JPS6023542B2 - High-density pixel prediction restoration method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a high-density pixel predictive restoration method for predicting and restoring an image from image information sampled by dividing an image into a regular grid pattern with one pixel at a time.

Conventionally, as this type of high-density pixel predictive restoration method, the following two methods have been generally used. '11
Divide the sampled image into a square grid, sample all grid points as sample points, and use the information obtained to predict and restore curves that are displayed in black and white, for example, so that curves are displayed more smoothly. How to make corrections.

■ Divide the sampled image into a regular grid pattern, sample these grid points in a checkered pattern, and in the restoration stage unsampled pixels (pixels skipped on the grid points) are replaced with sampled adjacent pixels. This is a method of predictive restoration based on information on grid points.

However, in such conventional high-density pixel prediction and restoration methods, the method [1'] does not compress any information on the sampled image as it is, and therefore does not contribute much to reducing transmission costs.

Furthermore, in the method (2), since the size of the sampling pixel and the size of the restored output pixel correspond one-to-one, it is impossible to improve the pixel density. There were problems such as: The present invention has been made in consideration of the above points, and
The present invention provides a high-density pixel prediction and restoration method that enables highly efficient transmission and improves image quality.

In order to achieve this object, the present invention provides a method for predicting and restoring an image from image information sampled by dividing an image into a square grid with one pixel at a time. The non-sampled pixel is calculated by interpolation prediction from the density information of four sampling pixels in the horizontal direction, and the non-sampled pixel is divided into four small pixels, and each small pixel is calculated so that the average density of each small pixel approximates the predicted density value. Each pixel is assigned white or black density information using the density information of the four adjacent sampling pixels, and each of the density information is output, and each sampling pixel is divided into four small pixels. White or black density information is assigned to each small pixel using predicted density values of the four neighboring non-sampling pixels so that the average density of each small pixel approximates the density of each sampling pixel. The present invention is characterized in that each density information is output. Embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

FIG. 1 is a state diagram of sampling pixels according to the present invention, in which the shaded area is the sampling pixel S, and the non-shaded area is the non-sampled pixel x.

i, i represent the address of each pixel. Sampling is performed from the sampled image plane as shown in FIG. 1, the sampling information is transmitted, and the receiving side converts the non-sampled pixel Xi,i to its four adjacent sampling pixels Si-i,1, Si,j-. 1, Sj 11”i,
Predictive restoration is performed from Si,j+1.

At this time, as shown in FIG.
, i is equally divided into four pixels ○, , 02, 03, and 04, respectively. The algorithm will be explained below. Second
As shown in the figure, a non-sampling pixel Xi,j is connected to eight adjacent pixels A, B, C, D, Si-1,i, Si,i
Considering the case where prediction is made from −1,Si,·i+1,Sj+1,i, it is considered that the following equation (1) exists.

Xi,j=K. A+K2B10K3C+KD10. Si-
1,i+〆bui,i-1Toba 30i,i11 Toba 4Si
11,i...m Here, A,B,C,D,Si,11,j,Si,i-
1,Si,i11,Si+1,i is the density information of each pixel, K,,K2,K3,K4,Yu,,〆2,So3,
Part 4 is its prediction coefficient.

In order to investigate the relationship between the prediction coefficients K, , K2, K3, K4, So, So2, So3, and So4 in equation {1}, we conducted experiments on various images. It was found that the value shown in the following formula (■) can be approximated.

》3≦Yumedanki 2≧Low) ...・The following formula obtained by applying the formula ■ '1}' [31X. ,j=d. Si-1,j+Yu2Si,j-1Ju〆bui,iJu1Toba4Si+1,i...{3}So,Sanba2...〆3...
The non-sampled pixel Xi,i is predicted using 0.25.

Let the predicted value of the non-sampled pixel Xi,i restored by prediction be SQ(Xi,j), and among the information that SQ(Xj,i) is the blackest and the whitest, divided into five, Detect the location of the object.

Here, to simplify the explanation, it is assumed that all sampling information exists between ○ and 1, with 1 being the highest black and 0 being the highest white. Using the predicted value 6Q(Xi, i), the following equation {4'~legs predicts the four equally divided pixels ○, 02, 03, 04.

'Force SQ (Xi, j) < 1/5, then 0, = 02 = 0
3=04=○(white) ......'4} [a) 1/
5SSQ(Xi,j)<2/50k=1 (black).

