JPH03109868A - Picture processor - Google Patents

Abstract

PURPOSE:To improve the picture quality and the processing efficiency by applying binarizing processing in response to a characteristic of each picture with respect to picture information where a character area and a photographic area exist in mixture. CONSTITUTION:The unit is provided with a binarizing means 4, a binarizing error calculation means 5, a weight coefficient storage means 6, a weight error calculation means 7, and a characteristic quantity extraction means 10 or the like. A maximum density difference of a picture within a prescribed range is calculated as a characteristic quantity, and whether a noted picture element is a character or a photograph is identified by the characteristic quantity, then simple binarization is implemented and when photograph is identified, the error spread method is used to apply binarizing processing and the weight error at the binarizing processing is dispersed to peripheral picture elements by the following binarizing processing. Thus, even when photograph and characters exist in mixture in a picture, binarizing processing in response to each characteristic is attained and normal error dispersion is attained even in the vicinity of the border.

Description

[Detailed Description of the Invention] [Objective of the Invention] (Industrial Field of Application) The present invention targets a document image in which a text portion and a photo portion are mixed, and improves the resolution of the text portion and the gradation of the photo portion. The present invention relates to an image processing device that performs binarization processing while maintaining high quality.

(Prior Art) Generally, in an image processing device such as a document image processing device that can handle not only code information but also image information, text and line drawings are processed based on the image information read by a reading means such as a scanner. Image information with contrast is simply binarized using a fixed threshold value, and image information with gradation such as photographs is binarized using a pseudo gradation means such as a dither method. This is because if read image information is uniformly subjected to simple binarization processing using a fixed threshold value, resolution will be preserved in areas such as text and line drawings, so image quality will not deteriorate, but in areas such as photographs, image quality will not deteriorate. On the other hand, if the read image information is subjected to uniform gradation processing using systematic dithering, etc., the gradation is preserved in areas such as photographs. However, in areas such as characters and line drawings, the resolution decreases, resulting in an image with degraded image quality. That is, when a single binarization method is applied to read image information, it is impossible to binarize while simultaneously satisfying the image quality of each region having different characteristics.

On the other hand, the error diffusion method has been proposed as a binarization method that satisfies the gradation of photographic image areas and has better resolution for text and line drawing areas than the systematic dither method. "Error Diffusion Method" is published in the journal Proceedings of the
S, 1. D Vol, 1725econd Quart
er 1.97B75-77J literature rAn Adapt
tive Algorithm for Spatial
As described in Gray 5caleJ, the value obtained by multiplying the binarization error by a predetermined weighting coefficient by the binarization error when the pixel of interest is binarized with a certain threshold value is calculated as the red color of a predetermined region around the pixel of interest. This method attempts to perform binarization by adding this dispersed binarization error as a correction value when binarizing pixels that have not been binarized.

FIG. 6 is a block diagram showing the configuration of an image processing apparatus using the above-described "error diffusion method."

In the figure, 410 is an input image signal, 40 is a correction means 411 for correcting image information of a pixel of interest, and 4 is a corrected image signal.
2 is a binarization means for binarizing the corrected image information of the pixel of interest, 421 is a binarized image signal, and 43 is a binarization error calculation for calculating the binarization error of the binarized pixel of interest. means, 4
31 is a binarization error signal, 44 is a weighting coefficient storage means for storing a weighting coefficient for calculating a weighting error, and 45 is a weighting coefficient storage means 44 for the binarization error 431 calculated by the binarization error calculation means 43. 451 is a weighted error signal, 46 is an error storage means for storing the weighted error calculated by the weighted error calculating means 45, and 481 is an image correction signal. The binarization process using the "error diffusion method" will be explained in detail below.

For example, an input image signal 410 obtained by reading an image with an input device such as a scanner is corrected by an image correction signal 481 in the correction means 40, and the corrected image signal 411
is output as The binarization means 42 to which the corrected image signal 411 is supplied generates the binarized image signal 421 by comparing the corrected image signal 411 and the binarization threshold Th.
Output. That is, for example, r80,6J (the subscript "16" indicates a hexadecimal number) is used as the binarization threshold Th, and the corrected image signal 4]1 is set to the binarization threshold T.
If it is thicker than h, ``1'' (black pixel) is output as the binarized image signal 421, and if it is smaller, ``0'' (white pixel) is output.

