JPH02156771A - Picture reducing device - Google Patents

Abstract

PURPOSE:To improve the reproducibility of a thin line by providing a means extracting information of a picture element reduced already as required and structure information of a reference picture element block of a low pass filter, and a means correcting an output value of the low pass filter. CONSTITUTION:The device consists of a section processing an inputted original picture 1, a low pass filter 2, a subsampling section 3, a section processing an obtained reduced picture 4, a correction value arithmetic unit 6 obtaining structure information from a low pass filter reference picture element of the inputted original picture, discriminating the presence of the correction and calculating the correction value and an adder 7. That is, the structure information of a reference picture element of the low pass filter 2, the output result of the low pass filter 2 as required and the information of the reduced picture element obtained already are used to correct the output of the low pass filter 2. Thus, a thin line is reproduced as a thin line to preserve well the information thereby improving the picture quality.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to an image reduction device that obtains a smoothed image from an original size image and obtains a reduced image by subsampling the pixels of the smoothed image. .

[Conventional technology]

2. Description of the Related Art Projection methods and the like are conventionally known as image reduction techniques for obtaining a smoothed image from a full-size image and subsampling the pixels of the smoothed image to obtain a reduced image. The projection method is configured as shown in FIG.

l is a full-size image and generally a multivalued image. 2 is a low-pass filter, 3 is a subsampling section, and 4 is the obtained reduced image. The low-pass filter unit 2 performs different low-pass filtering on the input original-size image depending on the reduced size, thereby obtaining smoothed pixels. The subsampling unit 3 obtains the number of pixels necessary to form the reduced image 4 from among the smoothed pixels. In addition, when the pixels of the input image are binary, as shown in FIG. 10, a binarization section 5 is added after the subsampling section 3 in FIG. 9, and the multi-value data calculated by the low-pass filter is Binarize again, 2
Get a reduced image of the value.

For the sake of explanation, let's take as an example the case where the image size is set to horizontal A. FIG. 11 shows the reference image of the low-pass filter in the original size image surrounded by a thick frame line centered on the pixel 10f. Pixel 10f, lOa to lOc,
foe, 10g, 10i-10 pixel values as shown in the figure. −x8. Output X of the low-pass filter. is X. It is expressed as in equation (1) using −x8.

X , ) == −(x 1 +2x 2 +x 3
+2x 4 +4x ,)+2x5 +x6 +2x7
+xB) - - - (1) Each reduced image can be obtained by subsampling the output result of this low-pass filter with the result marked with a circle in Fig. 11.

[Problem that the invention is trying to solve]

However, in the above-mentioned conventional example, if the image contains many thin lines (such as a character image), the thin lines may be broken or not reproduced in the reduced image, resulting in a significant deterioration of the image quality. Taking a binary image as an example, a diagonal thin line as shown in FIG. 12(a) is not reproduced at all by using the low-pass filter of the reference image in FIG. 11, subsampling of the circle pixels, and a threshold value of 0.5. Further, a thin line as shown in FIG. 12(b) is reproduced if the pixel position to be subsampled is on the thin line, but if it is shifted, it is not reproduced at all as in FIG. 12(a).

[Means to solve the problem]

According to the present invention, by providing a means for extracting the structure information of a reference pixel block of a low-pass filter and information of already reduced pixels as necessary, and by providing a means for correcting the output value of the low-pass filter, thin lines can be reduced. This is to improve reproducibility.

[First Embodiment] FIG. 1 is a functional block diagram showing the features of the present invention.

1 is the input original size image, 2 is the low-pass filter, 3 is the subsampling section, 4 is the obtained reduced image, 6 is the structural information obtained from the low-pass filter reference pixel of the input original size image, and determines whether or not to perform correction. A correction value calculator 7 is an adder for calculating a correction value.

The following explanation assumes that the pixel value of the current-input full-size image l is represented by 8 bits from 0 to 255, and that the size to be reduced is outside.