〆=〇m=○h=〇 …-・-[5] [U'2/
5 Mi SQ (Xi, j) < 3/50 k = M〒1 na ... state ○ m = ○ n = o 〇) 3/5 Mi SQ (X 3, j) < 4/50 k = 0 so:
0m=1} Seni 7}.

n=. Therefore, 4/5SSQ(Xi, j) 〇k=〇so; 〇m=〇n=1...[81 In this case, the subscripts k, so, m, and n have the following relationship.

T,=Si-1,i+Si,i-I T2=Si-1,i+Si,i+I T3=Si,i-1+Si+1,i T4=Si,i+1+Sj+1,i...
{9} Arrange these T, ~T4 in descending order, and subscript 1~ of T
4, correspond to k to n.

For example, the value of T is T. =0.33 T2=0.17, T3
=0.80 When T4=0.54, the inequality relationship of the following formula (2) is established, and the values of K=3, so=4, m=1, and n=2 are established. T32T4≧T,≧T2
...■ In this way, ○,,02 in Figure 2
, 03, 04 can be determined as black or white. Next, the sampling pixel Sii+1 is divided into four equal parts〇,,0′2
, 0'3, and 04 are determined as follows.

The predicted value SQ(Xi,j) of Xi,j is calculated by the above formula [3}
and the predicted values SQ(B) of B, D, and E obtained in the same way,
SQ(D), SQ(E) and sampling pixels Si, i+
From 1, determine ○',,02,0'3,0'4.

In other words, the above-mentioned SQ(
Xi, i) is replaced by Si, j 11, and the above equation '9} is replaced by the following equation (11), the above 0,,02,03
,04, ○′,,〇2,0′3,0′
Determine 4. T, = SQ (xi, i) 10 SQ (B) L = SQ (B) + SQ (E) T3 = SQ (Xi, i) 10 SQ (D) t = SQ (D) + SQ (E) ・・...(11)
) In this way, by sequentially performing the same processing on the entire sampled image, it is possible to transmit a small amount of information sampled at a low density, and it is possible to predictably restore the image to a higher density image.

FIG. 3 shows a block diagram of an embodiment that specifically implements the high-density pixel prediction and restoration method according to the present invention. In the figure, 1 is an address counter, 2 is an address indicator, 3 is a first memory, 4 is a predictor, 5 is a second memory,
Reference numeral 6 indicates a velcolator, and reference numeral 7 indicates a logic judgment circuit.

The first memory 3 stores the sampling signal SGi from the scanner, and the second memory 5 stores non-sampled pixels predicted by the predictor 4. Further, the level controller 6 determines the level value of the predicted value or the sampled value. In addition, the logic judgment circuit 7 selects four divided pixels 0, . . .
04 and outputs an output signal SGO to an output device such as a block. Firstly, in the high-density pixel predictive restoration method according to the present invention, in the method of predictively restoring an image from image information sampled by dividing the image one image at a time into a square grid pattern,
A non-sampling pixel is interpolated and predicted from four adjacent sampling pixels in the left and right direction, and using the information of the four adjacent sampling pixels, the non-sampling pixel is divided into four, and the white and black density information of each small pixel is calculated. In addition, using the predicted values of four adjacent non-sampled pixels, the sampled pixel is divided into four, and the white and black density information of each subpixel is predicted and output. Since the information of the sampled image is compressed and transmitted, the information can be transmitted with high efficiency and the transmission cost can be reduced.

Further, the present invention has various excellent advantages such as being able to restore a clear image because the pixel density can be increased.

[Brief explanation of the drawing]

1 and 2 are state diagrams of sampling pixels according to an embodiment of the present invention, and FIG. 3 is a block diagram of an embodiment that specifically implements the high-density pixel prediction and restoration method according to the present invention. 1...address counter, 2...address indicator, 3...
First memory, 4... Predictor, 5... Second memory, 6... Level collator, 7... Logic judgment circuit. Younger brother Z diagram Younger brother Z diagram 3rd diagram

Claims (1)

[Claims] 1 In a method of predicting and restoring an image from image information sampled by dividing an image into a square grid with one pixel at a time, the predicted density value of a non-sampled pixel is calculated from the density information of four adjacent sampling pixels in the vertical, horizontal, and vertical directions. The non-sampled pixel is divided into four small pixels, and the density information of the four adjacent sampling pixels is used to calculate the white color so that the average density of each small pixel approximates the predicted density value. Alternatively, each sampling pixel is divided into four subpixels, and the average density of each subpixel is equal to the average density of each sampling pixel. White or black density information is assigned to each of the small pixels using predicted density values of the four adjacent non-sampled pixels so as to approximate the density of the pixel, and each of the density information is output. A high-density pixel prediction restoration method.