Next, in the binarization error calculation means 43, the corrected image signal 41
1 and the binarized image signal 421 (here, when the binarized image signal 421 is 10, it is rO + 6J, and when it is "l", it is F F 16J). The weighting error calculation means 45 outputs the weighting coefficients A and B stored in the weighting coefficient storage means 44 as the binarized error signal 431.

CS, D (A=7/16, B=1/16, C=
5/16, D=3/16) is calculated. Here, "*" in the weighting coefficient storage means 44 indicates the position of the pixel of interest, and the binarization error of the pixel (note 1) is multiplied by the weighting coefficient ASBSC,D, and the surrounding four pixels of the pixel of interest (
The weighting error 451 of pixels corresponding to the positions of weighting coefficients A, B, C, and D is calculated. The error storage means 46 is for storing the weight error 451 calculated by the weight error calculation means 45, and the weight error 451 for 4 pixels calculated by the weight error calculation means 45 is for the pixel of interest "*". are added to the areas eAseB and ec%eD, respectively, and stored.

The image correction signal 481 described above is a correction signal for the position of "*", and is a signal obtained by accumulating weight errors for a total of four pixels calculated in the above procedure.

In this manner, the "error diffusion method" minimizes the binarization error by diffusing the binarization error generated by binarization of the pixel of interest to surrounding pixels to compensate for the error.

Therefore, if the gradation is important, such as an input image or a photographic image, it is possible to perform binarization processing that fully satisfies the gradation.

However, when a human image is a character image, in which 1M resolution is more important than gradation, error compensation processing is a problem, and the resolution of the character part deteriorates. In addition, in the case of an image with a mixture of photos and text, a weighted error is calculated even if the pixel of interest is text near the boundary between the text and photo areas, and the photo area adjacent to the text area is binarized. Since this is reflected in the correction value at the time of image processing, accurate error diffusion cannot be achieved at the edges of the photographic area, resulting in a decrease in image quality.

(Problems to be Solved by the Invention) In the present invention, since the "error diffusion method" performs binarization processing using a single method even for document images in which text areas and photo areas are mixed, each area If it is not possible to perform processing according to the characteristics of each image, and the input image is one that emphasizes gradation, such as a photographic image, perform binarization processing that fully satisfies the gradation. However, in images where resolution is more important than gradation, such as character images, there are disadvantages such as error compensation processing, deterioration of resolution in character areas, and This eliminates the drawback that accurate error diffusion is not possible near the boundary between the text and photo areas, resulting in a decrease in image quality, and the system can binarize image information according to the characteristics of each image, even if the image information includes text areas and photo areas. Through processing, it is possible to improve image quality and prevent deterioration of image quality even near the boundary between text areas and photo areas.Furthermore, by performing processing according to the characteristics of the image, various types of images can be improved. An object of the present invention is to provide an image processing device that can improve processing efficiency in processing.

[Structure of the Invention] (Means for Solving the Problems) An image processing apparatus of the present invention includes a binarization means for binarizing image information of a pixel of interest in an image to be processed, and a binarization means for binarizing image information of a pixel of interest in an image to be processed. binarization error calculation means for calculating a binarization error from the converted information and the image information of the pixel of interest, a weighting coefficient storage means for storing a weighting coefficient for calculating a weighting error, and this weighting coefficient storage. weighting error calculation means for calculating a weighting error by multiplying the binarization error calculated by the binarization error calculation means by a weighting coefficient stored in the means; a feature amount extraction means for extracting a feature amount representing the feature of the feature amount, and a processing means for truncating the weight error calculated by the weight error calculation means according to the feature amount extracted by the feature amount extraction means; an error storage means for storing the weighted error that has been truncated by the processing means; and an error storage means for storing the weighted error that has been truncated by the processing means; The present invention is characterized by comprising a selection means for selecting according to the amount, and a correction means for correcting the image information of the pixel of interest when the selection means selects to perform correction by weight error.

(Function) The present invention utilizes the characteristic that the maximum density difference between each pixel within a window within a predetermined range including the pixel of interest is large for text areas and small for photo areas.
First, the maximum density difference of images within a predetermined range is calculated as a feature amount, and this feature amount is used to identify whether the pixel of interest is a text or a photograph. When it is determined that the pixel of interest is a character, the image signal of the pixel of interest is binarized using a predetermined threshold without correction, so-called simple binarization.