The input original size image l is input to the global filter 2 in units of three lines, and a smoothed image is obtained by a 3×3 matrix operation according to the above-mentioned equation (1).

Further, the original size image l is input to the correction value calculator 6.

FIG. 2 shows a block diagram of the correction value calculator 6.

20a to 20c are terminals for inputting reference pixels of the global filter 2 of the original size image l for each line. 21a-2
11! is a latch, which latches the input pixel value and shifts the data to the right depending on the timing. 22.
26 is an adder that adds nine inputs, and 23 is a divider that divides the input by the number of reference pixels.

32 is a latch that latches data. 24 is a subtracter that inputs nine subtracted numbers, subtracts the separately input numbers from each, and outputs the results. 25 is a squarer which outputs the squared numbers of nine inputs. 27 is 9
This is a comparator that outputs the sign of the input with the largest absolute value among the two inputs. 29 is an input terminal, and 30 is a comparator that compares two inputs to determine their magnitude. 28 is 2
31 is an output terminal of a selector which inputs two select signals and switches between three inputs and outputs the selected signal.

When the pixel data input from the terminals 20a to 20c are stored in the latches 21a to 21i, the adder 22 calculates the sum of the pixel values of nine pixels and outputs the sum to the divider 23. The divider 23 divides the sum of pixel values by the total number of pixels to obtain an average value. At the next timing, the reference pixel block is latched 21d~
211', and the average value is latched into latch 32. The subtracter 24 calculates the error between each pixel value of the latches 21d to 211 and the average value of the latch 32, and outputs it to the squarer 25 and the comparator 27.

The comparator 27 detects the maximum error among the nine input errors and outputs the sign 1 bit of the maximum error (positive is “
0”, negative is represented by “B”. ). The squarer 25 squares each of the nine inputs to obtain a squared error, and the adder 26 squares each of the nine inputs.
Find the total multiplicative error. The comparator 30 compares the value entered manually from the terminal 29, for example 100,000°, with the total squared error, and if the total is larger, it is "BI", and if it is smaller, it is "0".
Output.

The selector 28 switches and outputs three inputs.

These three inputs are 0. +100. -100. When the output of the comparator 30 is “l”, correction is performed and the output is “0”.
No correction is made in this case. Further, when performing correction, if the sign of the comparator 27 is negative, that is, "1", negative correction is performed, and if it is positive, that is, "0", positive correction is performed. For example, when no correction is performed, the output of the selector 28 is 0, when negative correction is performed, it is -100, and when positive correction is performed, it is +100.
It is. These results are sent from the terminal 31 to the adder 7 in FIG.
is output to.

Adder 7 connects the output of low-pass filter 2 and correction value calculator 6
The sub-sampling section 3 performs addition with the correction value from the sub-sampling section 3 to obtain the value of the reduced pixel, and obtains a reduced image 4 that has been reduced vertically and outward in the horizontal direction.

As described above, in the first embodiment, the structural information in the low-pass filter reference pixel block is extracted, thin lines, etc. are detected, and correction is applied to the output of the low-pass filter, so that characters etc. that would disappear or fade with the low-pass filter alone are removed. It becomes possible to preserve thin lines and obtain a reduced image.

[Second Embodiment] FIG. 3 is a functional block diagram of a second embodiment of the present invention. 101 is an original size image, 102 is a low-pass filter, 106 is a correction value calculator that determines the presence or absence of correction and calculates a correction value, and 107 is an adder. 103 is a sub-sampling section, and 105 is a binarization section. Also, 10
9 is a buffer that stores a part of the reduced image; 1
04 is the obtained reduced image.

Now, let us assume that the pixel value of the input full-size image lot is a binary value of 0 (white) or l (black), and explain the size to be reduced as a percentage. The input original size image lot is input to the low-pass filter 102 in units of 3 lines, and the above-mentioned (1)
Find the smoothed image using the formula. , the original size image is also input to the correction value calculator 106 at the same time.

