JP4506954B2 - Image processing apparatus, image processing method, image processing program, and storage medium - Google Patents

Description

  The present invention relates to a technique for improving image quality when an input binary digital image is printed out or displayed on a printer or a display device.

  In an apparatus with a small number of colors that can be expressed, such as a printer, a low-resolution display, and a portable terminal, a pseudo-gradation image is generated by reducing the number of colors to a desired number by a pseudo-gradation process (screen process). Among pseudo gradation images, binary images are widely used in various devices such as printers and facsimiles, and the amount of data is much smaller than multi-value images (continuous gradation images). Is also suitable. Also, with the recent digitization of documents, it is widely performed to scan paper documents and store them as digital documents. Even in such a case, a binary image with a small amount of data is often used. Furthermore, documents and images intended only for printing are more resistant to tampering and the like when transmitted after being converted into a binary image than when the original data is transmitted. There are many advantages. From such a background, binary images are widely used.

  By the way, with the recent wide distribution of electronic documents, it is generally required to display or print digital images with high quality, and this requirement is the same for binary images. For example, high-definition printing of binary images such as scanned documents, facsimile images, maps, topographic maps, GIS (geographic information system) information, drawing images, or on-demand printing is one of the important requirements.

  In order to display or print with high quality, a technique for improving the image quality of a binary image is required. In particular, an enlargement technique for outputting a high-quality image to an output device having different resolutions is important. Therefore, a conventional method for enlarging a binary image will be described below.

  There are two types of binary images. One is an image such as a character or a line drawing, and these are images in which each pixel has a meaning. Such a binary image is hereinafter referred to as a character / line image. The other is an image obtained by subjecting a multi-value image to pseudo gradation processing. In this case, each dot does not have a meaning, but a set of dots represents the density of the image. Such a binary image is hereinafter referred to as a halftone image or a halftone image. Note that a character / line image may also be obtained by performing pseudo gradation processing on a multi-value image. In such a case, a simple binarization method in which, for example, less than a predetermined density is converted to 0 and a predetermined density or more is converted to 1 in a multi-value image, or a similar method is often used.

  There are various types of character line drawing images such as scanned documents, facsimile images, map images, topographic map images, and drawing images. On the other hand, many halftone images are binarized photographs. These classifications should be performed not by the contents of the image but by the method for halftone processing. For example, halftone images can be classified by a halftone dot method, an error diffusion method, or the like. As described above, since the meaning of the image itself is different between the character / line image and the halftone image, different processing is often performed when image processing is performed. Note that the above-described scanned document, facsimile image, and the like may include a character line image and a halftone image in one image. In such a case, a character / image separation unit is provided to There may be a case where a separate processing is used.

  When the above-described scanned document, facsimile image, map, topographic map, GIS (geographic information system), drawing image, or the like is provided as a binary image, it is often a character line image. This is because these images need to keep the detail information faithful, and the halftone image in which the detail information is collected in a certain range and converted into one shade information is not suitable for the purpose.

  When a binary character / line image is output to a binary high-resolution output device such as a printer, the output capability at a higher resolution than that of a multi-value image can be enjoyed. This is because, in the case of a multi-value image, even if the resolution is increased to the resolution limit of the printer, the actual resolution is reduced to the screen line number by the halftone process. On the other hand, in the case of a binary character line drawing image, it can be output at the resolution of the printer resolution limit itself. In other words, it can be said that a binary image is prone to image degradation unless resolution conversion according to the output device is appropriately performed. For example, when outputting a 400 dpi multi-valued image and a binary image to a 600 dpi printer, the former has almost no difference in image quality even when an arbitrary resolution conversion method is used, whereas the latter is a resolution conversion method. The difference clearly appears in the image quality. From these points, the method for enlarging the binary character line drawing image needs to be performed more precisely.

  As a method for enlarging a character / line image, there are a method of performing a binary image as it is and a method of performing a multi-valued image. The method for enlarging a binary image is often used when the input / output destination resolution is fixed. For example, Patent Document 1 describes a method of interpolating a black pixel or a white pixel based on a predetermined rule in a portion where pixels of different colors (black and white) are adjacent in a binary image. Yes. In such an enlargement method, the resolution conversion is performed with the binary image as it is, so the processing is fast, but it is often limited to a fixed resolution, and there is a problem in using it widely. In addition, although the image can be enlarged, since the image is not appropriately anti-aliased, there is a problem that the image cannot be enlarged with high image quality when the image is further enlarged by another enlargement method.

  As a method for performing enlargement after multi-value conversion, for example, in Patent Document 2, when a binary image is multi-valued, the relationship between the target dot and the surrounding pixel values in the original image based on a weighting factor set in advance. Describes a method for determining a multi-valued pixel value from. When a binary image is enlarged after being multi-valued by a method other than simple multi-value conversion, loss of detailed information may occur when multi-value conversion is performed, and detail image quality deterioration may occur. Even if an anti-aliased image is obtained, it is often not appropriate, and it cannot be enlarged with high image quality when the image is further enlarged by another enlargement method. In addition, in the case of enlarging after multi-valued by a simple multi-valued method, a method of enlarging a multi-valued image is generally used, but in a method of enlarging a multi-valued image such as a natural image, In general, there is a problem that it is not possible to remove jaggies in a simple multi-valued image.

  To sum up the above, when a binary character / line image is enlarged as it is, the anti-aliasing process cannot be performed even if it can be enlarged, and it is therefore difficult to enlarge the image quality to an arbitrary magnification. Further, when the binary image is enlarged after being multi-valued, the loss of detailed information occurs when the multi-value is made, and image quality deteriorates. Further, as a general problem of these conventional enlargement methods, there is a problem that an anti-aliasing effect suitable for enlarging the image quality using a different method cannot be obtained.

  In addition, there is a method of combining techniques described in Japanese Patent Application Nos. 2002-079388 and 2003-330076 filed earlier by the applicant of the present application with regard to improving the image quality and expanding the image quality of general images. Using this method, natural images can be enlarged with high image quality. In addition, the binary image can be anti-aliased or enlarged by using this method, and the image quality can be improved. However, since binary images are more jaggy than multi-valued images, jaggies may remain even if this method is used, and there are problems such as loss of fine lines as a result of anti-aliasing on the texture part. there were.

JP 2000-268166 A JP 2001-268351 A

  The present invention has been made in view of the above-described circumstances. An image processing apparatus, an image processing method, and an image processing program that can appropriately perform anti-aliasing while maintaining detailed information of a binary image and can enlarge the image quality. The object is to provide a storage medium storing the image processing program.

  The present invention relates to an image processing apparatus and an image processing method for obtaining an enlarged anti-aliased image by processing a binary original image represented by a large number of pixels, and enlarging the original image by n times (for example, 2 times) Further, multi-level n-fold original images are obtained by simple multi-leveling. It is possible to perform dot correction on a predetermined part of the n-fold original image in advance to improve image quality by subsequent processing.

  A plurality of domain block images are extracted from the n-fold original image in the size of the first block unit D1. Further, a range block image is extracted from the n-fold original image so as to include a domain block image. Further, an improved domain block image of the same first block unit as the domain block image is obtained from the extracted range block image according to the domain block image.

  The improved domain block image reduces the extracted range block image to the first block unit to create a reduced range block image, and approximates the pixel value of the generated reduced range block image to the pixel value of the domain block image by the least square approximation By doing so, an improved domain block image having a size of the first block unit arranged at a position corresponding to the domain block image is created. In this case, the least square coefficients a and b in the case of approximating the pixel value of the reduced range block image to the pixel value of the domain block image are obtained, and the least square coefficients a and b are used to calculate the reduced range block image. An improved domain block image is created by converting the pixel value by least square approximation of the pixel value to the pixel value of the domain block image by the linear expression az + b. Thereafter, the edge enhancement processing can be performed based on the relationship between the maximum value and the minimum value of the pixel values in the improved domain block image and the edge degree.

  When the domain block image extracted from the n-fold original image is extracted so as to have a common part on the n-fold original image, the improved domain block image also includes overlapping pixels. In this case, a division process corresponding to the degree of overlap is performed on the pixels where the respective improved domain block images overlap.

  Furthermore, after such a process is completed for all domain block images, the output image can be replaced with an n-fold original image, and the process can be repeated a predetermined number of times from the domain block image extraction process.

  Note that the pre-processing by dot correction performed on the n-fold original image, for example, determines the dot connection direction according to the number of colors of surrounding pixels for the 2 × 2 checkered pattern in the original image, and the n-fold original In the image, dots can be arranged in a checkered portion of a 2n × 2n pattern corresponding to the 2 × 2 checkered pattern so that a line is connected in the connecting direction of the dots. Also, the dot connection direction and the connection length of the edge portion in the original image are calculated, and the density value of the connection portion of the dot at the position corresponding to the edge portion in the original image is corrected in the n-fold original image. Can do.

  Further, when extracting a range block image, one or more range block images are extracted from one domain block image, each range block image is reduced, and a plurality of reduced range block images are extracted. Of these, the reduced range block image determined to have the highest similarity to the domain block image can be selected to create an improved domain block image. At this time, when the similarity value determination unit approximates the pixel value z of the reduced range block image to the pixel value of the domain block image by the linear expression az + b, the reduced range block image has the smallest least square error. Can be determined to be the most similar reduced range block image having the highest similarity.

  Further, post-processing such as edge enhancement processing can be performed on the output improved domain block image. When performing edge enhancement processing, an edge is added to the improved domain block image based on the relationship between the maximum and minimum pixel values in the improved domain block image (for example, the average value) and the edge degree of the improved domain block image. Emphasis processing can be performed. At this time, for example, according to the above-described average value and edge enhancement strength, a pixel value larger than the average value is converted to a larger value according to the distance from the average value, and a pixel value smaller than the average value is converted into the distance from the average value. The conversion can be made smaller depending on The intensity of edge enhancement at this time can be set to be strong when the standard deviation value of the pixel value in the domain block image is small and weak when the standard deviation value is large.

  The present invention also provides an image processing program for causing a computer to execute image processing for performing image quality improvement processing as described above. Furthermore, a storage medium storing such an image processing program is provided.

  As is clear from the above description, according to the image quality improvement processing of the binary image according to the present invention, the local collage processing or the equivalent processing with a smaller decimal size block can be performed on the original image. It is possible to appropriately correct jaggies while keeping image details faithful. In addition, since the anti-aliasing process and the double enlargement process are executed simultaneously, there is no loss of information due to the anti-aliasing of the binary image. Since the present invention is an image quality improvement process based on local self-similarity, if enlargement is further performed using local self-similarity in the subsequent stage, the occurrence of jaggies can be suppressed and the image quality can be increased arbitrarily. According to the present invention, there are such effects as well as various effects as described above.

  FIG. 1 is a block diagram showing an example of a system including an embodiment of the present invention. In the figure, 1 is an image input unit, 2 is a memory unit, 3 is an image output unit, 4 is a control unit, 5 is an image quality improvement processing unit, 51 is an initial enlargement processing unit, 52 is a preprocessing unit, and 53 is a domain block extraction. , 54 is a domain block classification unit, 55 is a range block extraction unit, 56 is a reduced range block creation unit, 57 is a similarity determination unit, 58 is an improved domain block creation unit, 59 is an edge enhancement processing unit, and 60 is an average value An operation unit 61 is an iterative processing unit. 1 may exist as a processing program module on a computer, or a part or all of them may be configured by dedicated hardware. Good.

  The image input unit 1 inputs binary image data digitized from outside the system shown in FIG. 1 and transfers it to the memory unit 2. The memory unit 2 holds the image data input from the image input unit 1, the image data after the image quality improvement, and various intermediate calculation results and processing parameters used for the image quality improvement processing. The image quality improvement processing unit 5 performs image quality improvement processing on the processing target image held in the memory unit 2. The image output unit 3 outputs image data whose image quality has been improved by the image quality improvement processing unit 5 to the outside of the system shown in FIG. The control unit 4 controls processing of each unit.

  The image quality improvement processing unit 5 includes an initial enlargement processing unit 51, a preprocessing unit 52, a domain block extraction unit 53, a domain block classification unit 54, a range block extraction unit 55, a reduced range block creation unit 56, a similarity determination unit 57, and an improvement. It includes a domain block creation unit 58, an edge enhancement processing unit 59, an average value calculation unit 60, an iterative processing unit 61, and the like. The improved domain block image output from the image quality improvement processing unit 5 is written in the memory unit 2. At this time, the pixel value of the new improved domain block image is added to the previously written value for the same pixel. And update the value. The outline of each part will be described below. Details will be described later.