At this time, the weight error is not stored by cutting off the binarization error caused by the binarization process, and therefore the weight error during the simple binarization process is dispersed to surrounding pixels in the following binarization process. I won't let you.

On the other hand, when it is identified that the pixel of interest is a photograph, the image signal of the pixel of interest is corrected with the binarization error information of surrounding pixels, and the corrected image information obtained is binarized using a predetermined threshold value of 0. Binarization processing is performed using a so-called error diffusion method, and weight errors during this binarization processing are dispersed to surrounding pixels in the following binarization processing. This makes it possible to perform binarization processing according to the characteristics of each image, even if the image contains both photographs and text, as well as accurate error diffusion near the boundary between text and photograph areas. It becomes.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 2 is a principle diagram showing the method of binarization processing according to the present invention. In the figure, a line buffer 1 stores image information of an image to be processed, and "*" indicates the position of a pixel of interest in the line buffer 1. The input image signal 21 from the line buffer 1 is supplied to the correction means 3.

The correction means 3 corrects the image information of the pixel of interest, and the corrected image signal 31 corrected by the correction means 3 is supplied to the binarization means 4 and the binarization error calculation means 5. The binarization means 4 binarizes the corrected image information of the pixel of interest using a predetermined threshold Thl, and the binarized image signal 41 binarized by the binarization means 4 is a binary The result of the conversion processing is outputted to the outside, and is also supplied to the binarization error calculation means 5. The binarization error calculation means 5 calculates the binarization error of the binarized pixel of interest from the corrected image signal 31 and the binarized image signal 41. The binarized error signal 51 thus obtained is supplied to the weighted error calculation means 7. The weighting error calculation means 7 calculates the weighting coefficients stored in the weighting coefficient storage means 6 and the binarization error signal 51.
is input, and the binarization error signal 51 calculated by the binarization error calculation means 5 is multiplied by the weighting coefficient of the weighting coefficient storage means 6 to calculate the weighting error. This weight error calculation means 7
The calculated weighted error signal 71 is supplied to the error truncation means 8. The error truncation means 8 selects whether to output the weighted error signal 71 or zero according to the identification signal 111 from the identification means 11 (described later), and outputs it as a 2 selection weighted error signal 81.

Selection weight error signal 81 outputted by this error truncation means 8
is supplied to the error storage means 9 and stored as a weighted error. The weighted error stored in the error storage means 9 is supplied to the selection means 12 as an image correction signal 91.

On the other hand, the feature calculation means 10 calculates the maximum density difference (ΔDmax) within a predetermined area of the line buffer 1 (the area surrounded by a large frame) from image information within the area. The maximum density difference signal 101 outputted by the feature quantity calculation means 10 is supplied to the identification means 11.The identification means 11 compares the maximum density difference signal 101 with a predetermined threshold Th2. , outputs an identification signal 111 indicating whether the pixel of interest is a photograph or a character.The identification signal 111 outputted by the identification means 11 is sent to the selection means J2 in such a manner that If the identification signal J, l ] indicates that it is a photograph, the selection means 12 outputs the weighted error stored in the error storage means 9 as the selected image correction signal 121, and 3 indicates that the selected image correction signal 121 is output as the selected image correction signal 121. Then, the input image signal 21 from the line buffer 1 is outputted as the selected image correction signal 121 in the correction means 3. The signal is corrected and supplied to the binarization means 4.

Next, in the above configuration, the binarization processing method of the present invention will be explained in detail.

For example, an input image signal 21 obtained by reading an image with an input device such as a scanner is corrected by the selected image correction signal 121 in the correction means 3 and output as a corrected image signal 31. The binarization means 4 to which the corrected image signal 31 is supplied is configured to combine the corrected image signal 31 and the binarization threshold Th.
A binary image signal 41 is output by comparing with l. That is, as the binarization threshold Thl, for example, r
80+6", the corrected image signal 31 is set to the binarization threshold Th
If it is larger than l, the binary image signal 41 is set to “1” (
If it is smaller, it outputs "0" (white pixel).

Next, in the binarization error calculation means 5, the corrected image signal 31 and the binarized image signal 41 (here, the binarized image signal 41 is rO+6J when the image signal 41 is "O", and rO+6J when it is "1"). is assumed to be "FF16"), and outputs this as a binarized error signal 51. Weight error calculation means 7
Now, the weighting coefficient ASB stored in the weighting coefficient storage means 6 is added to the binarization error signal 51.

