JPH0120834B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a high-density pixel predictive restoration method for original images in facsimile machines, copying machines, and the like.

Conventionally, when a sampled pixel (original image) is divided into a square lattice and all of the lattice points are sampled to predict and restore a curve displayed in black and white, etc., the black and white state (arrangement) of the surrounding pixels of the pixel of interest is When the pixel of interest is white (or black), the pixel of interest is divided into four equal parts and the small white pixel (or small black pixel) formed is corrected to the opposite small black pixel (or small white pixel). Method (Unexamined Japanese Patent Publication 1973-
No. 41115) is publicly known.

This restoration method is a method of attempting high-density restoration by so-called low-density sampling. For example, even if you sample at 4 pixels/mm x 4 pixels/mm and restore at 8 pixels x 8 pixels/mm, the pixel However, it is still insufficient to reproduce curved parts smoothly enough, and the resolution is also insufficient.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a predictive restoration method capable of restoring sampled pixels sampled in a checkered pattern at low density with high density and high quality.

The present invention will be explained below with reference to illustrative embodiments.

In the present invention, a document image is read by a reader and divided into a plurality of pixel groups in a square grid pattern, and sampled every other pixel group in a checkered pattern. This is shown in FIG. Among the pixel groups in FIG. 1, S indicates a sampling pixel, and X indicates a non-sampling pixel. The addresses of these pixels S,
Represented by n. In FIG. 1, the address of the pixel of interest to be predicted and restored is (i, j)
The addresses of other pixels are as shown in the figure, centering on the reference address. That is, FIG. 1 shows a case where the non-sampled pixel X _i,j is the pixel of interest.

Now, in predicting and restoring each of the sampled pixel S and the non-sampled pixel X as the pixel of interest, the pixel of interest is processed in the following steps.

[Step a]

The density of each non-sampling pixel X is predicted and determined based on the density information of four horizontal and vertical sampling pixels S adjacent to each non-sampling pixel X.

For example, when predicting the density of a non-sampled pixel X _i,j , four samplings S _i,j-1 , S _i-1,j , S _i+1,j , S _i,j+1 are used. Predictive determination is made based on pixel density information. Note that the density information here refers to 10-value quantized density.

This density prediction is based on 4 sampling pixels.
This can be done by simply averaging the 10-value quantization density of S _{i,j -1} , S i-1,j , S _i+1,j , S _i,j+1 , but it can be done as follows. It is preferable to do so.

First, sampling pixels S _i,j-1 , S _i-1,j , S _i+1,j ,
The 10-value quantized density of S _i,j+1 is corrected using the following formula, and each corrected density S′ _i,j-1 , S′ _i-1,j , S′ _i+1,j , S′ _{i ,j+1}
seek.

S′ _i,j-1 = (2×S _i,j-1 )−4 S′ _i-1,j = (2×S _i-1,j )−4 S′ _i+1,j = (2 ×S _i+1,j )−4 S′ _i,j+1 =(2×S _i,j+1 )−4 Next, these corrected densities are averaged as shown in the formula below, and the average value is calculated for the non-sampled pixels. Let it be the concentration.

X _i,j = (S′ _i,j-1 +S′ _i-1,j +S′ _i+1,j +S′ _i,j+1 )／
4 Predicting concentration in this way is preferable for the following reasons.

That is, if the surrounding pixel densities are averaged as they are, halftones are likely to occur. This is undesirable if you want to avoid halftones as much as possible and restore a sharp image. Therefore, by performing the above density correction, the occurrence rate of halftones in the predicted density value will be lowered,
Sharpness is obtained by being able to distribute as much as possible to either the black or white side.

After interpolating and predicting the density of all non-sampling pixels X as described above, all sampling pixels S and all non-sampling pixels
Perform the following treatment.

[Step b]

The pixel of interest is divided into 9 square grids, i.e. 3 vertically and horizontally.
Each pixel is divided into nine small pixels. R ₁ in Figure 2
~ _R9 is each subpixel.

Then, _{for the central small pixel R5 located at the center among the nine small pixels R1 to R9, the peripheral small pixels R1 to R4 , R6 to R9 adjacent in} each direction around the central small pixel R5 are Among sampling pixels S and non-sampling pixels X (hereinafter referred to as adjacent pixels) adjacent to the pixel of interest, from the center point of the central small pixel R ₅ to each of its surrounding small pixels R ₁ to R ₄ , R ₆ to R Correlate the objects located in the direction toward the center point of ₉ .

For example, when X _i,j is the pixel of interest, the surrounding small pixels R ₁ to R ₄ and R ₆ to _{R 9} are associated with each other as follows.