  The initial enlargement processing unit 51 enlarges the input binary original image by, for example, n times by the nearest neighbor method, and further performs simple multi-value conversion here to obtain a multi-value n-times original image. The enlargement method is arbitrary, and is not limited to the nearest neighbor method. Further, in the following description, for simplification, it is assumed that the image is enlarged twice, and the n-fold original image is described as a double-fold original image. Further, although the multi-value conversion method is arbitrary, it is assumed here that a simple multi-value conversion method in which binary values 0 and 1 are associated with the multi-value minimum value and maximum value as described above is adopted. . Note that the minimum value and the maximum value of the multivalue will be described as 0, 255 if necessary in the following description, but are not limited thereto.

  The preprocessing unit 52 performs dot correction on a predetermined part of the double original image obtained by the initial enlargement processing unit 51. Specifically, the line connection direction is corrected by placing small dots in the checkered portion of the double original image corresponding to the 2 × 2 checkered pattern of the original image. Furthermore, the processing is repeated until the image quality is sufficiently improved by the iterative processing unit 61 in the subsequent stage. In order to increase the speed of the repeated convergence, the edge portion of the double original image corresponding to the edge portion of the original image It is also possible to provide an effect of correcting the pixel value.

  The domain block extraction unit 53 extracts a domain block image with a size of the first block unit D1 from the double original image. The block unit means a general term for an entire arbitrary block determined by the block size. In addition to dividing the original image so that the domain block images do not overlap, the domain block image to be extracted can be extracted so as to have a common part on the original image. If the original image is a color image, the range block image is extracted from the relatively same position for each color component of the original image.

  The domain block classification unit 54 classifies the types of domain block images extracted by the domain block extraction unit 53 as necessary. The classification can be performed based on, for example, the standard deviation and the unevenness degree of the domain block image. As a kind to classify, it can classify into a flat part, a step edge part, a fine line part, a texture part, etc., for example. Further, according to the result of the classification, in the case of the thin line portion, the third block unit D2 below the first block unit D1, and in the case of the texture portion, the fifth block unit D3 below the third block unit D2. A plurality of divided domain block images are obtained from the respective domain block images. In the subsequent processing, this divided domain block image is treated as a domain block image.

  It should be noted that image quality deterioration such as jaggies mainly occurs in the step edge portion, the thin line portion, and the texture portion. Therefore, if the processing for improving the image quality is performed on these, it is possible to improve the image quality as a whole. For this reason, for example, in a flat part, it is possible to control to output the domain block image of the double original image as it is as an improved domain block image without performing the following processing, thereby shortening the processing time. In addition, the third block unit D2 and the fifth block unit D3 can be the same size as the first block unit D1, and in that case, as an effect, the divided domain block can be obtained simply by classifying the blocks. It is possible to prevent the creation of.

  Note that the above classification is an example, and may be further subdivided, or may be further classified if the processing target is narrowed down. In addition, if the control of the processing based on the classification of the domain block images is not performed, the domain block classification unit 54 can be configured without being provided.

  The range block extraction unit 55 extracts the range block image from the double original image in the second block unit R1 larger than the first block unit D1 with respect to the domain block image extracted by the domain block classification unit 54. Further, when the domain block classification unit 54 extracts the domain block image (divided domain block image) in the third block unit D2 or the fifth block unit D3, the fourth larger than the third block unit D2 respectively. The range block image is extracted from the double original image in the sixth block unit R3 larger than the second block unit R2 and the fifth block unit D3. Here, the range block image to be extracted is extracted so that the domain block image includes the domain block image, and the divided domain block is extracted so as to include the divided domain block image. At this time, a plurality of range block images may be extracted for one domain block image or divided domain block image.

  The reduced range block creation unit 56 converts the range block images of the second block unit R1, the fourth block unit R2, and the sixth block unit R3 extracted by the range block extraction unit 55 into the first block unit D1, The size is reduced to the size of the third block unit D2 and the fifth block unit D3, and a reduced range block image is created. The reduction method is arbitrary. When a plurality of range block images are extracted by the range block extraction unit 55, each range block image is reduced.

  The similarity determination unit 57 includes a reduced range block image reduced by the reduced range block creation unit 56, a domain block image extracted by the domain block extraction unit 53 (including a divided domain block image extracted by the domain block classification unit 54), and The degree of similarity is determined, and information on the degree of similarity is passed to the improved domain block creation unit 58. Although the determination of the similarity is arbitrary, for example, the least square error when the pixel value z of the reduced range block image is approximated to the pixel value of the domain block image by the linear expression az + b can be used. The least square coefficients a and b at this time can be configured to be passed to the improved domain block creation unit 58. When a plurality of range block images are extracted, a reduced range block image having the highest similarity with the domain block image is selected. When using the least square error as described above, a reduced range block image having the smallest minimum square error may be selected as the most similar reduced range block image. Further, when only one range block image is extracted, it is also possible to directly pass the least square coefficients a and b to the improved domain block creation unit 58 without determining the similarity.

  The improved domain block creation unit 58 performs pixel value conversion on the reduced range block image created by the reduced range block creation unit 56 based on information about the similarity obtained by the similarity determination unit 57, and The result of value conversion is output as an improved domain block image. For example, when the least square error is calculated by the similarity determination unit 57, the pixel value of the reduced range block image (the most similar reduced range block image) is obtained by using the linear expression az + b using the obtained least square coefficients a and b. An improved domain block image is created by converting with. When the similarity is not determined, the pixel value z of the reduced range block image may be directly approximated to the pixel value of the domain block image by the linear expression az + b, or the reduced range block image The image itself may be an improved domain block image.

  The edge enhancement processing unit 59 has a relationship between the maximum value and the minimum value of the pixel values in the improved domain block image with respect to the improved domain block image corresponding to the domain block image classified in the domain block classifying unit 52 other than the flat part. To perform edge enhancement processing. Specifically, the average value of the maximum pixel value and the minimum pixel value in the improved domain block image is taken, and the pixel value larger than the average value is converted to a larger value according to the distance from the average value according to the edge enhancement strength. The pixel value smaller than the average value can be converted to be smaller according to the distance from the average value. The intensity of edge enhancement at this time can be set to be strong when the standard deviation value of the pixel value of the domain block image is small and to be weak when the standard deviation value is large. In the case where the domain block classification unit 54 is not provided, the edge enhancement process may be performed only when the standard deviation value of the pixel value of the domain block image is a predetermined value or more.

  When the domain block extraction unit 53 extracts the domain block image and the domain block classification unit 54 extracts the divided domain block image, the average value calculation unit 60 calculates the average value for pixels in which the improved domain block images after processing overlap. Is calculated. This calculation adds each pixel value of the improved domain block image to the memory at the time of creating the improved domain block image, and at the same time, the overlap value storage memory prepared separately in the memory 2 for each overlap. After 1 is added and the processing is completed for all the domain blocks, the added pixel value can be divided by referring to the value in the overlap amount storage memory. Alternatively, when the overlap amount is known in advance from the position in the original image, the pixel value of the improved domain block image may be divided by the overlap amount and then added to the pixel value in the memory 2 and rewritten.

  The iterative processing unit 61 repeats the processing after the preprocessing unit 52 or the domain block extraction unit 53 again using the image quality improved image obtained by the end of the average value calculation unit 60 as a double original image. The iteration may be performed a predetermined number of times, or the feature amount may be extracted from the obtained image quality improved image and the end determination may be automatically performed.

  Note that the image quality improvement processing unit 5 of the present invention targets a binary character / line image. For this reason, a processing unit for determining whether or not the input image is a binary character line drawing image may be provided before the processing starts. If the input image is not a binary character line drawing image, the processing is performed. May be terminated.

  Hereinafter, several examples of the image quality improvement processing unit 5 which is a main part of the present invention will be described in detail with reference to the drawings. In each example, it is assumed that the input original image is a binary character line drawing image.

  First, the first example will be described. In this first example, the first block unit D1, the third block unit D2, and the fifth block unit D3 are assumed to be equal, and a plurality of range block images are extracted from one domain block image. An example of selecting an optimal reduced range block image is shown.

  FIG. 2 is a flowchart showing an example of the operation of the image quality improvement processing unit 5 according to the embodiment of the present invention. FIG. 3 is a specific example of the original image and the double original image until the end of the processing of the preprocessing unit 52. FIG. 4 is a schematic diagram for explaining an example of processing after the domain block extraction unit 53. As a preparation before starting the operation, a memory area for storing an 8-bit value image in the vertical and horizontal sizes of the original image to be subjected to the image quality improvement processing is secured in the memory unit 2, and the initial value is 0. Keep it. This is performed in order to add and write the improved domain block images (including those subjected to edge enhancement processing) output from the image quality improvement processing unit 5.

  In S101, the initial enlargement processing unit 51 enlarges the original image to a double size image by, for example, the nearest neighbor method, and further performs simple multi-value processing to obtain a double original image. That is, after the original image is enlarged twice by the nearest neighbor method, 0 is converted to 0 and 1 is converted to 255 to obtain a double original image. By the processing of the initial enlargement processing unit 51, for example, the original image shown in FIG. 3A becomes as shown in FIG. In FIG. 3B, black pixels and white pixels are shown as in FIG. 3A, but the pixel values are converted to 0, 255 because they are multi-valued.

  In S102, the pre-processing unit 52 corrects the double original image. FIG. 5 is a flowchart showing an example of the operation of the preprocessing unit, and FIG. 6 is an explanatory diagram of a specific example of a pattern corrected by the preprocessing unit. Here, an example is shown in which the 2 × 2 checkered pattern of the original image is corrected. In S121, a 2 × 2 pixel block is extracted from the original image, and it is determined in S122 whether the 2 × 2 pixel block is a checkered pattern. If it is not a checkered pattern, the 2x original image corresponding to the 2 × 2 pixel pattern is not corrected. For example, in the case of a checkered pattern such as the 2 × 2 pixel pattern in the center of FIG. 6A, the number of white pixels and black pixels in the 4 × 4 pixel pattern including the 2 × 2 pixel pattern in S123. Count the number.

  In S124, the counted number of white pixels and black pixels is determined, and one or more small dots are arranged in the corresponding portion of the double original image so that the smaller color pixels are connected. In this example, cases are classified according to whether black pixels are 6 pixels or less, white pixels are 6 pixels or less, or other than 4 × 4 (= 16) pixels. If the number of black pixels is 6 pixels or less, it is determined that a black line exists, and in S125, black dots are arranged at the black step portion of the double original image. The example shown in FIG. 6 corresponds to this case, and a double original image obtained by enlarging the original image shown in FIG. 6A twice is as shown in FIG. 6B. In the case of FIG. 6A, since the number of black pixels is 6 pixels and the number of white pixels is 10 pixels, the condition that the number of black pixels is 6 pixels or less is satisfied, and as shown in FIG. Two black dots are arranged. As a result, the image can be corrected to an image in which black lines are connected obliquely.

  Similarly, when the number of white pixels is 6 pixels or less, it is determined that a white line is present, and two white dots are arranged in the white step portion of the double original image in S126. In other cases, there is a possibility of erroneous determination, so correction processing is not performed here. Note that the determination condition is not limited to this example, and it can be configured to consider the arrangement of black pixels and white pixels in addition to adjusting the number of pixels. Further, the number and positions of dots to be arranged may be determined as appropriate.

  In S127, it is determined whether or not processing has been performed for all 2 × 2 pixel blocks. If an unprocessed 2 × 2 pixel block remains, the process returns to S121 to perform processing for the next 2 × 2 pixel block. . When the processing is completed for all 2 × 2 pixel blocks, the correction processing for the 2 × 2 checkerboard pattern in the preprocessing unit 52 is terminated. In addition to the checkerboard pattern as described above, the same correction processing can be performed on various patterns that can obtain better results if they are corrected in advance.

  By the processing of the preprocessing unit 52 as described above, the double original image shown in FIG. 3B is corrected as shown in FIG. The diagonal black lines are continuous and the white line portion between the black lines is also diagonally missing. Such correction processing is performed in order to prevent the occurrence of problems in the subsequent image quality improvement processing. In the subsequent image quality improvement processing, local collage processing using local self-similarity that the local region has the property of the local region is performed. However, the above-mentioned checkered pattern has local self-similarity. It is a part that collapses, and by correcting this part, it is possible to perform a good local collage process.

  Returning to FIG. 2, in S <b> 103, the domain block extraction unit 53 extracts a domain block image from the original image. FIG. 7 is an explanatory diagram of an example of a domain block image extracted by the domain block extraction unit 53. Here, the size of the first block unit D1 is 3 × 3 pixels, and a domain block image having a size of 3 × 3 pixels is extracted. In FIG. 4A as well, a 3 × 3 pixel domain block image is indicated by a bold line. Here, domain block images having a size of 3 × 3 pixels shifted by one pixel from the end of the original image are sequentially extracted. Then, the extracted domain block image is processed, the image quality is improved by the following processing, and the created improved domain block image is added to the position corresponding to the domain block image in the memory unit 2 and written. In the following description, processing for one domain block image will be described.