C, D (However, A = 7/16, B = 1/
16, C=5/16, D=3/16) is calculated. Here, "*" in the weighting coefficient storage means 6 indicates the position of the pixel of interest, and the binarization error of the pixel of interest is multiplied by the weighting coefficient ASBSC,D.
Pixel (pixel corresponding to the position of weighting coefficient A, B, C, D)
Calculate the weight error 71 of .

This weighted error 71 is then supplied to the error truncation means 8. In the error truncation means 8, when the identification signal 111 from the identification means 11, which will be described later, is "1", that is, indicating that the pixel of interest is a character pixel, the weighted error 71 is truncated and zero is selected and outputted as the weighted error 81. and identification signal 1.
When 11 is "0", that is, the pixel of interest is a photographic pixel, the weight error 71 is output as is as the selection weight error 81.

5. The storage means 9 stores the selection weight error 81 calculated by the weight error calculation means 7 and subjected to truncation processing by the weight error truncation means 8 according to the characteristics of the image. That is,
When the identification means 11 identifies the pixel as a photographic pixel,
Weight error 71 for 4 pixels calculated by weight error calculation means 7
are eAs B and e for the pixel of interest “*”, respectively.
It is stored cumulatively in each class area of CSep. On the other hand, if the identification means 11 identifies the pixel as a character pixel, zero is added to each area of eAz enSeCSeD for each pixel of interest "*". This means that in the case of character pixels, binarization errors are not diffused. The image correction signal 91 is a signal at the position of "*", and is a signal obtained by accumulating weight errors for a total of four pixels calculated in the above procedure.

On the other hand, the feature calculation means 10 calculates the maximum density difference in image density within the (4×4) window (large frame portion) including the pixel of interest “*”, and outputs a maximum density difference signal 101. The identification means 11 compares the maximum density signal 101 with a preset threshold Th2, and if the maximum density difference signal 101 is larger than Th2, it outputs "1" indicating the character part as the identification signal 111, and if it is smaller, it outputs "1" indicating the character part. Outputs "0" indicating the photo section. The selection means 12 determines whether or not to output the selected image correction signal 121 according to the identification signal 111. That is, the identification signal 111 or "O", that is, the selected image correction signal 12 if it is a photographic pixel.
The image correction signal 91 is set as "1", the identification signal is set as "1",
In other words, if it is a character pixel, zero is output as the selected image correction signal 121.

Next, an example of an image processing apparatus to which the above binarization processing method is applied will be described. Note that parts and signals having the same functions as those in the principle diagram shown in FIG. 2 will be described with the same reference numerals.

FIG. 1 is a schematic configuration diagram showing an image processing apparatus according to an embodiment of the present invention. This image processing device inputs image information obtained by reading a document with a reading device such as an image scanner as digital data of, for example, 8 bits per pixel, and binarizes this data. .

The line buffer 1 temporarily stores such image information and provides it for image processing (binarization processing) described below.

The delay means 2 inputs the image signal output from the line buffer 1 in synchronization with a predetermined clock, and processes the image signal for a predetermined period of time, that is, calculates an identification signal 111 to be described later and outputs a selected image correction signal 121. This is to delay the amount of time until the

The correction means 3 is composed of an adder, and corrects the image information of the pixel of interest. That is, the image signal 21 delayed by the delay means 2 is added to a selected image correction signal 121 from a selection circuit 12, which will be described later, and a corrected image signal 31 is output. This corrected image signal 31 is binarized by being compared with a predetermined threshold Thl by the binarizing means 4 and output as a binarized image signal 41. At this time, if the corrected image signal 31 is larger than the binarization threshold Thl, "1" (black pixel) is set as the binarized image signal 41, and if it is smaller, it is set to "0".
(white pixels) are each multiplied by M.

Next, the binarization error (EB) 51 generated in the binarization process is
Calculate. The binarization error calculating means 5 is composed of 8 subtracters, and performs a subtraction process between the corrected image signal (CI) 31 which the correction means 3 self-cuts and the binarized image signal (Bl) 41 to generate a binarization error. A signal 51 is calculated. That is, the binarization error (EB) is expressed as EB=CI-B I
...Find as (1).