R ₁ →X _i-1,j-1 R ₂ →S _i,j-1 R ₃ →X _i+1,j-1 R ₄ →S _i-1,j R ₆ →S _i+1,j R ₇ →X _i-1,j+1 R ₈ →S _i,j+1 R ₉ →X _i+1,j+1 [Step c] Then, when the 10-value quantization density of the pixel of interest is set as k. , the nine small pixels that make up this pixel of interest
Out of R ₁ to _{R 9} , k small pixels are output as black pixels, and the remaining (9-k) small pixels are output as white pixels. The black pixel and the white pixel are connected so that the k surrounding small pixels corresponding to those with relatively high value quantization density become black pixels and the remaining (9-k) small pixels become white pixels. Place.

Specifically, the output is determined as follows based on the density of the pixel of interest.

(1) When the 10-value quantization density k of the pixel of interest is 0.

All nine small pixels R ₁ to _{R 9} of the pixel of interest are output as white pixels.

(2) When the 10-value quantization density k of the pixel of interest is from 1 to 4.

When the 10-value quantization density k of the target pixel is from 1 to 4, among the 9 small pixels, the 10-value quantization density is first selected from among the 8 adjacent pixels. k small pixels to be output as black pixels are determined in order from the corresponding peripheral small pixels.

If this is not the case, then next to that process, for example, one of the eight adjacent pixels with a relatively high level of 10-value quantization density is determined as a black pixel, starting from the nearest surrounding small pixel. k small pixels to be output as black pixels.

The remaining (9-k) small pixels are assumed to be white pixels.

(3) When the 10-value quantization density of the pixel of interest is between 5 and 8.

When the 10-value quantization density k of the target pixel is from 5 to 8, among the 9 small pixels, the 10-value quantization density is first selected from among the 8 adjacent pixels with a relatively low level. (9-k) small pixels to be output as white pixels are determined in order from the corresponding peripheral small pixels.

If this is not the case, then next to that process, for example, one of the eight adjacent pixels with a relatively low level of 10-value quantization density is determined as a white pixel, starting from the nearest surrounding small pixel. Then, (9-k) small pixels to be output as white pixels are determined.

The remaining k small pixels are assumed to be black pixels.

(4) When the 10-value quantization density k of the pixel of interest is 9.

All nine small pixels R ₁ to _{R 9} of the pixel of interest are output as black pixels.

With this processing method, a black pixel consisting of small pixels is positioned on the side where the 10-value quantization density of the adjacent pixel is higher, and a white pixel consisting of the small pixel is positioned on the side where the 10-value quantization density of the adjacent pixel is higher. The output of the pixel of interest is determined by locating it on the lower side to obtain a sharp pixel.

Here, two specific examples will be explained.

[First Example] For example, assume that X _i,j is the pixel of interest as described above, and this is x ₁ in FIG. 3. Considering the density of this pixel X _i, j as SQ _i,j , the density of each of the eight neighboring pixels around it is T ₁ =SQ _i-1,j-1 =0...X _{i-1 ,j-1} concentration T ₂ =SQ _i,j-1 =0...S _i,j-1 concentration T ₃ =SQ _i+1,j-1 =0...X _i+1,j-1 concentration T ₄ = SQ _i-1,j = 7...Concentration of S _i-1,j T ₆ = SQ _i+1,j = 0...Concentration of S _i+1,j T ₇ = SQ _{i-1,j+ 1} = 9...X Concentration of _i-1,j+1 T ₈ = SQ _i,j+1 = 9...S _i,j+1 concentration T ₉ = SQ _i+1,j+1 = 2...X _i The concentration is _+1,j+1 . Note that T ₁ to _{T 9} are symbols established for easy comparison with R ₁ to _{R 9} .

Here, since the density of x ₁ is "4", the procedure is to first determine 4 black pixels and make the rest white pixels.In order of the density level among the adjacent pixels, it is as follows:
First the concentrations T ₇ and T ₈ , then the concentrations T ₄ and T ₉
And the rest are 0.

The corresponding small pixels are R ₇ , R ₈ , and R ₄ ,
_R9 , and these are determined as black pixels. The remaining small pixels are then determined as white pixels.
The output mode is as shown in FIG.

[Second example] Also, if the pixel of interest is x ₂ in Figure 3, and its density is set as SQ _i,j , the surrounding 8
The densities of adjacent large pixels are T ₁ = SQ _i-1,j-1 = 0 T ₂ = SQ _i,j-1 = 3 T ₃ = SQ _{i+1, j-1} = 7 T ₄ = SQ _i-1,j = 0 T ₆ = SQ _i+1,j = 9 T ₇ = SQ _{i-1, j+1} = 4 T ₈ = SQ _i,j+1 = 9 T ₉ = SQ _{i+1 ,j+1} =9.