In S104, the domain block classification unit 54 classifies the types of domain block images extracted in S103. FIG. 8 is a flowchart showing an example of domain block image classification processing in the domain block classification unit 54. First, in S131, the pixel average value Dv = Σd _ij / 9 and the pixel standard deviation VDv = Σ (d _ij −Dv) ^{2 for} the pixel values {d _ij | i, j = 1, 2, 3} in the domain block image. Ask for. In S132, it is determined whether or not the pixel value standard deviation VDv obtained in S131 satisfies Sv1 <VDv with respect to the predetermined threshold value Sv1 for edge portion determination. Classify block images into edge portions. If not, the domain block image is classified as a flat part in S134. In this example, the types of domain block images are classified only into edge portions and flat portions, but they may be classified into more types as will be described later. The determination threshold value Sv1 is arbitrary and may be a fixed value or a variable value. Note that the value of the standard deviation VDx here is precisely a value called fluctuation, and a value obtained by removing the constant term and the square root from the true standard deviation value is used.

  The classification result of the domain block image is determined in S104, and the following image quality improvement process is performed for the domain block image whose classification result is an edge portion. For domain block images whose classification result is determined to be other than an edge portion, that is, in this example, a flat portion, the process proceeds to S111 without performing image quality improvement processing, and the domain block image is directly used as an improved domain block image as a memory. Addition writing is performed in the memory area for storing the image after the image quality improvement of the unit 2.

  In S105, the domain block image is divided into smaller blocks according to the result classified in S104, and a divided domain block image is extracted. However, in this operation example, the domain block image is classified into only the edge portion at this time, and the divided domain block is not extracted here. Alternatively, it may be considered that a divided domain block image having the same size as the domain block image is extracted.

  In S106, the range block extraction unit 55 extracts a range block image including a domain block image from the original image. Here, as an example, the second block unit R1 larger than the first block unit D1 is set to 4 × 4 pixels. FIG. 9 is an explanatory diagram of an example of a range block image extracted by the range block extraction unit 53. Assuming that the size of the range block image is 4 × 4 pixels and the range of 5 × 5 pixels is the extraction range of the range block image, as the range block image including the domain block image having the size of 3 × 3 pixels, It is possible to extract four range block images as shown in 9 (A) to (D). 4A shows the example shown in FIG. 9B.

  In S107, all the range block images extracted in S106 are respectively reduced to the same size as the domain block image by the reduced range block creation unit 56 to obtain reduced range block images. As described above, since the size of the domain block image is 3 × 3 pixels, the range block image of 4 × 4 pixels is reduced to a reduced range block image of 3 × 3 pixels. Although the reduction method is arbitrary, a projection method is used here as an example. In particular, when the projection method (linear reduction) is used, the anti-aliasing process (jaggy correction) in the present invention is particularly effectively performed because the adjacent pixel values are reduced by weighted addition according to the area ratio. be able to.

FIG. 10 is an explanatory diagram of an example of reduction processing by the projection method in the reduction range block creation unit 56. In the reduction process by the projection method here, for example, as shown in part in FIG. 10A, the pixel value is calculated based on the area ratio when a 4 × 4 pixel image is projected onto a 3 × 3 pixel image. This is done by weighted addition. The pixel value of 4 × 4 size shown in FIG. 10B is {P _ij | i, j = 1, 2, 3, 4}, and the pixel value of 3 × 3 size after reduction shown in FIG. When {Q _ij | i, j = 1,2,3}, each Q _ij is
Q ₁₁ = (9 × P ₁₁ + 3 × P ₂₁ + 3 × P ₁₂ + 1 × P ₂₂ ) / 16
Q ₂₁ = (6 × P ₂₁ + 6 × P ₃₁ + 2 × P ₂₂ + 2 × P ₃₂ ) / 16
Q ₃₁ = (3 × P ₃₁ + 3 × P ₄₂ + 9 × P ₄₁ + 1 × P ₃₂ ) / 16
Q ₁₂ = (6 × P ₁₂ + 6 × P ₁₃ + 2 × P ₂₂ + 2 × P ₂₃ ) / 16
Q ₂₂ = (4 × P ₂₂ + 4 × P ₃₂ + 4 × P ₂₃ + 4 × P ₃₃ ) / 16
Q ₃₂ = (6 × P ₄₂ + 6 × P ₄₃ + 2 × P ₃₂ + 2 × P ₃₃ ) / 16
Q ₁₃ = (9 × P ₁₄ + 3 × P ₁₃ + 3 × P ₂₄ + 1 × P ₂₃ ) / 16
Q ₂₃ ≡ (2 × P ₂₃ + 2 × P ₃₃ + 6 × P ₂₄ + 6 × P ₃₄ ) / 16
Q ₃₃ = (3 × P ₄₃ + 3 × P ₃₄ + 9 × P ₄₄ + 1 × P ₃₃ ) / 16
Given in.

  In general, a domain block image and a range block image including the domain block image are expected to be similar (local self-similarity). When such an assumption holds, jaggy appearing at the edge portion is reduced to some extent by weighted addition. However, this is based on the premise that the domain block image and the range block image are similar. Therefore, a range block image similar to the domain block image is selected in the next step.

Returning to FIG. 2, in S <b> 108, the similarity determination unit 57 determines the similarity between the reduced range block image created from each range block image in S <b> 107 and the domain block image. Here, by performing pixel value conversion on the reduced range block image, selection of the reduced range block image having the pixel value and pattern most similar to the domain block image and acquisition of information on the similarity obtained at that time are obtained. Do. First, for the pixel values {r _ijk | i, j = 1, 2, 3, k = 1,..., 4} of all the reduced range block images, the pixel average value Rv _k = Σr _ijk / 9 and the pixels, respectively. The standard deviation VRv _k = Σ (r _ijk −Rv _k ) ² is obtained. Next, the conversion coefficients a _k and b _k when the pixel value z of the reduced range block image is approximated to the pixel value of the corresponding domain block image by the primary conversion az + b, respectively, and the conversion error E _k = Σ (d _ij seek _{-a k × r ijk -b k)} 2. In the least square method, a _k , b _k , and E _k can be directly _obtained by the following formulas. As Dv and VDv, those calculated by the domain block classification unit 54 can be used.
a _k = (Σ (r _ijk −Rv _k ) × (d _ij −Dv)) / VRv _k
b _k = Dv−a _k × Rv _k
E _k = VDv−a _k × a _k × VRv _k

Here, E = min {E _k | k∈G} is set, and a = a _k and b = b _k are set for _k where E = E _k . The reduced range block image indicated by k is the most similar to the domain block image. Actually, b _k is not necessary for determining E _k , so _{k k} that minimizes E k is obtained, and b _k may be calculated after a _k is finally determined.

  In S109, the transform coefficient a, obtained by the similarity determination unit 57 for all the pixels z of the 3 × 3 size reduced range block image selected by the similarity determination unit 57 by the improved domain block creation unit 58. The pixel value conversion az + b is performed by b. This result is an improved domain block image of a 3 × 3 pixel domain block image as shown in FIG.

  In such processing, the obtained improved domain block image is created from the range block image. However, if the domain block image and the range block image are similar as described above, the reduced range block image is an image in which jaggies of the domain block image are reduced to some extent by the reduction. Furthermore, the improved domain block image created by the improved domain block creation unit 58 is an image that is similar to the domain block image by pixel value conversion (approximate the least squares). Therefore, it can be said that the improved domain block image is an image obtained by improving the image quality of the domain block image.

  Further, in order to improve the image quality of the edge portion, the edge enhancement processing by the edge enhancement processing portion 59 can be performed. In S110, the edge enhancement processing unit 59 performs edge enhancement processing on the improved domain block image. Here, the edge enhancement processing is performed by making the pixel value larger than av larger and the pixel value smaller than av smaller than the intermediate value av of the pixel values in the improved domain block image.

FIG. 11 is an explanatory diagram of an example of edge enhancement processing in the edge enhancement processing unit 59. For the 3 × 3 pixel improved domain block image p _ij (i, j = 1,2,3) shown in FIG. 11A, the minimum pixel value minp and the maximum pixel value maxp are set to minp = min {p _ij }.
maxp = max {p _ij }
And av = (maxp + minp) / 2
l = (maxp−minp) / 2
far.

At this time, the edge enhancement pixel value q _ij (i, j = 1, 2, 3) shown in FIG. 11C is obtained by the following calculation.
q _ij = av + f ((p _ij −av) / l) · l
Here, as the edge enhancement function f, as shown in FIG. 11B, an S-shaped function centered on the origin (0, 0) is used. For example,
f (x) = x (1−x ² ) · α + x (α = 0.0 to 1.0)
And other functions can be used. The variable α is a parameter that defines the strength of edge enhancement. In this case, the variable α can take a value of 0.0 to 1.0. In the case of α> 0.5, the function does not increase monotonously, so that overshoot / undershoot is likely to occur, but the edge enhancement effect can be increased. When α = 0.0, edge enhancement is not practically performed, and a stronger edge enhancement effect is given as the α value increases. The α value may be fixed, or may be variable so as to be proportional to the inverse of the pixel value standard deviation VDv value corresponding to the strength of the edge portion. In this way, adjust so that the blurred portion (that is, the portion with a small standard deviation) is subjected to strong edge enhancement and the originally sharp portion (that is, the portion with a large standard deviation) is subjected to weak edge enhancement. Can do. Of course, other functions can also be used as f (x), and edge enhancement parameters can be determined uniquely for each function. For example, a configuration that switches whether or not to perform edge enhancement is also possible.

  In S111, the improved domain block image subjected to the edge enhancement processing as described above is added and written to the image quality improved image area in the memory unit 2 (FIG. 4F). When the edge enhancement processing is not performed, the domain extracted by the domain block extracting unit 53 when the improved domain block image itself generated by the improved domain block generating unit 58 or the classified type is a flat part. The block image itself is added and written to the image quality improvement area in the memory unit 2. Addition writing is performed by reading the pixel value of the writing area from the memory unit 2, adding the read pixel value and the pixel value of the obtained improved domain block image, etc., and writing again at the same position.

  In S112, it is determined whether or not processing has been performed for all domain block images to be extracted by the domain block extraction unit 53. If there is an unprocessed domain block image, the process returns to S103 to repeat the processing.

  When the processing is completed for all domain block images, the average value calculation unit 60 performs processing for calculating the average value for the image written in the image quality improvement area in the memory unit 2 in S113. FIG. 12 is an explanatory diagram of an example of an average value calculation process in the average value calculation unit 60. When the processing on the domain block image as described above is performed, the overlapping amount (number of additions) of the domain block image in the original image becomes a value as shown in FIG. Therefore, the average value calculation unit 60 rewrites the image obtained as a result of dividing the value stored in the image quality improved image area of the memory unit 2 by the value of the overlap amount into the image quality improved image memory, and outputs the pixel value. Update. At this time, if the pixel value is 0 or less, the pixel value is replaced with 0, and if it exceeds the maximum value (for example, 255), it is replaced with the maximum value (for example, 255).

  By performing the above-described processing, an anti-aliased double-magnified image is output. By repeating such processing a plurality of times, the anti-aliasing effect can be further enhanced. In S114, the iterative processing unit 61 determines whether or not the above-described process has been repeated a predetermined number of times. If it has not been repeated a predetermined number of times, the output double enlarged image is replaced with the double original image in S115, and the process returns to S103 to repeat the processing. If it has been repeated a predetermined number of times, the image is output to the image output unit 3 and the image quality improvement processing is terminated. Here, as an example, 5 times are designated as the number of iterations, and when 5 iterations are completed, the process is terminated.

  Note that the averaging processing by the average value calculation unit 60 may be sequentially performed and output to pixels for which all overlap calculations have been completed, thereby saving memory. In the case of a binary image, there are many flat portions (pure white or pure black). The domain block image is classified as a flat portion by the domain block classification unit 54 and extracted by the domain block extraction unit 53 without performing image quality improvement processing. In many cases, an improved domain block image is used. The domain block image without such change is not added to the value in the memory 2 in S111, the flat part is not processed at all, and instead the edge part is added to the value in the memory 2 in S111. 2 can be added one by one to the overlap amount storage area prepared in advance. In this case, when the average value is calculated in S113, the average value can be calculated by referring to the overlap amount storage area and dividing by the overlap amount. In other words, only new pixels that have actually undergone image quality improvement processing are added to the value in the memory 2 and at the same time, the amount of overlap processing of each pixel is also added, and finally each pixel is divided by the amount of overlap. In addition, for the portion where the overlap amount is 0 in the overlap amount storage area, the value of the preprocessed double original image may be output as it is. In the case of the method of performing partial addition in this way, there is a portion where the pixel value slightly changes from the result of the above example in which addition is performed for all pixels, but there is no large difference in enlarged image quality, and there are many flat portions. The effect of speeding up can be obtained.

