JPS59133774A - Adaptive quantizing circuit - Google Patents

Abstract

PURPOSE:To perform effective quantization in real time and at the same time to reduce the scale of hardware by detecting the feature quantity of the luminance level distribution in a luminance level histogram and then detecting a range where an input picture signal is changed to a low density level from a high density level. CONSTITUTION:The picture signal sent from a sensor 21 is digitized by a quantizing circuit 22, and the number of picture elements is counted every luminance level within a block divided in the main scanning direction by a luminance level histogram circuit 23. Then the count value of picture elements is supplied to a deciding circuit 24 for segmenting position. The circuit 24 detects the position where the picture signal is changed to a low density level from a high density level on the basis of the supplied count value and with reference to the segmenting position of a preceding line or a preceding block. Thus the segmenting position is decided by the circuit 24. A quantizing circuit 24 receives the designation for the segmenting position and converts the picture signal into a low level.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention extracts the luminance level in the range from dark to light, which is considered to contain important information when converting a 9-tone image into binarization or rough multi-value conversion. Regarding adaptive quantization circuits aimed at . When reproducing a binary image from an image signal containing halftones, the most important information is contained in the halftones.

Changes within the perfect white or black level (may not contribute much to binarization or may even become noise).Determine the range of quantization from the brightness level histogram, taking into account the above characteristics. However, these methods have the ability to divide the screen into small areas and use a brightness level histogram for each area.
can.

In the method using the brightness level histogram of the entire screen,
Localized abnormalities on both sides! ! ! ! This method has disadvantages in that it cannot be tracked and cannot be processed in real time.

Furthermore, although the method using a brightness level histogram within a small area can eliminate the above-mentioned drawbacks, it has the disadvantage that the hardware becomes enormous and the connections between each small area become unnatural.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an adaptive quantization circuit comprising relatively small-scale hardware capable of eliminating such conventional drawbacks and performing effective quantization in real time.

According to the present invention, there is provided a means for creating a brightness level histogram of an input image signal, a means for generating a brightness level histogram of an input image signal; The present invention provides an adaptive quantization circuit which is specially equipped with means for detecting the range where the brightness changes from dark to light, and means for extracting the luminance level of the previously detected range.

According to the adaptive quantization circuit having the configuration described above, image signals successively corrected for each part of the screen can be obtained in real time using small-scale hardware.

The present invention will be described below with reference to the drawings showing one embodiment.

FIG. 1 is a brightness level histogram obtained from a general binary image. The horizontal axis shows the brightness level, and the vertical axis shows the frequency (
Frequency). In this embodiment, the analog image signal output from the sensor is converted into a 64-value digital image signal by the first quantization circuit. - Next, the adaptive quantization circuit of the present invention extracts the 16 values of the part where the artist's name indicated by diagonal lines in FIG. 1 changes from deep to light.

FIG. 2 is a block diagram of the adaptive quantization circuit of the present invention.

The analog image signal obtained from the sensor 21 is converted into a 6-bit digital image signal by a first quantization circuit 22f consisting of a video amplifier and an analog/digital converter. A brightness level histogram circuit 23 counts the number of pixels for each brightness level within a block divided in the main scanning direction. When counting for one block is completed, counting for the next block is started, and the count value for the previous block is read out according to the read address generated by the cutout position determination circuit 24, and the count value is sent to the cutout position determination circuit 2. 4f
Send this. The cropping position determination circuit 24 has a microprocessor, and based on the count value of the luminance level histogram, also refers to the cropping position of the previous line or block, etc., and determines the position where the image signal changes from dark to light (!: The second quantization circuit 25 receives the designation of the cutting position from the cutting position determining circuit 24, and determines the cutting position.
In this embodiment, the first quantizer converts the 64-value quantized image signal into 16-value data. FIG. 3 is a block diagram showing a brightness level histogram circuit. One blog consists of n pixels in the main scan. Terminal 31ζ
This 64-value digital image signal is input, and a clock synchronized with this image signal is input from the terminal 38. Control circuit 39
Then, the input selection of multiplexers 33a and 33b is switched every n pixels. 33a (when the image signal 31 is selected here, the address corresponding to the brightness level I of the memory 35a is accessed, and after reading, the counter 36a
+Thus, 1 count up is performed, and multiplier 3.4
The data is written to the same address in the memory 35a again via the address a.

Here, the multiplier 34a multiplies l, so the counter 36a
The output is written directly into the memory 35af. After creating a luminance level histogram for n pixels in this way, the control circuit 39 controls the second part of the memory 35a.
Terminal 32 receiving the output of the cutting position determining circuit 24 in the figure
is selected by the multiplexer 33a. and memory 35
The content of a, that is, the count value of the brightness level histogram created in the previous block, is read out and output to the terminal 37. Also,
At the time of reading, the multiplier 34a multiplies the result by a coefficient α (<1) and writes the result to the memory 35ai. Therefore, the count value of the brightness level histogram that is read out has a weight of 1. The previous line is α, the line before the previous one is α2, etc., and the value multiplied by the count value ζ of m lines before becomes the integrated value. While reading the memo 35a, the control circuit 39 controls the multiplex +! j33
In b, the image signal input terminal 31 is selected, and the memo 1J
35b, counting is performed in the block next to the block previously counted in Domemo 1L 35a. When the counting in this block is completed, the control circuit 39i reads out the data from the memory 35b and multiplies it by the coefficient α, then writes the data into the memory 35b as before. During this time, the multiplexer 33a selects the image signal input terminal 31 and creates a brightness level histogram in the memory 35af; Create a level histogram.