The weight error calculation means 7 is composed of a multiplier, and the binarization error 5
1 and the weighting coefficient stored in the weighting coefficient storage means 6 are input and multiplied, and a weighting error 71 is output. The weighting coefficient storage means 6 stores the aforementioned four weighting coefficients (A=7/16,
B=1/16, C=5/16, D=3/16) in accordance with the corresponding positions of the four surrounding pixels of the pixel of interest. The weight error of 4 pixels is e
p, = A X E B, =
(2) e B= B X E B
-(3)e(=CXEB---(
.

It is obtained as eD=DxEB-(5). In this case, e13 is eB =EB - (eA
It may also be obtained as +e(+eD)-(8)9. Then, each weight error is stored at a corresponding position in the error storage means 9. Error storage means 9
is composed of line memory for two lines. The image correction signal 91 is a signal read from the "*" position of the error storage means 9. At the position of "*" in the error storage means 9, weighted errors for four pixels that have already been processed are stored.

Meanwhile, in parallel with the above operation, an identification signal 111 for identifying the type of image is calculated. The identification signal 111 is obtained by determining the maximum density difference as feature amount information in a local area including the pixel of interest from the image information output from the line buffer 1, and
Based on this feature amount information, it is determined whether the image information of the local area exhibits characteristics specific to the Literature Department or characteristics of the Photography Department, and an identification signal 111 is output.

That is, the feature amount calculation means IO calculates the maximum image density within a (4×4) area including the pixel of interest (pixel indicated by diagonal lines) from the image information output from the line buffer 1, as shown in FIG. Find the value and minimum value. Next, these are subtracted to extract the maximum density difference signal in the 0 (4×4) area, that is, the feature amount information. Therefore, the line buffer 1 needs to be composed of line memories for four lines. The comparator 11 uses this feature information and a preset threshold Th2.
If the maximum density difference signal is larger than Th2, "1" representing a character is outputted as an identification signal, and if it is smaller, "0" representing a photograph is outputted.

Now, the above-mentioned feature amount information is obtained as follows.

FIG. 3 is a block diagram showing an example of the configuration of the feature value calculating means 10. As shown in FIG. As shown in FIG. 5, this feature quantity calculation means 10 calculates the maximum value and minimum value of the density within a (4×4) pixel area including the pixel of interest, respectively, for the pixel of interest in the image to be processed, as shown in FIG. , and subtract them to find the maximum density difference.

First, the feature value calculation means (20) receives image information (8 bits/8 bits/8 bits) that is sequentially input from the line buffer 1 in the column direction in units of 41 pixels in synchronization with the clock CLK, as shown in the timing chart of FIG. 4, for example. pixel)
The comparator fob and IOC through the selector 10a.

It is distributed sequentially to 10d and 10e. Note that the comparator 1 uses the selector loa of the image information input column by column.
The distribution to 0bSlOc, 10d, and 10e is controlled by selection signals SE1 and SE2 from a 2-bit counter LOh that operates in response to a clock CLK.

And comparators 10b, 10c, lOd, 10e
The image information is compared in the column direction in units of four pixels, and the maximum density (MAX terminal output) and minimum density (MIN terminal output) in that column are respectively determined.

The next stage comparator 10f, Log is the above comparator 10b110
The signals from c % 10d and 10e are input at FTRI timing, and the maximum and minimum values are further determined from the maximum and minimum values determined in the column direction. As a result of the above comparison process, the maximum density value D max within a 4×4 pixel area is shown in FIG.
and the minimum value D min are respectively determined and output at the timing of FTR2.

The subtracter 101 calculates the maximum density difference ΔD max ΔD max = D max −D mtn which is the difference between the maximum value D max and the minimum value D 111n of the density determined in this way.
- (7) is sought.

The above-mentioned comparator 11 determines the type of image based on the feature information obtained in this way, that is, the maximum density difference ΔD maX, and outputs an identification signal 111.

Based on the identification signal 111, the selection circuit 12 determines whether or not to select the image correction signal 91 read from the error storage means 9 described above. That is, the identification signal 111
is “0”, the pixel of interest is determined to be a photographic area, and the image correction signal 91 is used as the selected image correction signal 121.
If the identification signal 111 is "1", it is determined that the pixel of interest is a character part, and "0" is output as the selected image correction signal "21". The correction means 3 uses the selected image correction signal 121 and the image signal 21 calculated by the method described above.
Perform the addition of . This means that if it is a photographic pixel, correction processing 3 is performed and binarization processing is performed using the error diffusion method, whereas if it is a text pixel, correction processing is not performed and simple binarization processing is performed. do.