Here, the buffer 109 temporarily stores the already reduced pixels of the previous raster and the raster being processed among the already reduced reduced images. Among these reduced pixels, as shown in FIG. 4, the values of the reduced pixel obtained immediately before the reduced pixel to be obtained and the reduced pixel at the same position in the previous raster are also input to the correction value calculator 106. Ru. In FIG. 4, hatched pixels 110a to 110x are reduced pixels that have already been obtained, and 1lla is a reduced pixel that is about to be obtained.

At this time, the correction value calculator 106 includes reduced pixels 110o and 1
lox is input. FIG. 5 shows a block diagram of the correction value calculator 106. 40 is an input terminal for inputting reference pixel information of the low-pass filter, and 41 is an input terminal for inputting information about surrounding reduced pixels from the buffer 109. 42 is R
LUT (Look Up Table) composed of OM (Read Only Memory)
It is.

43 is a selector that switches and outputs one of the three inputs, and 44 is an output terminal.

The reference pixels (9 pixels) input from the input terminal 40 and the surrounding reduced pixels (2 pixels) input from the input terminal 4I are LU
It is input as the address of T42. This LUT 42 judges whether or not to make a correction based on the input 11-bit information, and when making a correction, stores in advance 2-bit information indicating whether it is a negative correction or a positive correction.

For example, when a reference pixel as shown in FIG. 6(a) is input, the reduced pixel value X is calculated according to equation (1). When xo is calculated, xo becomes 0.375, and when the threshold value T is set to 0.5, the reduced pixel becomes white. However, when considering the continuity of line segments, the reduced pixels must be black since preserving information in the reduced image is important. Similarly, when a reference pixel as shown in FIG. 6(b) is input, the reduced pixel must be black in consideration of the continuity of the line segment. However, the sixth
In the case shown in Figure (C), when the subsampled ○ mark is taken as the center pixel of the reference pixel block, the result of the Bobo projection method for the reference pixel block on the left is black, and the result for the reference pixel block on the right is black. It's black for a reason.

In this case, a line segment with a width of 2 pixels in the original size image is 2 pixels wide even in the reduced image.
Since a problem related to the pixel width will occur, no correction is made to the projection method for the reference pixel block on the right side. That is, even if the structures of the reference pixel blocks are exactly the same, some may or may not be corrected depending on the circumstances of the surrounding pixels. In the reference pixel block as shown in FIG. 6(b), if the previous reduced pixel is white, positive correction is performed, and if it is black, no correction is performed. Use this information as an LUT
The meanings expressed by the two bits output from the LUT 42 are as shown in Table 1.

Table 1 According to the output of this LUT 42, the selector 43 switches between three inputs and outputs. Input is 0. +0.5°-0,5625. If the input is without correction, 0 is output, if positive correction is to be performed, +0.5 is selected, and if negative correction is to be performed, -0,5625 is selected and output to terminal 44.
The signal is then output to the adder 107 in FIG. Adder 107
adds the output of the low-pass filter and the correction value to obtain the value of the reduced pixel, and the sub-sampling unit 103 reduces the number of pixels vertically and horizontally to %, which is input to the binarizing unit 105, and the threshold value T =0.5 to obtain a reduced binary image 104. This result is also output to the buffer 109 and fed back to the correction value calculator 106.

As described above, in the second embodiment, by using the structure information in the low-pass filter reference pixel block and the information on the already obtained reduced pixels, thin lines that would disappear with the low-pass filter alone can be preserved, and the structure in the reference pixel block can be preserved. It is possible to prevent thin lines from becoming thicker, which would otherwise occur when making decisions based only on information.

[Third Embodiment] FIG. 7 is a functional block diagram of a third embodiment of the present invention.