Here, since the density of pixel x ₂ is "8", 1
After determining white pixels, the remaining pixels are determined as black pixels.

Among the adjacent pixels, here, if we say the density levels in ascending order, first, the density T ₁ and T ₄ are "0".
Therefore, one white pixel is not determined.

Therefore, a process of selecting peripheral small pixels closer to those having a relatively low level of 10-value quantization density among the eight adjacent pixels is performed. This results in
A small pixel R ₄ corresponding to an adjacent pixel with a density of T ₄
is determined as a white pixel, and the remaining small pixels are determined as black pixels. The output mode is shown in FIG.

FIG. 5 is a block diagram of an apparatus for implementing the above restoration method.

100 is an input pixel signal, 101 is a 10-value quantizer, 102 is a memory, 103 is a density prediction calculator,
104 is a determiner, 105 is a pixel level generator, 1
06 is an output buffer, 107 is an address indicator,
108 is an address signal, 109 is an output pixel signal,
110 is a timing signal.

In this device, density information of each sampling pixel represented by an input pixel signal is collected by a quantizer 10.
1, and this 10-value quantized density data is applied to the data input terminal of the memory 102.

This memory 102 takes in this 10-value quantized density data to the write address indicated by the address indicator 107 at the timing indicated by the address indicator 107.

The arithmetic unit 103 measures the timing using the timing signal 110, and when a predetermined amount of data has been accumulated in the memory 102, uses the data in the memory 102 to carry out the process of step a, that is, calculate the vertical and horizontal peripheries of each non-sampled pixel. A density prediction calculation is performed for each non-sampled pixel based on the 10-value quantized density information of the four sampling pixels adjacent to the , and is written into the memory 102. If this process is performed for all non-sampled pixels,
This calculator 103 sends a predetermined amount of concentration data to a determiner 104 in synchronization with a timing signal 110.

The determiner 104 performs processing corresponding to step b and step c in synchronization with the timing signal 110.
That is, based on the data sent from the arithmetic unit 103, each of the sampling pixel and the non-sampling pixel is set as a pixel of interest, and a black/white determination is made for that small pixel, and the determination results are sequentially output.

Pixel level generator 105 receives timing signal 11
Based on the determination results sequentially sent in synchronization with the timing signal 110, the white pixel signal and black pixel signal for each small pixel of each sampling pixel and non-sampling pixel are output in synchronization with the same timing signal 110.

The data from this determiner 104 is transferred to the buffer 10.
6 as an output pixel signal.

As described above, according to the present invention, an image sampled at low density in a checkered pattern can be restored with high density and high quality.

[Brief explanation of drawings]

Fig. 1 is an explanatory diagram showing the sampling state of a group of original pixels, Fig. 2 is an explanatory diagram showing the state in which non-sampled pixels are quantized with 10 values, and Figs. 3 and 4 are explanatory diagrams showing the restoration state of the original image. FIG. 5 is a block diagram of a prediction restoration device according to the present invention. S...sampling pixel, X...non-sampling pixel, X _i,j ... pixel of interest, _R1 to _R9 ...quantized small pixel, P...original, Pr...restored image, 100...
...Input pixel signal, 101...Quantizer, 102...
...Memory, 103...Arithmetic unit, 104...Determiner, 105...Pixel level generator, 106...Output buffer, 107...Address indicator, 108
...Address signal, 109...Output.

Claims (1)

[Claims] 1. A document image is divided into a square grid, and the divided pixel groups are sampled to form a checkered pattern of sampled pixels and non-sampled pixels,
In the method of restoring the original image to a binary image having black pixels and white pixels based on the sampling result, each of the non-sampled pixels is set as a density prediction target pixel, and four samplings adjacent to the density prediction target pixel are Predicting and determining the density of the density prediction target pixel based on the density information of the pixel, making each pixel of the sampled pixel and the non-sampling pixel a target pixel for output determination, and dividing the target pixel into 3 pixels vertically and horizontally. Each of the eight peripheral subpixels located around the central subpixel located at the center among the nine subpixels, and among the sampling pixel and the non-sampling pixel. Among the eight adjacent pixels adjacent to the pixel of interest, those located in the direction from the center point of the central small pixel to the center points of its surrounding small pixels are associated, and the 10-value quantization density of the pixel of interest is determined. is set as k, k of the nine small pixels constituting the pixel of interest are output as black pixels, and the remaining (9-k) small pixels are output as white pixels; And in the output, the k surrounding small pixels corresponding to those whose 10-value quantization density is relatively high among the eight adjacent large pixels become black pixels, and the remaining (9-k) A high-density pixel prediction and restoration method characterized by: arranging the black pixel and the white pixel so that each small pixel becomes a white pixel.