  In the above-described image quality improvement processing, the local collage processing is performed on the edge portion to improve the image quality. In other words, a 4 × 4 pixel range block image including a 3 × 3 pixel domain block image is selected, the 4 × 4 pixels are reduced to 3 × 3 pixels, brightness adjustment is performed, and after edge enhancement, a domain block image is obtained. Replaced. When 4 × 4 pixels are reduced to 3 × 3 pixels, the edge portions where jaggy is conspicuous are averaged and added with some blurring to be anti-aliased, and further sharpened by edge enhancement processing. In the case of a binary image, there is a portion where jaggies cannot be completely removed by a single process. Therefore, as described above, the iterative processing unit 61 performs iterative processing to gradually improve the image quality of the double image. Of course, the process may be ended only once.

  Furthermore, although the domain block image and the range block image are both square, of course, either one or both may be rectangular. For example, the size of the domain block image may be 2 × 3 pixels, and the size of the range block image may be 3 × 4 pixels, or the size of the domain block image may be 2 × 3 pixels, and the range block image corresponding thereto The size may be 4 × 4 pixels. A second example is shown below as an example of changing the size of the range block image.

  A second example will be described. In the first example described above, the size of the range block image is 4 × 4 pixels, but in the second example, an example in which the size is 5 × 5 pixels will be described. The size of the domain block image is 3 × 3 pixels as in the first example. Further, in the second example, it is assumed that the similarity determination process is not performed in accordance with the search process of the range block image, and one range block image is always selected from the same position with respect to the domain block image.

  This second example differs from the first example described above in the processing of the range block extraction unit 55, the reduced range block creation unit 56, and the similarity determination unit 57, that is, S106 of the processing shown in FIG. ~ S108 processing. Hereinafter, this part will be specifically described.

  As a result of classifying the domain block image by the domain block classification unit 54, when the domain block image is classified into the edge portion, the range block extraction unit 55 extracts the range block image from the original image. FIG. 13 is an explanatory diagram of another example of the range block image extraction process in the range block extraction unit 55. In this second example, as shown in FIG. 13, a 5 × 5 block is selected so as to include the domain block image at the center position. In this case, only one range block image is extracted. Of course, if the range block image extraction range is expanded, a plurality of range block images can be searched. However, as described above, in this example, one range block image is fixedly extracted as shown in FIG. When extracting a range block image, a 5 × 5 pixel range block image cannot be taken so that a 3 × 3 pixel domain block image is included in the center in the green portion of the image. In such a case, for example, a range block image may be selected from a position shifted by one pixel.

  When the domain block image is classified as a flat portion, the range block image is not extracted, and the domain block image itself is selected as the improved domain block image. For example, the process proceeds to S111 in FIG. Addition writing is performed.

Next, the extracted 5 × 5 pixel size range block image is reduced to a 3 × 3 size reduction range block by the reduction range block creation unit 56. The reduction method here also uses a projection method. FIG. 14 is an explanatory diagram of another example of the reduction process by the projection method in the reduction range block creation unit 56. In the reduction processing by the projection method, for example, as shown in part in FIG. 14A, pixel values are weighted based on the area ratio when a 5 × 5 pixel size image is projected onto a 3 × 3 pixel size image. Add. As shown in FIG. 14B, the pixel value of the 5 × 5 pixel size range block image is {P _ij | i, j = 1, 2, 3, 4, 5}, as shown in FIG. 14C. When the pixel value of the reduced range block image of 3 × 3 pixel size after reduction is {Q _ij | i, j = 1,2,3}, each Q _ij is
Q ₁₁ = (9 × P ₁₁ + 6 × P ₂₁ + 6 × P ₁₂ + 4 × P ₂₂ ) / 25
Q ₂₁ = (3 × P ₂₁ + 3 × P ₄₁ + 2 × P ₂₂ + 2 × P ₄₂ + 6 × P ₃₂ + 9 × P ₃₁ ) / 25
Q ₃₁ = (6 × P ₃₁ + 6 × P ₄₂ + 9 × P ₄₁ + 4 × P ₃₂ ) / 25
Q ₁₂ = (3 × P ₁₂ + 3 × P ₁₄ + 2 × P ₂₂ + 2 × P ₂₃ + 6 × P ₂₃ + 9 × P ₁₃ ) / 25
Q ₂₂ = (1 × P ₂₂ + 1 × P ₄₂ + 1 × P ₂₄ + 1 × P ₄₄ + 3 × P ₃₂ + 3 × P ₂₃ + 3 × P ₄₃ + 3 × P ₄₄ + 9 × P ₃₃ ) / 25
Q ₃₂ = (3 × P ₅₂ + 3 × P ₅₄ + 2 × P ₄₂ + 2 × P ₄₄ + 6 × P ₄₃ + 9 × P ₅₃ ) / 25
Q ₁₃ = (9 × P ₁₅ + 6 × P ₁₄ + 6 × P ₂₅ + 4 × P ₂₄ ) / 25
Q ₂₃ = (3 × P ₂₅ + 3 × P ₄₅ + 2 × P ₂₄ + 2 × P ₄₄ + 6 × P ₃₄ + 9 × P ₃₅ ) / 25
Q ₃₃ = (6 × P ₅₄ + 6 × P ₄₅ + 9 × P ₅₅ + 4 × P ₄₄ ) / 25
Given in.

Next, the similarity determination unit 57 performs a similarity determination process. Since only one range block image is selected in the second example, the process of selecting the most similar reduced range block image is as follows. It is not necessary for practical use. That is, the reduced range block image itself created by the reduced range block creating unit 56 may be selected as the most similar reduced range block image. First, the pixel average value Rv = Σr _ij / 9 and the pixel standard deviation VRv = Σ (r _ij −Rv) for the pixel values {r _ij | i, j = 1,2,3} of all the reduced range block images. Ask for ² . Next, conversion coefficients a and b when the reduced range block pixel value is approximated to the corresponding domain block image pixel value by the primary conversion az + b are obtained from the following equations.
a = (Σ (r _ij −Rv) × (d _ij −Dv)) / VRv
b = Dv−a × Rv
In the second example, since the process of selecting the most similar reduced range block image is not necessary, the calculation of the least square error performed in the first example is not necessary.

  The subsequent processes of the improved domain block creation unit 58, the edge enhancement processing unit 59, the average value calculation unit 60, and the iterative processing unit 61 are the same as those in the first example described above, and thus the description thereof is omitted here. In the second example, the division amount of the average value calculation unit 60 may be stored using the memory 2. That is, only when a domain block image is classified and processed as an edge portion, a new pixel value is added, and at the same time, 1 is added to a previously prepared overlap amount storage area, and the final average value calculation unit 60 The division may be performed with reference to this overlapping area deep area when dividing. By doing so, the addition writing to the memory 2 performed in S111 regarding the flat portion can be omitted, and the speed can be increased.

  In the second example, an example is shown in which the size of the range block image, which is 4 × 4 pixels in the first example described above, is changed to 5 × 5 pixels. When the size is reduced from 5 × 5 pixels to 3 × 3 pixels, the antialiasing effect is higher than when the size is reduced from 4 × 4 to 3 × 3. Therefore, in this second example, the effect of improving the edge portion is higher than in the first example, and a good processed image can be obtained. In addition, since the 5 × 5 block including the 3 × 3 block at the center position is always selected as the most similar reduced range block image without searching for the range block image, the four of the 4 × 4 blocks are selected. Compared to the case of selecting one, uniform collage processing is performed over the entire image, and the image quality is stabilized. In particular, in the case of a binary image, this has a great effect on the image quality, and there is a case where it is better to reduce the 5 × 5 block without performing the search. Of course, the speed can be increased by not performing the search process.

  Here, the effects of the present invention described in the first and second examples are summarized. The part of the local collage processing in which a range block image including a domain block image is extracted, reduced and written as an improved domain block image as in the present invention is based on the assumption of local self-similarity of the image. However, this local collage process can improve the image quality of natural images, but in the case of a binary image, an appropriate effect may not always be obtained. The reason for this is that in the case of binary images, it is necessary to pursue local self-similarity more appropriately and appropriately because jaggy is strong and there are many thin lines. In normal local collage processing, it is sufficient to pursue local self-similarity. This is probably because it is not. For this reason, the present invention makes significant improvements in this regard.

  In the present invention, the original image is subjected to enlargement processing by the initial enlargement processing unit 51 in advance. When local collage processing is performed on an original image using an integer size block without performing the enlargement processing by the initial enlargement processing unit 51, local self-similarity is insufficient in the binary image as described above. A part arises. Therefore, in the present invention, by performing the enlargement process by the initial enlargement processing unit 51, it is possible to perform local collage processing using a smaller decimal size block on the original image and more appropriately pursue local self-similarity. is made of.

  FIG. 15 is an explanatory diagram of the relationship between the domain block image and the range block image in the original image and the double image in the first and second examples. In the figure, black pixels are indicated by hatching. In the specific example described above, the initial enlargement processing unit 51 enlarges the original image twice by the nearest neighbor method, and the 3 × 3 block size domain block image and 4 × 4 from the double original image shown in FIG. Alternatively, a 5 × 5 block size range block image was extracted. In the double original image, these block sizes are integer sizes, but in the corresponding original image (FIG. 15A), a 1.5 × 1.5 block size domain block image and 2 × 2 or 2. The range block image has a size of 5 × 2.5 blocks.

  In this way, by processing with a decimal block size smaller than the integer block size on the original image, use small blocks even for parts where local self-similarity collapses in large blocks, especially thin lines. Thus, processing can be performed while maintaining local self-similarity, and the image quality can be greatly improved.

  Since the initial enlargement processing unit 51 performs the enlargement process instead of applying the decimal size block to the original image, the anti-aliasing effect is obtained and at the same time an image enlarged twice is obtained. Can do. This not only means that anti-aliasing and double enlargement can be performed simultaneously, but also has the effect of suppressing the loss of information when anti-aliasing is applied to a binary image by enlarging the image. The effect is enhanced. Of course, when it is desired to obtain an image having the same size as the original image, a reduction process may be performed after the process.

  As described above, in the first and second examples, there is an effect similar to that when local collage processing is performed on the original image with a small-size block, and it is possible to enlarge while maintaining detailed information faithfully. it can. On the other hand, since a small block is used, there is a disadvantage that an anti-aliasing effect obtained by one processing is reduced. Therefore, in the first and second examples described above, the iterative processing unit 61 performs a process of repeating the local collage process a plurality of times to gradually increase the antialiasing effect. Iterating in this way is preferable from the viewpoint of image quality, but it is preferable to reduce the repetition as much as possible from the viewpoint that the processing speed becomes slow. Therefore, in the following third and fourth examples, a method for reducing the number of iterations will be described in detail.

  A third example will be described. In the first and second examples described above, processing is performed with all block sizes fixed. In this third example, a processing method will be described in which the block size is changed according to each part of the image, and an anti-aliasing effect is appropriately obtained in each part with a small number of iterations. In the third example, similarly to the second example, the range block image search process is not performed, and one range block image is always selected from the same position with respect to the domain block image.

  In this third example, what is essentially different from the second example described above is the processing of the domain block classification unit 54, the range block extraction unit 55, and the reduced range block creation unit 56. That is, it is the process of S104 to S106 among the processes shown in FIG. Hereinafter, this part will be specifically described.

  First, the size of the first block unit D1 is set to 4 × 4 pixels, and the domain block extraction unit 53 extracts the domain block image from the double original image.

  In the third example, unlike the first and second examples described above, the domain block classification unit 54 converts the domain block image into a flat part, a step edge part, a thin line part, an isolated point part, a tip part, and a texture part. Classify into: That is, in the first and second examples, the portion that is commonly handled as the edge portion is subdivided and classified as different portions.

  FIG. 16 is a flowchart showing another example of domain block image classification processing in the domain block classification unit 54. Here, the classification of the edge portion and the flat portion in S141 to S144 is the same as the flowchart shown in FIG. 8 in the first and second examples. If the edge portion is classified in S143, the unevenness amount of the domain block image, the number of white pixels, and the number of black pixels are calculated in the next S145. FIG. 17 is an explanatory diagram of an example of the calculation process of the unevenness amount. As the calculation of the unevenness amount, for example, as shown in FIG. 17, the previously extracted 4 × 4 domain block image (thick frame line in FIG. 17A) is placed on the original image shown in FIG. Reverse projection. The image on the original image corresponding to the 4 × 4 domain block image on the double original image is an image of 2 × 2 block size. In addition, when the position at the time of back projection is a decimal unit, the block of the position where the decimal number was cut off may be used.