Reference numeral 201 indicates an input original size image, 202 indicates a low-pass filter, 208 indicates a latch for timing adjustment, 206 indicates a correction value calculator that determines the presence or absence of correction and calculates a correction value, and 207 indicates an adder. 203 is a sub-sampling section, and 205 is a binarization section. 204 is the obtained reduced image, and 209 is a buffer that stores a part of it.

Now, let us assume that the pixel value of the input original size image 201 is binary, 0 (white) or l (black), and assume that the size to be reduced is l. The input original size image 201 is input to the low-pass filter 202 in units of 3 lines, and the above-mentioned (1)
The image value of the smoothed image is determined by the formula and latched in the latch 208. At the same time, the output result of the low-pass filter, the reference pixel block, and the buffer 2 are sent to the correction value calculator 206.
From the reduced pixel already reduced from 09, surrounding reduced pixels 110o and 110x as shown in FIG. 4 are input.

FIG. 8 shows a block diagram of the correction value calculator 206.

240 is a terminal for inputting reference pixel information of the low-pass filter, and 241 is an input terminal for inputting information about surrounding reduced pixels from the buffer 209. 242 is an LUT composed of ROM. 243 is an input terminal for inputting the result of the low-pass filter, 244 is a binarizer that binarizes with a threshold value T=0.5, 245 is an exclusive OR circuit, 246, 247
is a two-manpower, one-output selector, and 248 is an output terminal.

The reference pixels (9 pixels) input from the input terminal 240 and the surrounding reduced pixels (2 pixels) input from the input terminal 241 are LU
It is input as the address of T242. ~This LUT2
42 contains the input 11-bit information, or if correction is not necessary in each case, the result of binarization without correction, or if correction is necessary, the result of binarization after correction. Predetermined results are stored to be derived. Taking FIG. 6 as an example, in the case of FIG. 6(a), 1, that is, black is output regardless of the surrounding reduced pixels. Figure 6 (
In case b), when the result of the reduced pixel on the left side is 1 (black), O (white) is output, and when the result is 0 (white), l (black) is output.

The output of the low-pass filter is inputted from an input terminal 243 and is binarized by a binarizer 244 . Exclusive OR circuit 24
5, it is checked when the outputs of the binarizer 244 and the LUT 242 are different, that is, whether or not correction is necessary. Exclusive OR 245
No correction is performed when the output is 0, and correction is performed when the output is l. The selector 246 is a two-man power, one-output selector. LUT
If the result of 242 is 0, it is a negative correction value, i.e. -0,31
25, and in the case of l, a positive correction value, that is, +0.25 is output. These correction values are determined as follows.

The block represented by the reference pixel block in FIG. 6(b) has the smallest low-pass filter output ((0, 25), threshold T
If the difference value with = 0.5 is the largest, the positive correction value will be 0.25. On the other hand, the negative image in FIG. 6(b) requires negative correction in the same way, and the low-pass filter output is the thickest ((0,75), If the difference value from the threshold value T=0.5 is the largest, the negative correction value will be -0,3125 because the value after correction is less than 0.5.

The selector 247 selects the correction value determined in this way,
The correction value output 0 in the case of no correction is switched by the output of the exclusive OR circuit 245, and is output from the terminal 248 to the adder 207 in FIG.

The adder 207 adds the output of the low-pass filter and the correction value to obtain the value of the reduced pixel, and the subsampling unit 203 thins out the number of pixels in the vertical and horizontal directions.
The image is input to the binarization unit 205 and compared with a threshold value T=0.5 to obtain a reduced binary image 204. This result is also output to the buffer 209, and is used as information for determining whether subsequent pixel correction is necessary.

As described above, in the third embodiment, the structure information in the low-pass filter reference pixel block and the already obtained reduced pixel information are used, and the output of the low-pass filter is used to preserve thin lines that would disappear with the low-pass filter alone. This prevents the thickening of thin lines that occurs when determining only the structural information in the reference pixel block, and furthermore, the weighting coefficient of the low-pass filter,
Image quality and density can be controlled by changing the binarization threshold.

