JPH0475176A - Image processor - Google Patents

Abstract

PURPOSE:To obtain the best ideal binary-coding level by generating the histogram of an image signal, dividing among-class variance values which are obtained by density values by the frequencies of the respective density values, and calculating the binary-coding level according to the maximum value. CONSTITUTION:An among-class variance value arithmetic means 1 calculates the among-class variation values by the respective density values of the input image signal and a histogram generating means 3 detects the frequencies by the density values of the image signal. A dividing means 4 divides the among- class variance values obtained by the among-class variance value arithmetic means 1 by the frequencies of the corresponding density values obtained by the histogram generating means 3. A maximum value detecting means 2 detects the maximum value among corrected among-class variance values outputted by the dividing means 4 and sends it out as the binary-coding level for the image signal. Consequently, the best binary-coding level which nearly matches a binary-coding level obtained by a mode method can be calculated.

Description

[Detailed Description of the Invention] [Industrial Field of Application] This invention relates to an image processing device, and in particular to a method for determining a binarization level for binarizing an image signal for image recognition based on an inter-class dispersion value. The present invention relates to an image processing device suitable for making decisions.

[Conventional technology] Binarization processing has traditionally played a very important role among the image processing techniques used for image recognition, and the results of subsequent image processing are determined by the selection of the binarization level that serves as the criterion. It is so important that it is influenced by Generally, a calculation method called the Otsu discriminant analysis method is widely used as a method for determining the binarization level.

What is the Otsu discriminant analysis method? When a histogram is divided into two groups (classes) at a certain binarization level, the binary value is calculated so that the variance between these two divided groups (marginal variance) is maximized. This method determines the conversion level and divides the histogram into two.

Here, if the frequencies of groups 1゜2 divided by the temporary binarization level are μ and μ, and the average values are Ml and M2, the formula for calculating the interclass variance value σ2 is σ −= μ ・μ (M−M )/(μl ten μ2)・・・・(
1) It will look like this.

FIG. 2 is a block diagram of an image processing device constructed from the above-mentioned viewpoint, and particularly shows the image signal binarization processing function in blocks. In the figure, (
1) is an inter-paper variance calculator that calculates the variance between groups of image signals to derive an inter-class variance value, and (2) is an inter-paper variance calculator that calculates the maximum value of the inter-class variance value obtained by the inter-class variance calculator (1). This is a maximum value detector that detects and outputs the density value at that time as a binarized level.

The operation of the above configuration will now be described.

First, an image signal is input to the inter-sheet variance calculator (1), and it is assumed here that the input image signal is a 256-gradation gray image that has already been subjected to noise removal and brightness conversion. The inter-paper dispersion calculator (1) executes calculation based on the arithmetic expression (1) to obtain the inter-class dispersion value of each density value. Next, this interclass variance value is sent to the maximum value detector (2), which searches for the maximum value of the interclass variance value and outputs the density value at that time as a binarized level.

Regarding binarization processing of image signals, for example, as shown in the document "Image Recognition Theory" (Nagaashi Genshu, February 15, 1980, published by Corona Publishing, pp. 38-45), two histograms are used. Various methods are known, including the moat method, which uses the concentration value in the valley between peaks as a binary level, but - Otsu discriminant analysis method is widely used in boat fishing.

[Problem to be solved by the invention] Is the conventional image processing device configured as described above? Does the binarization level obtained by the Otsu discriminant analysis method necessarily match the binarization level obtained by the mode method? It is not always the best 2
There is a problem that needs to be solved that a value level cannot be obtained.

This invention was made in order to solve the above-mentioned problems, and the object is to obtain an image processing device that can calculate an optimal binarization level that substantially matches the binarization level obtained from the mode method. .

[Means for Solving the Problems] In order to achieve the above object, the present invention provides an interclass variance calculation means for calculating an interclass variance value for each density of an input image signal, and a histogram of the image signal. a histogram generating means for detecting the frequency number for each density, and dividing the interclass variance value which is the output of the interclass variance value calculating means by the frequency number of the corresponding density value which is the output of the histogram generating means. and a maximum value detection means for determining a binarization level for the image signal based on the maximum value of the corrected inter-class variance output from the dividing means. A processing device is provided.

C Effect] In the above means, the image processing apparatus of the present invention calculates the interclass variance value for each density of the input image signal by the interclass variance calculation means, and calculates the interclass variance value for each density of the image signal by the histogram generation means. The frequency number for each density value is detected by the dividing means, and the inter-class variance value obtained by the inter-class variance calculation means is divided by the frequency number of the corresponding density value obtained by the histogram generating means to calculate the corrected class interval. A dispersion value is determined, and the maximum value of the corrected interclass dispersion value, which is the output of the dividing means, is detected by the maximum value detection means and sent as a binarization level for the image signal.

[Example] Hereinafter, the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram of an image processing apparatus according to an embodiment of the present invention. In the figure, (3) is a histogram generator that takes a histogram of the input image signal and outputs the frequency number for each density value, and (4) is the output signal of the paper variance calculator (1) between groups of image signals. The histogram generator (3) calculates the interclass variance value for each density value obtained by calculating the variance of
The divider (5) smoothes the output of the histogram generator (3) and sends it to the maximum value detector (2). ), and (6) is the histogram generator (3
) is a switch that selects whether to give the output to the divider (4) through the A-side terminal or to the smoothing function section (5) through the B-side terminal, and (7) is the histogram generator (3). ,
This correction function section is composed of a divider (4), a smoothing function section (5), and a switch (6), and performs correction of the interclass dispersion value calculated by the edge dispersion calculation unit (1).