According to the image processing device configured in this way, when a pixel of interest is binarized, it is possible to select whether or not to correct the pixel of interest based on the binarization error of surrounding pixels, depending on the characteristics of the image information. As a result, it is possible to suppress the deterioration in resolution of character images seen in the conventional "error diffusion method", and it is possible to perform binarization processing that simultaneously satisfies the resolution of character images and the gradation of photographic images. It is possible to do so. In addition, the binarization error that occurs when character pixels are simply binarized is not stored as a weighted error by being discarded. It is designed so that it is not distributed to This prevents image quality from deteriorating at the boundary between the text and image portions of a document image in which text and photo portions coexist, making it possible to obtain a clear image.

Note that the present invention is not limited to the above embodiments.

For example, the area surrounded by reference node 4 from which feature quantities are extracted is not limited to (4×4), and the range may be changed as appropriate.

Further, in the above embodiment, an example was shown in which a photograph/character is identified in units of one pixel, but identification may be performed in units of (N x N) blocks (an integer N22). As a result, it is possible to improve the processing speed and realize a high-speed image processing device.

In the above embodiment, the maximum density difference ΔDma
Although we have shown an example using Features with different properties may be used.

Furthermore, although the above embodiment describes the case where binary output is obtained, multi-value output is also possible by setting a plurality of threshold values Thl. Gradation expression becomes possible. Furthermore, in the present invention, the value of special 5 trace amount and the judgment threshold are calculated based on the image signal read by the reading means, that is, the amount corresponding to the reflectance of the image information, and this amount is calculated based on the image density. Identification may be performed using a converted value (logarithm of the reciprocal of the reflectance) or further based on a converted signal that takes human visual characteristics into consideration.

[Effects of the Invention] As detailed above, according to the present invention, even if image information includes a text area and a photo area, the image quality can be improved by performing binarization processing according to the characteristics of each image. In addition to improving image quality, it is possible to prevent image quality from deteriorating even near the border between text areas and photo areas.Furthermore, by performing processing according to the characteristics of the image, processing efficiency in various image processing can be improved. It is possible to provide an image processing device that can perform the following functions.

[Brief explanation of drawings]

1 to 5 show an embodiment of the present invention,
FIG. 1 is a block diagram showing the general configuration of an image processing device;
Fig. 2 is a diagram for explaining the principle of binarization processing, Fig. 3 is a block diagram showing the configuration of the feature amount calculation means, Fig. 4 is a timing chart showing the operation of the special weight calculation means, and Fig. 4 is a timing chart showing the operation of the special weight calculation means. FIG. 5 is a diagram showing the concept of a pixel region in image processing, and FIG. 6 is a diagram for explaining the principle of the conventional "error diffusion method." 1... line buffer, 2... delay means, 3...
Correction means, 4... Binarization means, 5... Binarization error calculation means, 6... Weighting coefficient storage means, 7... Weighted error calculation means, 8... Error truncation means ( processing means), 9.
...error storage means, 10...feature extraction means, 11.
... Comparator (identification means), 12... Selection circuit (selection means).

Claims (1)

[Claims] Binarization means for binarizing image information of a pixel of interest in an image to be processed, and binarization from the information binarized by the binarization means and the image information of the pixel of interest. A binarization error calculation means for calculating an error; a weighting coefficient storage means for storing a weighting coefficient for calculating a weighting error; and a binarization error calculation means for using the weighting coefficient stored in the weighting coefficient storage means. weighted error calculation means for calculating a weighted error by multiplying the calculated binarization error; and feature amount extraction means for extracting a feature amount representing a feature of an image from image information within a predetermined range including the pixel of interest; processing means for truncating the weight error calculated by the weight error calculation means according to the feature extracted by the feature extraction means; and an error for storing the weight error truncated by the processing means. a storage means; a selection means for selecting whether or not to perform correction based on the weighted error stored in the error storage means according to the feature quantity extracted by the feature quantity extraction means; An image processing apparatus comprising: a correction means for correcting image information of the pixel of interest when correction is selected.