and .DELTA.B.sub.Y are used as the maximum values and the color balance is variably corrected in proportion to the luminance of each component pixel of the image. This is because if the correction amounts .DELTA.R.sub.Y and .DELTA.B.sub.Y are merely subtracted from the color differences R-Y and B-Y of each component pixel of the image, the correction is performed too much as the luminance of the pixel becomes low. Therefore, less correction amounts are subtracted for the pixels having a low luminance with regard to all of the pixels as explained in FIG. 25A by performing the following correction.

Y, R-Y, and B-Y denote the image signals of each of the pixels constituting the image. Y.sub.max denotes the maximum luminance. FIG. 28 shows a procedure for the correction of the color balance. By the foregoing method, the image whose color balance was deviated in the image memory as shown in FIG. 25A is corrected as shown in FIG. 25B.

As described above, the resultant data obtained by multiplying the correction amounts obtained from the highlight points with the luminance/maximum luminance is subtracted from the color difference amounts of each component pixel of the image, so that the color balance can be effectively corrected at a high speed by the simple calculations. The color balance correction amounts linearly change in accordance with the luminance in the foregoing embodiment. However, the invention may be also applied to non-linear correction amounts by applying the weights corresponding to the luminances.

The correction ROMs 303 and 304 constituting the correction table for the correction of the color balance will now be explained. These ROMs intend to realize the high-speed process by executing the operations in step S21 in FIG. 28 in a hardware manner in place of the CPU 308.

As shown in FIG. 24, with respect to R-Y, the Y signal, R-Y signal, and signal of the correction amount .DELTA.R.sub.Y are input to the ROM, so that the corrected R-Y' signal is output. Similarly, the B-Y' signal is output with respect to the B-Y signal. The correction amount .DELTA.R.sub.Y and .DELTA.B.sub.Y are previously calculated on the basis of the following equations and stored.

The processes can be performed at a high speed by executing the table conversion in this manner. FIG. 24 shows a block diagram of the color video printer using the table conversion. For example, with respect to R-Y, only the R-Y signal, .DELTA.R.sub.Y signal, and Y signal are input to the ROMs 303 and 304 shown in FIG. 24. The Y.sub.max signal is not input. This is because the value of Y.sub.max is fixed in order to reduce the number of input bits to the ROMs. For example, when the luminance Y consists of eight bits, the value of Y.sub.max is fixed to, e.g., "255". The values obtained by performing the foregoing equations using Y.sub.max =255 are stored into the ROMs. The correction accuracy does not deteriorate even by this method and, accordingly, the scale of each ROM can be reduced by only the amount corresponding to the decrease in number of input bits.

By use of the table conversion on the basis of three input signals of correction amount (.DELTA.E), luminance, and color difference amount, it is possible to realize the processing apparatus which can effectively correct the color balance at a high speed with a simple constitution.

FIG. 29 shows a modified example of the correction circuit. In this example, in order to reduce the scale of the ROM, only the calculations of .DELTA.R.sub.Y .times.Y/Y.sub.max and .DELTA.B.sub.Y .times.Y/Y.sub.max in the foregoing equations are executed in the ROM and the subtractions are performed in subtracters 322 and 323.

As described above, according to the embodiment, the color difference amounts of the pixel of a predetermined luminance are regarded as the deviation of the color balance and the color correction is performed in accordance with the luminance on the basis of this deviation. Therefore, it is possible to provide the color image processing apparatus which can correct the color balance by a constant algorithm. If the predetermined luminance is set to the maximum luminance, the color balance can be more effectively corrected.

As described above, according to the invention, the useful color image process can be realized.

The present invention is not limited to the foregoing embodiments but many modifications and variations are possible within the spirit and scope of the appended claims of the invention.