The amount of unevenness can be calculated using pixels around the 2 × 2 block on the original image, for example, a 4 × 4 block image. Of the 4 × 4 block image, the center of the 2 × 2 other pixels _{(p 1 ~p 12, p 13} = p 1, each p _i 0 or 1 value) with respect to,
Concavity and convexity = Σ _{i = 1} ¹² | p _i −p _{i + 1} |
The amount of unevenness can be calculated with.

  Further, the number of white pixels and the number of black pixels in the 4 × 4 block image on the original image are counted. In this way, the unevenness amount, the number of black pixels, and the number of white pixels in the original image can be obtained from the pixels of the original image corresponding to the domain block image in the double original image and the surrounding pixels. In the example shown in FIG. 17, the unevenness amount = 4, the number of black pixels = 5, and the number of white pixels = 9.

Returning to FIG. 16, in S <b> 146, the process branches according to the unevenness value calculated in S <b> 145. First, in the case where the unevenness amount = 0, it is classified as an isolated point in S150. Next, when the unevenness amount = 4, the thin line portion is classified in S151. Next, when the unevenness amount> 4, the texture portion is classified in S152. Finally, when the unevenness amount = 2, the classification is further performed according to the difference between the number of black pixels and the number of white pixels in S147. That is, if the difference between the number of black pixels and the number of white pixels is less than or equal to a predetermined value, it is classified as a step edge portion in S148, and if it is greater than or equal to a predetermined value, it is classified as a tip portion in S149. For example, the predetermined number is 4,
| Black pixel count-White pixel count | ≤ 4
Can be classified as a step edge portion.

  FIG. 18 is an explanatory diagram of a specific example of a classification result according to another example of domain block image classification processing in the domain block classification unit 54. The results classified according to the flowchart shown in FIG. 16 are collectively shown in FIG. Note that FIG. 18 shows an image on the original image. First, in the example of FIG. 18A, there are no black pixels around the central 2 × 2 block. Accordingly, the unevenness amount becomes 0 and is classified as an isolated point. In the example of FIG. 18B, when the unevenness amount is calculated from the pixels around the center 2 × 2 block, it becomes 4. Therefore, in this case, it is classified as a thin line. In the example shown in FIG. 18C, the unevenness amount is calculated as 6 from the pixels around the center 2 × 2 block. Therefore, in this case, it is classified as a texture part. In the example shown in FIGS. 18D and 18E, since the unevenness amount is calculated as 2 from the pixels around the central 2 × 2 block, the difference between the number of white pixels and the number of black pixels is further calculated. . In the example shown in FIG. 18D, since the number of white pixels and the number of black pixels is 8, the difference is 0, and the difference is determined to be equal to or less than a predetermined value, and is classified into the step edge portion. In the example shown in FIG. 18E, the number of white pixels is 13 and the number of black pixels is 3, so the difference is 10, and it is determined that the difference is equal to or greater than a predetermined value, and is classified as the tip portion. In these examples shown in FIG. 18, each classification result would be considered valid.

  If the domain block images are classified, it is determined whether or not the domain block image is a flat portion in S104 of FIG. In the description of S105 described above, it is described only that the domain block image is divided into smaller blocks according to the classification result. In the first example, since only the edge portion is classified, no division is actually performed. In this third example, at this point, the step edge portion, the fine line portion, the isolated point portion, the tip portion, and the texture portion are classified. According to these classifications, if necessary, division into smaller blocks is performed.

  FIG. 19 is an explanatory diagram of an example of domain block image division processing and range block image extraction according to the classification result. If the classification result by the domain block classification unit 54 is the step edge part, the block unit of the domain block image remains the first block unit D1 (= 4) as shown in FIG. . In this case, the domain block image is not divided.

  In addition, as shown in FIG. 19B, the domain block images classified into the thin line portions are set to a third block unit D2 equal to or less than the first block unit D1 as a block unit. Here, the third block unit D2 = 3. In this case, there are four methods for acquiring a 3 × 3 block size divided domain block image from the original 4 × 4 block size domain block image. That is, as shown in FIG. 19B, in addition to the position shifted to the upper left, a 3 × 3 size block positioned at the lower left, upper right, and lower right can be acquired as a divided domain block image.

  Further, for the domain block images classified into the tip portion, isolated point portion, and texture portion, as shown in FIG. 19C, the block unit is a fifth block unit D3 that is equal to or less than the second block unit D3. . Here, the fifth block unit D3 = 2. In this case, there are nine methods for acquiring a 2 × 2 block size divided domain block image from the original 4 × 4 block size domain block image, each of which is acquired as a divided domain block image.

  In addition, four divided domain block images obtained from the domain block images classified into the thin line portions, and nine divided domain block images obtained from the domain block images classified into the tip portion, the isolated point portion, and the texture portion. In the subsequent processing, instead of the original domain block image extracted by the domain block extraction unit 53, it is treated as a new domain block image and the processing proceeds.

  Next, in S106 of FIG. 2, the range block image extraction processing is performed by the range block extraction unit 55. As a result of the classification by the domain block classification unit 54, in the case of the step edge part, one domain block image of the first block unit D1 is obtained, and the thin line part is four divided domain blocks of the third block unit D2, the tip 9 parts, isolated point parts, and texture parts, respectively, are extracted nine divided domain blocks of the fifth block unit D3. Therefore, the second block unit R1, the fourth block unit R2, and the sixth block unit R3 (R1> D1 with respect to the first block unit D1, the third block unit D2, and the fifth block unit D3, respectively. , R2> D2, R3> D3).

That is, in the case of the step edge portion, the range block image is extracted so as to include the domain block image in the second block unit R1. Further, in the case of a thin line portion, a range block image is extracted so as to include each divided domain block image in the fourth block unit R2 for four divided domain block images, and four range block images are obtained. Get. Further, in the case of the tip portion, isolated point portion, and texture portion, a range block image is extracted so that each divided domain block image is included in the sixth block unit R3 with respect to nine divided domain block images. 9 range block images are obtained. As the second block unit R1, the fourth block unit R2, and the sixth block unit R3, for example,
Step edge portion: second block unit R1 = 6
Thin line portion: fourth block unit R2 = 5
Tip, isolated point, texture: 6th block unit R3 = 4
It can be. When the block unit of the range block image is determined in this way, as in the second example described above, the range block is searched so as to include the domain block or the divided domain block image in the center without searching for the range block image. Only one image can be extracted for each. Therefore, the range block image is not searched.

  An example of the range block image extracted according to each classification is shown in FIG. FIG. 19A shows a case of a step edge portion, and a 6 × 6 size range block image including a 4 × 4 size domain block image is extracted. FIG. 19B shows an example in which a range block image is extracted for a 3 × 3 block at the upper left of the domain block image as a divided domain block image in the case of a thin line portion. A range block image is similarly extracted for each divided domain block image. FIG. 19C shows the case of the tip portion, isolated point portion, and texture portion, and shows an example in which the range block image is extracted for the 2 × 2 block at the upper left of the domain block image as the divided domain block image. Range block images are similarly extracted for each of the nine divided domain block images.

  The remaining processing in this third example is essentially the same as in the second example. That is, when the domain block image extracted by the domain block extraction unit 53 is classified as the step edge portion, the range block image of the second block unit R1 is reduced to the first block unit D1, and the least square approximation is performed. Adopted as an improved domain block image. In addition, when classified into the thin line portions, four range block images of the fourth block unit R2 are respectively converted into the third block unit D2 for the four divided domain blocks of the third block unit D2. Are reduced to 2 and approximated by least squares to be adopted as four improved domain block images. Further, when classified into the tip portion, isolated point portion, and texture portion, nine range blocks of the sixth block unit R3 are respectively obtained for the nine divided domain block images of the fifth block unit D3. The image is reduced to the fifth block unit D3, approximated by least squares, and adopted as nine improved domain block images. These results are all added to corresponding positions in the double original image update storage area in the memory 2 regardless of the block unit. When each range block image is reduced, a projection method similar to that shown in FIGS. 9 and 13 may be used according to the block size.

  In addition, since the average value calculation unit 60 in the third example adds the improved domain block images with different block sizes, the division processing is fixedly performed as in the first and second examples. Therefore, as described in the last part of the first and second examples, the storage area for storing the division value (the overlapping amount) for the division processing in the average value calculation unit 60 When the pixel value of the improved domain block image is added, 1 is added to the same position in the overlap amount storage region, and after the processing is finished, the pixel is set with the value of the overlap amount storage region. What is necessary is just to comprise so that a value may be divided.

  In the case of the third example, the domain blocks are classified using the original image, but it is also possible to classify from the double original image. However, since the double original image is updated each time it is repeated by the iterative processing unit 61, the classification result may change each time it is repeated. Further, as the iteration is repeated, the pixel value of the double original image takes a value of 0 to 255, which may make the classification method more complicated. In such a case, for example, when classifying a double original image, the block is reduced to 1/2 size, and further simplified binarization (0 for 127 or less, 1 for 128 or more), and the above classification method. Can also be used.

  In the third example, a large block size is used for the step edge portion, a slightly smaller block size is used for the thin line, and a small block size is used for the tip portion, isolated point portion, and texture portion. When applying, more appropriate effects can be obtained. When a large block is used in a fixed manner, the amount of improvement in the image quality at the step edge portion is large, but problems such as crushing occur at the fine line portion, the texture portion, and the like. On the other hand, when a small block size is used in a fixed manner, the image quality can be improved in all parts and there are few problems such as crushing, but the image quality improvement amount is reduced as a whole. For this reason, the effect is reduced particularly in a portion such as a step edge portion, and many iterations are required to obtain a sufficient image quality improvement. In the third example, considering these points, an appropriate block size is selected for each portion. In other words, the local self-similarity of the image is appropriately increased in each part.

  In the third example, even in comparison with the first and second examples, a higher quality image is often obtained when the same number of repetitions (for example, five times) is used. On the other hand, the process of classifying domain block images is complicated, the classification is not performed properly, an incorrect block size is selected, and the image quality changes at the boundary portion where the block size is switched. In addition, when the number of iterations is the same, the processing time tends to increase as compared to the first and second examples. Therefore, in the following fourth example, an example in which the block size is fixed as in the first and second examples, and appropriate local collage processing can be performed on each part as in the third example will be described. To do.

  A fourth example will be described. In the fourth example, processing is performed with the block size fixed as in the first and second examples, but the processing is performed with the domain block image size set to 2 × 2 and the range block size set to 4 × 4. In the fourth example, the range block image search process is not performed as in the second example, and one range block image is always selected from the position including the domain block image as the center. The fourth example is essentially different from the second example only in the preprocessing unit 52, that is, the process of S102 in the process shown in FIG. Therefore, the preprocessing unit 52 will be described below.

  The purpose of the preprocessing unit 52 in the first to third examples so far has only been to correct the part where the problem occurs in the subsequent local collage processing. In other words, it was only a process of connecting checkered portions in which local self-similarity easily collapsed in an appropriate direction. In this fourth example, in addition to this, the purpose of the preprocessing unit 52 is to add another purpose of speeding up the convergence of the subsequent local collage processing. The processing method of the preprocessing unit 52 of the fourth example will be specifically described below.

  FIG. 20 is a flowchart illustrating another example of the operation of the preprocessing unit 52, and FIG. 21 is an explanatory diagram of an example of a pattern to be processed by the preprocessing unit 52 in the fourth example. In the pre-processing unit 52 of the fourth example, scanning is performed in units of 2 × 2 pixels, and pre-processing correction is performed when 2 × 2 dots are checkered, L-shaped, and inverted L-shaped. Here, checkered, L-shaped, and inverted L-characters are in the form as shown in FIGS. 21A to 21C, and those that are reversed in black and white are also checkered, L-shaped, and inverted L-shaped, respectively. Note that the subsequent processing is exactly the same for the reverse of black and white, and the description thereof is omitted.

  In S161, a 2 × 2 pixel block image is extracted from the double original image, and in S162, it is determined whether or not the pattern corresponds to any of the checkered pattern, the L shape, and the inverted L shape shown in FIG. FIG. 22 is an explanatory diagram of an example of a checkerboard, L-shape, and reverse L-shape determination method in the preprocessing unit 52. In the three patterns shown in FIG. 21, the values of the upper two pixels of 2 × 2 are different. Therefore, when determining checkered, L-shaped, or inverted L-shaped, it is first determined whether the values of the two pixels above 2 × 2 are different, and then determine which pattern is the pattern. You may do it. For example, when both pixels are black pixels as shown in FIG. 22A, or when both pixels are white pixels as shown in FIG. 22C, checkered, L-shaped, inverted L-shaped. It can be determined that none of the patterns is applicable. In addition, as shown in FIG. 22B, when the values of two pixels are different, the checkered, L-shaped, or inverted L-shaped pattern shown in FIGS. 22D to 22F is used. There is a possibility. Therefore, as shown in FIG. 22B, when the values of two pixels are different, a checkered pattern, an L-shaped pattern, and an inverted L-shaped pattern may be detected. Even in the case of FIGS. 22A and 22C and FIG. 22B, if the pattern is not a checkered pattern, an L shape, or an inverted L shape, the process returns to S161 and the next 2 × 2 Extract a pixel block image.