As mentioned above, 'f4' of multiple programs is stored in one memory.
An m-level histogram is created, and the brightness level histogram created in the previous block is read out from the other memory.

FIG. 4 is a diagram of probe 4 of the cutting position determination circuit 24.

The address for reading the brightness level histogram is connected to the terminal 3 via the brightness level histogram circuit interface 41.
21 outputs, and the counted value is taken into the main memory 43 from the terminal 37. The microprocessor 42 reads out the counted values of the luminance level histogram stored in the main memory 431, determines the range in which the image signal changes from false to pale, and outputs the counted value to the terminal via the interface 44 with the second quantization circuit. 454ζ output. There are various ways in which the microprocessor 42 determines the cutting position, and examples will be given below. The maximum value, minimum value, and mode of the brightness level are calculated as the feature values of the brightness level histogram, and the most frequent value is the brightness level representing the background part of the input image, and the minimum value is the brightness level representing the object such as text. One possible method is to set a reference value f between the two and cut out 16-value image signals around this reference value. In Figure 1, Y is the mode, 5frS
Major and minor values, X is the maximum value. Here, (Y+8)/2 is set as the reference value.

However, if there is only the background, such as between lines of text, S
cannot be used as is. Therefore, looking at the relationship between the level Z and S obtained by folding X to Y, if S is larger than Z-m, it is determined that the block is between lines, and the reference value in the previous block or line is set. use

Here m is a positive constant. However, if you update the reference value for each block, the values may jump too much between blocks and become discontinuous, causing unnaturalness such as sudden changes in line thickness. The image signal cutting reference value is changed by at most 1 between blocks that match. Even if the somewhat complicated processing described above is performed, existing microprocessors can be used because the processing is performed within one block.

Next, we will discuss an example of somewhat complicated processing. Here, it is assumed that the input image is a binary image, that the created histogram consists of two peaks, and that the peaks due to the white level of the background are symmetrical with respect to the mode. Although this assumption was made in the previous example, it is actively used in the example shown here. From the histogram read out as shown in Figure 6 fat, subtract the minimum value of a count value that is higher than the mode value from the count value at a symmetrical position with the mode as the center, and then All counts at the upper level are zero. If the peak due to the background white level is symmetrical, the count value due to the background white level f will not be reflected in the histogram f by the above operation.In other words, the remaining f on the histogram is due to the black characters etc. Only. In other words, a histogram as shown in Figure 6 (bl) is obtained.Then, look at the shape of the histogram that remains next.If there is a mountain with a count value above a certain level, its central value or mode will be expressed as a letter. Assuming that the level indicates black,
The median value between the mode value of the white level detected earlier is cut out and set as the reference value. Further, if the peak does not form a clear mountain shape, it is assumed that black does not exist or the number of samples is too small to be reliable, and the black level determined in the previous block or line is used. Subsequent updates of reference values between blocks are performed in the same manner as in the example shown above.

In general documents, there are many line drawings, and the peaks caused by the black level are spread out, making it difficult to find the center of the peaks when viewed at the same time as the white level peaks. However, according to the above method, the noise caused by white is reduced and the value that should be representative of the black level can be relatively accurately determined. Although the above processing can be performed by hardware, it would be very complicated, so it would be wise to use a microprocessor as shown in this embodiment.

In addition, if the extraction reference value is stable, the same reference value may be used from now on, and the feature 1lfI
As long as S does not change, updating of the reference value will be stopped and unstable lζ
It is also possible to avoid this.

The base value V determined as described above is inputted to the terminal 451 of the second quantization circuit shown in FIG.

An image signal X quantized to 64 values by the first quantization circuit lζ is input to the terminal 311ζ. When the output y to the terminal 26 in the arithmetic circuit 51 is Y=Oi-(x-v)+a<0=(x-v)+
When a; O<Cx v)+a<15" 15
Calculation is performed so that i (x-v)+a>15, and 16-value conversion is performed.

Here, a is a constant and is a parameter that determines the positional relationship on the luminance level axis between the reference value in Cζ in Figure 1 and the portion to be cut out. Compared to uniform quantization, quantization can describe parts that change from dark to light with a smaller number of bits, and can better track level fluctuations in the overall input image signal and compensate for differences in contrast. An image signal can be obtained. In addition, it has the same effect as the commonly used method of calculating histograms in small areas. This can be done without using additional hardware (line memory, etc.).

[Brief explanation of the drawing]

Figure 1 shows the portion of the brightness level histogram that should be extracted from dark to pale, the base value for extraction, and the level referenced for determining the reference value.
The figure is a block diagram showing the entire adaptive skeletonization circuit of the present invention.
The third block diagram is a block diagram of the brightness level histogram circuit, the 454th block diagram is a cutout prediction circuit, 23...Brightness level histogram circuit 24...The cutout position determination circuit , 25... Second quantization circuit s 33a + 33b + 33c... Song multiplexer 34a, 34b... Multiplier, 35g+35b... ...Memory, 36a, 36
b・・・・・・Counter%41・・・・−・・・
・Brightness level histogram circuit interface, 42
......Microprocessor, 43...
. . . Main memory, 44 Surroundings/Interface with second quantization circuit, 51 . . . Second quantization circuit. Representative Patent Attorney I's Shinnen 1 Zufu Shiheru

Claims (1)

[Claims] In a circuit for adaptively quantizing a grayscale image signal, means for creating a brightness level histogram of an input image signal;
The brightness level histogram is read line by line or block by block, the feature amount of the brightness level distribution in the read brightness level histogram is detected, and the input image signal is determined from the feature amount.