[Other Examples]

The subsampling method in each embodiment is not limited to this, and it is also possible to select the center pixel position of the low-pass filter and perform it simultaneously with the low-pass filter. Each correction value,
The threshold value for binarization is not limited to this. The shape and weighting coefficient of the low-pass filter are not limited to these. The structural information extraction method of the first embodiment is not limited to this, and it is also possible to use a Laplacian filter, orthogonal transformation, or the like.

The nature of the original size image and the reduction ratio are not limited to these. Also,
The positions of the surrounding reduced pixels to be referenced are not limited to this.

〔Effect of the invention〕

As explained above, the structure information of the reference pixel of the low-pass filter and the output result of the low-pass filter as necessary,
By correcting the output value of the low-pass filter using the already obtained reduced pixel information, thin lines in the reduced image are prevented from fading or disappearing, and by reproducing thin lines as they are, information is better preserved and image quality is improved. It can be performed.

Furthermore, by changing the binarization threshold value, it is possible to adjust the density while maintaining the image quality. Furthermore, even if reduction is repeated several times to obtain even smaller reduced images, information can be better preserved.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the functions of the first embodiment of the present invention, FIG. 2 is a block diagram of the correction value calculator of the first embodiment, and FIG.
Figure 4 is a block diagram showing the functions of the second embodiment of the present invention, Figure 4 is a diagram showing the positions of surrounding reduced pixels, Figure 5 is a block diagram of the correction value calculator of the second embodiment, Fig. 6 is a diagram showing an example of a reference pixel block to be corrected, Fig. 7 is a block diagram showing the functions of the third embodiment implementing the present invention, and Fig. 8 is a correction value calculation of the third embodiment. Block diagram of the device, No. 9
10 is a block diagram showing the functions of the conventional example, FIG. 11 is a diagram showing the matrix of the low-pass filter when the reduction rate is %, and FIG. 12 is a diagram showing the problems of the conventional example. 1, 101.201 is the original size image, 2,102,202
3, 103, and 203 are sub-sampling units; 4. is a low-pass filter; 104.204 is a reduced image, 5. 1
05°205 is a binarization unit, 6. 106.206 is a correction value calculation unit, 7.107, 207 is an adder, 109.20
9 is a buffer.

Claims (5)

[Claims] (1) In an image reduction device equipped with a low-pass filter for obtaining a smoothed image from a full-size image and a subsampling means for subsampling pixels of the smoothed image, structural information in a pixel block consisting of reference pixels of the low-pass filter An image reduction device comprising: structural information extracting means for extracting structural information; and means for correcting an output value of a low-pass filter using the structural information. (2) In an image reduction device equipped with a low-pass filter for obtaining a smoothed image from a full-size image and a subsampling means for subsampling pixels of the smoothed image, a structure in a pixel block consisting of reference pixels of the low-pass filter. A means for obtaining information, a means for obtaining information on reduced pixels that have already been obtained around the reduced pixel to be obtained by the subsampling means, and a means for correcting the output value of the low-pass filter using this information. An image reduction device characterized by: (3) In an image reduction device equipped with a low-pass filter for obtaining a smoothed image from a full-size image and a subsampling means for subsampling the pixels of the smoothed image, a pixel block consisting of reference pixels of the low-pass filter is means for obtaining structural information; means for obtaining information on reduced pixels that have already been obtained around the reduced pixel to be obtained by the subsampling means; and an output value of the low-pass filter based on these information and the output of the low-pass filter. An image reduction device comprising: means for correcting. (4) When both the original size image and the reduced image are binary images,
4. The image reduction device according to claim 1, wherein the correction value is a value equal to or larger than the threshold value. (5) When both the original size image and the reduced image are binary images,
4. The image reduction device according to claim 1, wherein the correction value is a maximum value of the difference between the output value of the low-pass filter of the reference pixel block to be corrected and the threshold value.