  If the block image of 2 × 2 pixels is a checkered pattern, the number of white pixels of the 4 × 4 block including 2 × 2 and the black in S163, as described in the flowchart of FIG. 5 in the first example. The number of pixels is counted and compared with a predetermined value in S164. For example, if the predetermined value is 6, and if the number of black pixels ≦ 6, it is classified as L-shaped in S165, and if the number of white pixels ≦ 6, it is classified as inverted L-shaped in S166 and the process proceeds. Otherwise, the process ends.

  When classified into L-shaped and inverted L-shaped in S162, and when classified into L-shaped in S165 and inverted L-shaped in S166, using the method for L-shaped or inverted L-shaped respectively in S167, The connection direction (horizontal or vertical) and connection length of dots in the original image are extracted. Further, in S168, based on the extraction result in S167, pixels are interpolated at corresponding positions of the double original image. In step S169, it is determined whether or not processing has been completed for all 2 × 2 pixel block images. If an unprocessed 2 × 2 pixel block image remains, processing is performed on the unprocessed block image. Therefore, the process returns to S161. When the processing is completed for all the 2 × 2 pixel block images, the processing in the preprocessing unit 52 is terminated.

  FIG. 23 is a flowchart showing an example of the extraction process of the connection direction and connection length of the block image classified into the L shape in the pre-processing unit 52, and FIG. 24 is a specific example of the extraction of the connection direction and connection length and the pixel interpolation process. It is explanatory drawing of an example. For example, when it is determined in S162 and S164 in FIG. 22 that the 2 × 2 pixel block image is L-shaped, the outer side of the 2 × 2 pixel block image shown by hatching in FIG. Refer to the pixel value of. First, in S171 and S180, the dot connection direction (horizontal or vertical) is determined. As shown in FIG. 24, if the dot at the upper left of 2 × 2 pixels (the dot indicated by r1 in FIG. 22B) is white, it is assumed that the dot is in the horizontal direction, and the process proceeds from S171 to S172. If the dot at the lower right of 2 × 2 pixels (the dot indicated by q2 in FIG. 22C) is white, it is determined to be the vertical direction, and the process proceeds from S180 to S181. If both are not satisfied, the process is terminated for the 2 × 2 pixel block image.

For example, when the dot connection direction is the horizontal direction, the dot connection length is calculated for each of the left and right sides. For this purpose, i = 2 is set in S172, and it is determined whether or not p _i-1 = white and p _i = black in S173. If the condition is met, 2 is added to i in S174, and the flow goes to S173. Returning, the determination of two pixels adjacent to the left side is repeated. In this way, the pairs of p _i-1 = white and p _i = black continuous in the horizontal direction on the upper left side of the block image of 2 × 2 pixels are counted by i, and i is halved in S175. Left connection length.

Further, i = 2 is set in S176, and it is determined in S177 whether q _i-1 = white and q _i = black. If the conditions are met, 2 is added to i in S178, and the process returns to S177. The determination of two pixels adjacent to the right side is repeated. In this way, a pair of q _i-1 = white and q _i = black continuous in the horizontal direction on the right side of the 2 × 2 pixel block image is counted by i, and i is halved in S179. Left connection length.

Thereby, the connection lengths in the left and right directions are obtained. Note that the positional relationship between p _i and q _i is shown in FIG.

Similarly, when the dot connection direction is the vertical direction, the dot connection length is calculated for each of the upper and lower sides. For this purpose, i = 2 in S181, it is determined whether or not r _i-1 = white and r _i = black in S182. If the condition is met, 2 is added to i in S183 and the flow goes to S182. Return and repeat the determination of the two adjacent pixels on the upper side. In this way, the pairs of r _i-1 = white and r _i = black continuous in the vertical direction on the upper left side of the 2 × 2 pixel block image are counted by i, and i is halved in S184. The upper connection length.

Further, i = 2 in S185, it is determined whether or not s _i-1 = white and s _i = black in S186. If the condition is met, 2 is added to i in S187 and the process returns to S186. The determination of two pixels adjacent to the lower right side is repeated. In this way, a pair of s _i-1 = white and s _i = black continuous in the vertical direction on the lower right side of the 2 × 2 pixel block image is counted by i, and i is halved in S188. The lower connection length.

As a result, the connection lengths in the upper and lower directions are obtained. Note that the positional relationship between r _i and s _i is shown in FIG.

After extracting the connection lengths in the horizontal direction or the vertical direction, pixel interpolation processing in S168 of FIG. 20 is performed. When the connection direction is the horizontal direction, for example, as shown in FIGS. 24D to 24G, the pixel value corresponding to the connection length is set in the double original image with respect to the horizontal direction as the connection direction. To distribute. For example, when the connection direction is horizontal and the connection length in the left direction is k, k pixels are newly arranged at corresponding positions. The pixel values N ₁ ,..., N _k of the pixels to be distributed are
N ₁ = (0 × k + 255 × 1) / (k + 1)
N ₂ = (0 × (k−1) + 255 × 2) / (k + 1)
:
N _k-1 = (0 × 2 + 255 × (k−1)) / (k + 1)
N _k = (0 × 1 + 255 × k) / (k + 1)
Should be calculated. In other words, the density up to 255 may be evenly distributed and arranged in the L-shaped portion. Similarly, the pixels may be arranged in the right direction.

  24D to 24G show examples of dot arrangements when the left and right connection lengths are the same. When the left and right connection lengths are different, for example, when the left direction is 2 and the right direction is 4, the left direction is 2 dots (density 85 and 170) arranged on the left side of FIG. 24E, and the right direction is Four dots (density 51, 102, 153, 204) arranged on the right side of FIG. 24G, that is, a total of six dots may be arranged at corresponding positions.

  In addition, even when the connection direction is the vertical direction, it is possible to arrange dots in the same manner as in the horizontal direction. An example of the pixel in that case is shown in FIGS. Thus, the arrangement process for the L-letter is completed.

  FIG. 25 is an explanatory diagram of a specific example of connection direction and connection length extraction and pixel interpolation processing of block images classified into an inverted L shape in the preprocessing unit 52. The case where the connection direction is an inverted L-shape is the same as the case of the L-shape described above, and is determined by the right pixel of the two pixels above 2 × 2 pixels and the upper pixel of the two left pixels. It is determined whether the connection direction is the horizontal direction or the vertical direction. Then, in the left-right direction, continuity is determined for each of the two pixels shown with rough hatching in FIG. 25B, and in the vertical direction, 2 shown in FIG. 25C with rough hatching. The continuity is determined for each pixel, and each connection length is obtained. Then, according to each connection length, as shown in FIGS. 25D to 25G in the horizontal direction and as shown in FIGS. 25H to 25K in the vertical direction, the pixels Can be interpolated. Thus, the arrangement process in the case of the inverted L shape is also completed. Thus, the processing of the preprocessing unit 52 in the fourth example is completed, and the edge portion of the double original image is corrected.

  In a binary image, edges that should originally be gentle curves cannot be expressed, and the connection lengths of L-shaped and inverted L-shaped portions may be slightly shifted. When the preprocessing unit 52 arranges dots in such a portion, it is preferable to slightly increase the connection length to blur the edge. Specifically, when only the left and right or upper and lower connection lengths differ by 1 for a portion classified as L-shaped or inverted L-shaped, it is preferable to increase the connection length by one and process.

  FIG. 26 is an explanatory diagram of a specific example when the preprocessing unit 52 performs the interpolation process by increasing the connection length. For example, a 2 × 2 pixel block image surrounded by a thick line in FIG. 26A is determined to be L-shaped, and the left connection length at this L-shaped portion is 1 and the right connection length is 2. In this case, when interpolation processing is normally performed by the preprocessing unit 52, as shown in FIG. 26B, 1 dot is arranged on the left side and 2 dots are arranged on the right side. On the other hand, by increasing this connection length by 1 to the left and right and arranging 2 dots on the left side and 3 dots on the right side as shown in FIG. 26C, the number of auxiliary dots can be increased. .

  Alternatively, the number of dots to be arranged may be changed without changing the number of dots (only when the connection length is increased by one step). At this time, as shown in FIG. 26C, in consideration of the connection with the surrounding pixels, when arranging the dots on the 2-dot side, the position is shifted by one dot on the 3-dot side. To do. The shift of the start position for one dot is not limited to this example, and it is preferable to always perform the shift on the side where the connection length is short.

  The above processing is performed in the same manner for blocks that have been classified as L-shaped or inverted L-shaped from checkerboard. Since blur is added when such processing is added, the image quality at the end of the pre-processing unit 52 is not necessarily better than when no change is made. However, since the directionality of the angle of the edge portion is averaged overall due to this slight blurring, when the local collage processing is performed later, only the blurring is corrected while the angle remains uniform, and the final image quality is improved. Can be improved.

  In the same way, when the left and right or top and bottom connection lengths differ by 2 for parts classified as L-shaped or inverted L-shaped, the longer connection length is reduced by one and the shorter connection length Is increased by one, and when the dots are arranged, the start position may be shifted by one dot to the side where the connection length is longer.

  Moreover, although the process by the above-mentioned pre-processing part 52 was about L character and reverse L character, you may perform a process specially about a checkered part. The L-shaped process shown in FIG. 23 is performed under the condition that r1 in FIG. 24B is white when determining that the connection direction is the horizontal direction, for example. In this case, since r2 in FIG. 24B is not determined, it is considered that r2 is a black pixel, and the 2 × 2 pixel block image that is one dot above has a checkered pattern. In such a case, the block image on the upper side is processed by being classified into any one of L letter / reverse L letter / other three. In other words, if r2 is allowed to be black, different processing may be performed around the checkerboard, and image quality degradation may occur. In order to avoid such a problem, it is preferable that the pixel value of r2 is added on the condition that it is white. In this case, for a block image processed as a checkered pattern, a 2 × 2 block image at a position where the block is shifted in a predetermined direction by one dot may be further processed as an L shape or an inverted L shape. Similarly, in the determination in the vertical direction, it is also possible to use the condition that q1 is black in addition to q2.

  FIG. 27 is an explanatory diagram of a difference between an actual L-shape and a checkerboard determined as an L-shape. As described above, even in the case of a checkered pattern, it may be processed as an L-shape or an inverted L-shape, but the difference will be described in comparison with an actual L-shape pattern. FIG. 27A shows the case of a truly L character, and it is determined as an L character in two 2 × 2 pixel block images surrounded by a thick line. Note that the block image on the lower left side is an L-shaped image that is reversed in black and white. By performing L-shaped processing on these two block images, pixels are interpolated and corrected in the upper and lower portions of the black line extending in the left-right direction as a result.

  On the other hand, even in the checkered portion as shown in FIG. 27B, a checkered pattern is detected in the upper block image in two 2 × 2 pixel block images surrounded by bold lines, and the lower block image is black and white. An inverted L-shaped pattern is detected. Therefore, interpolation processing is performed on these block images.

  At this time, since the lower block image (L-shaped) has a white dot at a position corresponding to r2 in FIG. 24 (black and white inversion), when the above-described r2 condition is added, Processing for the L-shape is not performed on the block image. Therefore, when the checkered portion is processed, the block image of 2 × 2 pixels at the position where the block is further shifted by one dot in the predetermined direction is further processed as an L-shape or an inverted L-shape. In this case, pixel interpolation processing is performed on the lower block image as a processing target. By such partial modification, it is possible to further improve the image quality of the checkered portion.

  FIG. 28 is an explanatory diagram of an example of edge determination in the 45-degree direction, and FIG. 29 is an explanatory diagram of an example of interpolation processing for the edge in the 45-degree direction. The pre-processing unit 52 can be configured to perform interpolation processing for a 45-degree direction. The example shown in FIG. 28A shows the case of an L-shaped pattern in which black and white are reversed. In this case, when the two pixels on the right side of the 2 × 2 pixel block image are black pixels, there is a possibility that an edge of 45 degrees exists. In such a case, a search is made for a pair of two pixels diagonally at the lower right of the block image, in which the left pixel is white and the right pixel is black. This search is the same as in the horizontal and vertical directions described above. Thereby, a connection length in an oblique direction is obtained. For example, if the connection length is 0, the edge shown in FIG. 28 (B) is the edge shown in FIG. 28 (C). If the connection length is 1, the edge shown in FIG. 3 is considered to have the edges shown in FIG. In either case, if the number of pixels in the vertical direction is n, the number of pixels in the horizontal direction can be considered as an edge having an inclination of n + 1. Based on this idea, pixel interpolation in an oblique direction is performed.

Here, the pixel value (density value) to be interpolated is determined according to the area ratio divided by the edge. Since the inclination of the edge is the horizontal direction n + 1 and the vertical direction n as described above, the area ratio of the A0 portion to the entire pixel in FIG.
Entire pixel: A0 = (n + 1) ² : (n · (n + 1) / 2)
It is. Accordingly, the density value P (A0) is obtained when the density value width is 0 to 255.
P (A0) = (n · (n + 1) / 2) / ((n + 1) ² )
= (N / 2 (n + 1)) × 255
It is. Similarly, in FIG. 29, the area ratio of the A1 portion to the entire pixel is
Entire pixel: A1 = (n + 1) ² : ((n−1) ² · (n + 1) / 2n)
It is. Accordingly, the density value P (A1) has a density value range of 0 to 255.
P (A1) = ((n−1) ² · (n + 1) / 2n) / ((n + 1) ² )
= ((N-1) ² / 2n (n + 1)) × 255
It is. Generally speaking, the area ratio of the Ak part to the entire pixel is
Whole pixel: Ak = (n + 1) ² : ((n−k) ² · (n + 1) / 2n)
It is. Therefore, the density value P (Ak) has a density value range of 0 to 255.
P (Ak) = ((n−k) ² · (n + 1) / 2n) / ((n + 1) ² )
= ((N−k) ² / 2n (n + 1)) × 255
Can be obtained.

In FIG. 29, the area ratio of the B0 portion to the entire pixel is
Whole pixel: B0 = (n + 1) ² : ((n + 1) ² − (n + 1) / 2n)
It is. Accordingly, the density value P (B0) is obtained when the density value width is 0 to 255.
P (B0) = ((n + 1) ² − (n + 1) / 2n) / ((n + 1) ² )
= (1-1 / 2n (n + 1)) × 255
It is. Similarly, in FIG. 29, the area ratio of the B1 portion to the entire pixel is
Entire pixel: B1 = (n + 1) ² : ((n + 1) ² −2 (n + 1) / n)
It is. Therefore, the density value P (B1) has a density value range of 0 to 255.
P (B1) = ((n + 1) ² −2 (n + 1) / n) / ((n + 1) ² )
= (1-2 / n (n + 1)) × 255
It is. Generally speaking, the area ratio of the Bk portion to the entire pixel is
Whole pixel: Bk = (n + 1) ² : ((n + 1) ² − (k + 1) ² (n + 1) / 2n)
It is. Accordingly, the density value P (Bk) has a density value width of 0 to 255.
P (Bk) = ((n + 1) ² − (k + 1) ² (n + 1) / 2n) / ((n + 1) ² )
= (1- (k + 1) ² / 2n (n + 1)) × 255
Can be obtained.

  By such interpolation processing, it is possible to correct even an edge in a 45-degree direction. Note that the same interpolation processing can be performed not only in the illustrated lower right direction but also in the upper left direction and the lower left direction in the case of an inverted L character. In addition to this, the preprocessing unit 52 can appropriately perform various necessary correction (interpolation) processes.

  The remaining processing in the fourth example can be performed in exactly the same way as in the second example, using a domain block image in units of 2 × 2 blocks and a range block image in units of 4 × 4 blocks. At this time, the process for reducing the reduced range block image from 4 × 4 to 2 × 2 can be created in the same manner as in FIGS. In the fourth example, the number of iterations of the iteration processing unit 61 may be set to about 5, for example. Of course, the number of iterations can be set arbitrarily.

  Thus, according to the fourth example, before the local collage processing is performed, the pre-processing unit 52 corrects the checkered portion, and at the same time, the temporary collage is focused on the low-angle line that is slowly converged by the local collage processing. By performing correction (anti-aliasing), convergence of the subsequent local collage processing can be accelerated. In this fourth example, since the block used in the local collage is as small as 2 × 2, it is possible to obtain an image in which detailed information is kept faithful and jaggies are appropriately corrected by preprocessing and local collage processing. . If the processing of the present invention is compared to engraving, the preprocessing section has a relationship such as “chisel” and the local collage processing has a “file”. A rough form is prepared by the pre-processing, and the local collage is processed in detail. Regarding the speed, although the pre-processing is slightly burdened, the local collage process itself uses a small block size and tends to converge to a sufficient image quality even if the number of iterations is small, so the overall speed is It is faster than the first to third examples.

  In the pre-processing unit 52 of the fourth example, since the upper, lower, left, and right are referred to 6 pixels, the connection length is 4 at the maximum. You may make it refer to. By doing so, a line at a lower angle can be picked up, jaggies can be corrected by pre-processing by the connection length, and the image quality can be reliably improved. When newly arranging pixel values, the density value width (for example, 0 to 255) is equally divided into k + 1 with respect to the connection length k, and k pieces are arranged. However, the number of pixels to be arranged and the pixel values are not limited to this. is not. For example, the total number of dots to be arranged may be reduced and all black dots or white dots may be arranged. In the fourth example, when the connection direction is determined, it is first determined whether or not it is in the horizontal direction, and then it is determined whether or not it is in the vertical direction, but this may be reversed. It should be noted that when referring to these peripheral pixels, the amount of scanning may be reduced as appropriate so that the scanning area does not protrude beyond the top, bottom, left, and right edges of the image.

  A fifth example will be described. In the fifth example, a case where the present invention is simply configured will be described. FIG. 30 is a block diagram illustrating a simple configuration example of the image quality improvement processing unit 5. The reference numerals in the figure are the same as those in FIG. FIG. 30 shows a configuration that is at least required for a simple configuration out of the configurations of the image quality improvement processing unit 5 shown in FIG.

  First, in the case where only one processing is required, the repetitive processing unit 61 can be omitted. In addition, as described in the explanation of each part, if the configuration is such that the pixels in which the improved domain block image overlaps are divided in advance and added to the pixel value in the memory 2 and written back, the average value calculation unit 60 is not provided. Can be configured.

  If a certain level of image quality is sufficient, there is no need to make corrections to the checkered pattern or the like in advance by the edge enhancement processing by the edge enhancement processing unit 59 or by the preprocessing unit 52. Therefore, the edge enhancement processing unit 59 and the preprocessing unit 52 can be configured without being provided.

  Furthermore, as described in the description of each unit described above, the configuration without the domain block classification unit 54 can be provided if the process control based on the classification of the domain block images is not performed. Similarly, as described in the second example above, when the range block extracting unit 55 extracts only one range block image in a fixed manner, each range block image, domain block image, It is not necessary to determine the degree of similarity. Therefore, it can be configured without providing the similarity determination unit 57. In this case, in the above description, the least square coefficient is calculated from the unique range block image and the domain block image, but the same processing is performed by the improved domain block creating unit 58 or the least square coefficient is used. If the improved domain block image is created without the similarity determination unit 37, the similarity determination unit 37 may not be provided.

  Furthermore, the process of generating the improved domain block image from the range block image in accordance with the domain block image in the improved domain block creating unit 58 is not limited to the above example, and can be generated by various other processes. . For example, referring to the domain block image and the range block image, a gradation domain block image having the same size as the domain block image can be directly generated according to a conversion formula or the like. In this case, the reduced range block creating unit 56 can be configured without being provided.

  FIG. 31 is a flowchart illustrating an example of an operation when the image quality improvement processing unit 5 has a simple configuration. In addition, the code | symbol of each step attaches | subjects the code | symbol of the step corresponding to the process shown in FIG. 2, and the overlapping description is abbreviate | omitted. First, as described with reference to FIG. 2, an area is secured in the memory unit 2 and initialized with 0.

  In S101, the initial enlargement processing unit 51 enlarges the original image to an n-fold size image to obtain an n-fold original image. In the same manner as in the above examples, it is assumed that an image enlarged twice is obtained with n = 2. In S103, the domain block extraction unit 53 extracts a domain block image from the n-fold (double) original image. In step S106, the range block extraction unit 55 extracts one range block image including the domain block image from the original image.

  In S109, the improved domain block creation unit 58 creates an improved domain block image having the same size as the domain block image from the range block image extracted in S106 according to the domain block image extracted in S103. For example, each pixel value of the improved domain block image can be obtained by calculation using a domain block image and a range block image.

  In S111, each pixel value of the improved domain block image created in S108 is divided by the number of times of addition, and addition writing is performed on the image quality improved image area in the memory unit 2.

  In S112, it is determined whether or not processing has been performed for all domain block images to be extracted by the domain block extraction unit 53. If there is an unprocessed domain block image, the process returns to S103 to repeat the processing.

  If processing is completed for all domain block images, an anti-aliased n-fold (2 times) enlarged image can be obtained. Of course, the image quality of the processing shown in the first to fourth examples is better, but the configuration of the processing can be greatly simplified and the processing speed can be improved in the fifth example. Can do.

  An intermediate configuration between the fifth example and the above-described first to fourth examples is also possible, and each part not provided in the fifth example can be provided and configured as appropriate.

  In any of the above-described examples, for example, after performing double enlargement together with the anti-aliasing processing according to the present invention, the double enlargement may be performed arbitrarily by the method of Japanese Patent Application No. 2002-079388, and finally simple binarization is performed. it can. In the case of the method shown as the fourth example, a more appropriate anti-aliased image can be obtained. Therefore, if the subsequent enlargement magnification is 2 times or less, a simpler method such as linear interpolation may be used. , Jaggies are unlikely to appear when binarized at the end. Further, it may be possible to repeat the process of enlarging the image that has been simply binarized by the method of the fourth example and further enlarging the image by the method of the fourth example. By doing so, it is possible to convert to an arbitrary magnification without causing any blur. By using such a combination, it is possible to convert the resolution of an arbitrary binary image to a high image quality at an arbitrary magnification.

  The above-described image processing apparatus and image processing method according to the present invention are as follows: “In an ideal image, a range block image similar to a target block (domain block) image exists in the very same vicinity in the same direction. The similarity (conversion parameter a in the least squares method is independent of scale) ”is based on the assumption of local self-similarity of the image. This assumption of local self-similarity is satisfied when the edge portion is visually good, and conversely, if the image is modified to satisfy local self-similarity, it will be visually good. Can do. The above-mentioned local collage method uses a block that is larger in size than the block of interest while matching the brightness using the least squares method (edge enhancement as necessary), and then puts it on the block of interest to overlap the block. The part is averaged at the end. This process (especially the reduction process) makes jaggy anti-aliased.

  In digital images, the assumption of local self-similarity is well established for visually good edge portions, and conversely, the edge portion can be improved in image quality by processing that increases local self-similarity. On the other hand, since local self-similarity tends to collapse in a thin line or the like, even if processing for increasing local self-similarity is performed, the image quality is not always improved. This problem is particularly noticeable in the thin line of a binary image. However, if the thin line is sufficiently enlarged, it can be regarded as a thick line with a step edge. That is, local self-similarity can be established if the scale at which the thin line is also seen is made smaller, and it is only necessary to have local self-similarity in a smaller region. The present invention was made from this point of view. When local collage is performed on the original image, a smaller fractional size block is used instead of the conventional integer size block, and the local self-similarity is more accurately pursued. is doing. In addition, since the binary image is an image having extremely low local self-similarity, it is made closer to an ideal image by repeating this local collage process.

  At this time, handling the fractional pixel size complicates the processing and decreases the processing speed. Here, using a fractional block when performing local collage processing on the original image is almost the same as performing local collage processing with integer size blocks on the original image enlarged by the nearest neighbor method. . It is easier to perform the actual operation with an integer size by enlarging it than when using a decimal block. In addition, by performing enlargement, an image with an anti-aliasing applied to the enlarged image is obtained, and not only performing double enlargement and anti-aliasing simultaneously, but also suppressing information loss when anti-aliasing is applied by double enlargement. be able to. In other words, since the image information is appropriately maintained and the image quality is improved, it is more convenient for further enlargement / reduction at a later stage. In particular, if the original purpose is enlargement of a binary image, it is advantageous to enlarge the original image and perform local collage processing with an integer size. Of course, when only anti-aliasing is required or when there is a memory limitation, a method of processing with a decimal size can be used. In that case, a double-enlarged memory may be held every several lines, and the pixel values obtained in the memory may be averaged to output an anti-aliased image.

  FIG. 32 is an explanatory diagram of an example in which the original image and the original image are simply enlarged by 4 times by the nearest neighbor method, and FIG. 33 is enlarged by 4 times by repeating the 2 times enlargement and antialiasing processing according to the fourth example twice. It is explanatory drawing of an example of the done image. As can be seen from the image shown in FIG. 32B, jaggies are generated in the edge portion of the image obtained by enlarging the original image shown in FIG. 32A four times by the nearest neighbor method.

  By performing the above-described processing of the present invention on the same original image, a high-quality image that has been subjected to anti-aliasing processing can be obtained, and by performing further enlargement processing, as shown in FIG. A high-quality enlarged image can be obtained.

  FIG. 34 is an explanatory diagram of an example of a computer program and a storage medium storing the computer program when the functions of the image processing apparatus or the image processing method of the present invention are implemented by the computer program. In the figure, 71 is a program, 72 is a computer, 81 is a magneto-optical disk, 82 is an optical disk, 83 is a magnetic disk, 84 is a memory, 91 is a magneto-optical disk apparatus, 92 is an optical disk apparatus, and 93 is a magnetic disk apparatus.

  The functions of the image quality improvement processing unit 5 described in the above description of the embodiment of the present invention and the first to fourth examples can also be realized by a program 71 executable by a computer. In that case, the program 71 and the data used by the program can be stored in a computer-readable storage medium. A storage medium is a signal format that causes a state of change in energy such as magnetism, light, electricity, etc. according to the description of a program to a reader provided in the hardware resources of a computer. Thus, the description content of the program can be transmitted to the reading device. For example, a magneto-optical disk 81, an optical disk 82, a magnetic disk 83, a memory 84 (including an IC card and a memory card), and the like. Of course, these storage media are not limited to portable types.

  By storing the program 71 in these storage media and mounting these storage media in, for example, the magneto-optical disk device 91, optical disk device 92, magnetic disk device 93, or memory slot (not shown) of the computer 72, the computer 71 The program 71 can be read to execute the function of the image processing apparatus or the image processing method of the present invention. Alternatively, a storage medium may be mounted on the computer 72 in advance, and the program 71 may be transferred to the computer 72 via a network, for example, and the program 71 may be stored and executed on the storage medium. Alternatively, the program stored in the storage medium in another computer may be read out and transferred to the computer 72 for execution.

  The computer 72 may be integrated with the memory unit 2 and the control unit 4 shown in FIG. 1, or may be integrated with the image input unit 1, the image output unit 3, and the like. Of course, some of the functions of the present invention may be configured by hardware, or all may be configured by hardware.

It is a block diagram which shows an example of the system containing one Embodiment of this invention. It is a flowchart which shows an example of operation | movement of the image quality improvement process part 5 in one Embodiment of this invention. It is explanatory drawing of the specific example of the original image by the end of the process of the pre-processing part 52 in one Embodiment of this invention, and a double original image. It is a schematic diagram for demonstrating an example of the process after the domain block extraction part 53 in one Embodiment of this invention. It is a flowchart which shows an example of operation | movement of a pre-processing part. It is explanatory drawing of the pattern specific example corrected by the pre-processing part. It is explanatory drawing of an example of the domain block image extracted by the domain block extraction part 53. FIG. It is a flowchart which shows an example of the classification process of the domain block image in the domain block classification part. It is explanatory drawing of an example of the range block image extracted by the range block extraction part 53. FIG. It is explanatory drawing of an example of the reduction process by the projection method in the reduction range block creation part. It is explanatory drawing of an example of the edge enhancement process in the edge enhancement process part 59. FIG. It is explanatory drawing of an example of the average value calculation process in the average value calculating part. It is explanatory drawing of another example of the range block image extraction process in the range block extraction part 55. FIG. It is explanatory drawing of another example of the reduction process by the projection method in the reduction range block creation part. It is explanatory drawing of the relationship between the domain block image in the original image in a 1st and 2nd example, and a 2 times image, and a range block image. 12 is a flowchart showing another example of domain block image classification processing in the domain block classification unit 54. It is explanatory drawing of an example of the calculation process of uneven | corrugated amount. It is explanatory drawing of the specific example of the classification result by another example of the classification process of the domain block image in the domain block classification part 54. FIG. It is explanatory drawing of an example of the division process of a domain block image according to a classification result, and extraction of a range block image. 10 is a flowchart illustrating another example of the operation of the preprocessing unit 52. It is explanatory drawing of an example of the pattern made into the process target of the pre-processing part 52 in a 4th example. It is explanatory drawing of an example of the determination method of a checkerboard, L character, and reverse L character in the pre-processing part 52. FIG. It is a flowchart which shows an example of the extraction process of the connection direction and connection length of the block image classified into the L character in the pre-processing part 52. FIG. It is explanatory drawing of the specific example of the extraction of the connection direction and connection length of the block image classified into the L character in the pre-processing part 52, and a pixel interpolation process. It is explanatory drawing of the specific example of extraction of the connection direction and connection length of the block image classified into the reverse L character in the pre-processing part 52, and pixel interpolation processing. It is explanatory drawing of the specific example in the case of performing an interpolation process by increasing the connection length in the pre-processing part. It is explanatory drawing of the difference in the case where it actually determines with the case of L shape, and a checkerboard as L shape. It is explanatory drawing of an example of the edge determination of a 45 degree direction. It is explanatory drawing of an example of the interpolation process with respect to the edge of a 45 degree direction. 3 is a block diagram illustrating a simple configuration example of an image quality improvement processing unit 5. FIG. It is a flowchart which shows an example of operation | movement in case the image quality improvement process part 5 is made into a simple structure. It is explanatory drawing of an example which expanded the original image and the original image simply 4 times by the nearest neighbor method. It is explanatory drawing of an example of the image expanded by 4 times by repeating 2 times expansion and anti-aliasing processing by the 4th example twice. FIG. 2 is an explanatory diagram of an example of a computer program and a storage medium storing the computer program when the function of the image processing apparatus or the image processing method of the present invention is realized by the computer program.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Image input part, 2 ... Memory part, 3 ... Image output part, 4 ... Control part, 5 ... Image quality improvement process part, 51 ... Initial expansion process part, 52 ... Pre-processing part, 53 ... Domain block extraction part, 54 ... domain block classification unit, 55 ... range block extraction unit, 56 ... reduced range block creation unit, 57 ... similarity determination unit, 58 ... improved domain block creation unit, 59 ... edge enhancement processing unit, 60 ... average value calculation unit, DESCRIPTION OF SYMBOLS 61 ... Repetitive processing part, 71 ... Program, 72 ... Computer, 81 ... Magneto-optical disk, 82 ... Optical disk, 83 ... Magnetic disk, 84 ... Memory, 91 ... Magneto-optical disk apparatus, 92 ... Optical disk apparatus, 93 ... Magnetic disk apparatus .

Claims (16)

In an image processing apparatus for performing image quality improvement processing of a binary original image, initial enlargement means for enlarging the original image n times by a predetermined method to obtain an n-fold original image, and a first block unit from the n-fold original image A domain block extracting means for extracting a domain block image with a size of a range block for extracting a range block image including the domain block from the n-fold original image in a second block unit larger than the first block unit Extraction means, reduced range block creation means for creating a reduced range block image by reducing the range block image extracted by the range block extraction means to the size of the first block unit, and reduced range block creation means the first from the reduced range block image in created is disposed at a position corresponding to the domain block image Possess improved domain block forming means for forming an improved domain block image of the block unit size, the improved domain block preparation means a least squares approximation pixel values of the reduced range block image to the pixel values of the domain block image The improved domain block image is created by using the least square coefficients a and b in the case of generating the improved domain block image by approximating the pixel value of the reduced range block image to the pixel value of the domain block image by a linear expression az + b. An image processing apparatus.   2. The image processing apparatus according to claim 1, wherein the initial enlargement unit enlarges the original image by a factor of 2 and performs simple multi-value processing to obtain the multi-value n-times original image.   3. The image processing apparatus according to claim 1, further comprising preprocessing means for performing dot correction on a predetermined portion of the n-fold original image obtained by the initial enlargement means using the original image. Image processing apparatus.   The preprocessing means determines the connection direction of dots according to the number of colors of surrounding pixels for a 2 × 2 checkered pattern in the original image, and in the n-fold original image obtained by the initial enlargement means, 4. The image according to claim 3, wherein dots are arranged in a checkered portion of a 2n × 2n pattern corresponding to the 2 × 2 checkered pattern so that a line is connected in a connecting direction of the dots and dot correction is performed. 5. Processing equipment.   The preprocessing means calculates the dot connection direction and connection length of the edge portion in the original image, and in the n-fold original image obtained by the initial enlargement means, the preprocessing means is at a position corresponding to the edge portion in the original image. The image processing apparatus according to claim 3, wherein the density value of a connection portion of a certain dot is corrected.   The domain block extracting means extracts the domain block image so as to have a common part on the n-fold original image, and further, the domain block extracting means corresponds to each domain block image extracted by the domain block extracting means. 6. The average value calculating means for performing division processing according to the degree of overlap with respect to pixels in which the respective improved domain block images output from the improved domain block creating means overlap. The image processing apparatus according to claim 1.   Furthermore, it has repetitive processing means for performing a repetitive process for a predetermined number of times by replacing the image output after the processing for all the domain block images with the n-fold original image and causing the domain block extracting means to perform the process again. The image processing apparatus according to claim 1, wherein:   Furthermore, it has a domain block classification means for classifying the type of the domain block image extracted by the domain block extraction means, the domain block classification means for the domain block image classified into a predetermined type Extracting one or a plurality of divided domain block images from the domain block image in a third block unit equal to or less than the first block unit, and making each of the divided domain blocks a new domain block image. The image processing apparatus according to any one of claims 1 to 7. Further, the image processing apparatus further comprises similarity determination means for determining the similarity between the reduced range block image created by the reduced range block creation means and the domain block image, and the range block extraction means adds one domain block image to the domain block image. One or more range block images are extracted, the reduced range block creating means reduces each of the one or more range block images, and the similarity determination means includes one or more reduced range ranges. A reduced range block image determined to have the highest similarity with the domain block image is selected from among the block images, and the improved domain block creation means is created by the reduced range block creation means and selected by the similarity judgment means similar the similarity determination means with respect to said reduced range block image Using a least squares coefficients a and b obtained in determining the image processing according to any one of claims 1 to 8, characterized in that creating and outputting the improved domain block image apparatus. The similarity determination means, when the pixel value z of the reduced range block image is approximated to the pixel value of the domain block image by a linear expression az + b, the reduced range block image having the smallest least square error is most similar. The image processing apparatus according to claim 9 , wherein the image processing apparatus is determined to be a reduced range block image having a high degree. The image processing apparatus further comprises edge enhancement means for performing edge enhancement processing on the improved domain block image based on a relationship between a maximum value and a minimum value of pixel values in the improved domain block image and an edge degree of the improved domain block image. the image processing apparatus according to any one of claims 1 to 1 0, characterized. The edge enhancement means takes an average value of the maximum pixel value and the minimum pixel value in the improved domain block image, and increases the pixel value larger than the average value according to the distance from the average value according to the edge enhancement intensity. the image processing apparatus according to claim 1 1, conversion, and converting smaller in accordance with pixel value smaller than the average value of the distance between the average value. The strength of edge enhancement in the edge enhancement means is set to be strong when the standard deviation value of the pixel value in the domain block image is small and weak when the standard deviation value is large. the image processing apparatus according to claim 1 2,. In the image processing method for performing image quality improvement processing of a binary original image, the original image is enlarged n times by an initial enlargement unit to obtain an n-fold original image, and a first block unit is obtained from the n-fold original image. A domain block image having a size is extracted by a domain block extracting unit, and a range block image including the domain block is extracted from the n-fold original image by a second block unit larger than the first block unit by a range block extracting unit. extracted, the range block image by reducing the size of the first block to create a reduced range block preparation means a reduced range block images, the pixel of the reduced range block image of pixel values the domain block image calculated by the least squares coefficients a and b of the similarity determination unit case of the least square approximation of the value, the least squares coefficients a With fine b, corresponding to the domain block image by converting the pixel values in improved domain block creating means by least squares fit to the pixel values of the domain block image pixel values of the reduced range block image by a linear expression az + b An image processing method comprising: generating and outputting an improved domain block image having a size of the first block unit arranged at a position to be processed. An image processing for image quality improvement processing of the original binary image by an image processing program executed by a computer, functions or claim 1 of the image processing apparatus according to any one of claims 1 to 1 3 An image processing program for causing a computer to execute the image processing method according to claim 4 . The computer readable storage medium storing a program for executing image processing on a computer to perform the image quality improvement processing of the original binary image, the image processing apparatus according to any one of claims 1 to 1 3 function or claim 1 4 computer-readable storage medium characterized by an image processing method for storing a program to be executed by the computer according to.

Images