The operation of the above configuration will now be described.

First, the input image signal is sent to a histogram generator (3).
is input to the edge-to-edge dispersion calculator (1).

However, at this point, it is assumed that the image signal is a 256-level grayscale image that has been subjected to noise removal and brightness conversion. The histogram generator (3) generates a histogram of the density value from the image signal, and the switch (6)
It is sent to the divider (4) through the side terminal. On the other hand, the edge dispersion calculator (1) calculates the interclass dispersion value for each density value and sends it to the divider (4). A divider (4) divides the interclass variance value of each density value sent from the edge variance calculator (1) by the frequency number of the density value of the histogram sent from the histogram generator (3). . This operation can be expressed numerically as follows.

σh (i) - σ (i) / H (i) (2) where σ (i) is the corrected interclass dispersion value of concentration value i, σ2 is the interclass dispersion value of concentration value i, H( i) is the frequency of density value i in the histogram. The corrected interclass dispersion value obtained as described above is sent to the maximum value detector (2), where the maximum value is detected. Then, the density value of the interclass variance value that is the maximum value is set as the binarization level.

In addition, in the above operation, the case where the switch (6) is switched to the A side has been explained, but next, the operation when the switch (6) is switched to the B side will be explained.

The histogram generator (3) takes a histogram of the input image, and the obtained frequency number for each density value is given to the smoothing function unit (5) through the switch (6).
By smoothing the histogram representing the number of frequencies for each density as shown in the explanatory diagram of FIG. 3(b), a smoothed histogram with fine irregularities smoothed out as shown in the explanatory diagram of FIG. This smoothed histogram is given to the divider (4), where the interclass variance value of each density value sent from the edge variance calculator (1) is sent from the smoothing function unit (5). Divide by the frequency of each density value in the smoothed histogram. This operation can be expressed mathematically as follows.

σha (i)-σ2(i)/H(i)... (2
) where σ (i) is the corrected interclass variance value of density value 5, aσ2 is the interclass variance value of density value i, and H(i) is the frequency number of density value i in the smoothed histogram.

The corrected interclass dispersion value obtained as described above is sent to the maximum value detector (2), where the maximum value is detected.

Then, the density value of the interclass variance value that is the maximum value is output as a binarization level.

By smoothing the histogram in this way, it is possible to smooth out small irregularities in the frequency of each density value, make the obtained corrected interclass dispersion value more stable, and bring it closer to the ideal binarization level. .

FIGS. 4(a) and 4(b) are explanatory diagrams of processing results for two types of density distributions obtained as a result of the above processing.

In the figure, p is the histogram obtained by the histogram generator (3), q is the interclass variance value calculated by the edge variance calculator (1), and r is the value corrected by the action of the correction function unit (7). It is an interclass dispersion value, and the value of i represents the maximum value of each density value. By the way, in the drawing, the horizontal axis is O~
255 density values, and the vertical axis shows the ratio when the maximum density value is 1. In the examples shown in Figures 4(a) and (b), the ideal binarization levels are both 1-175, and therefore the maximum interclass variance values are 1-116 and 1-126, respectively. On the other hand, the maximum values of the corrected interclass dispersion value are 1-176 and 1-173, so it is clear that the corrected interclass dispersion value is closer to the ideal value.

By binarizing the image signal using the binarization level obtained as described above, it can be applied to image processing in all fields such as shape recognition, character recognition, automatic inspection, monitoring, industrial measurement, etc. can.

[Effects of the Invention] As described above, according to the present invention, by taking a histogram of an image signal and dividing the interclass variance value obtained for each density value by the frequency number for each density value, the corrected class variance is calculated. By finding the variance value and calculating the binarization level based on this maximum value, it is possible to obtain an optimal and ideal binarization level.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an image processing device according to an embodiment of the present invention, FIG. 2 is a block diagram of a conventional image processing device, and FIG. 3 is an explanatory diagram of a histogram and a smoothing histogram in the configuration of FIG. 1. FIGS. 4(a) and 4(b) are explanatory diagrams of processing results for two types of concentration distributions. In the figure, (1) is an edge variance calculator, (2) is a maximum value detector, (3) is a histogram generator, (4) is a divider, (5) is a smoothing function unit, and (6) is a switch, and (7) is a correction function section. In addition, in the figures, the same reference numerals indicate the same or equivalent parts. Agent Patent Attorney Yoshi 1) Kenji (and 2 others) Lock diagram (of the $ processing device) in Figure 2, Figure 4 3. Cover 6, scope of claims of the specification subject to amendment Column for detailed description of the invention. 7. Amendment to procedures subject to amendment (voluntary) January 1991 180

Claims (1)

[Claims] interclass variance calculation means for calculating the interclass variance for each density of the input image signal; histogram generation means for detecting the frequency number for each density of the image signal; and the interclass variance calculation means dividing means for dividing the output of the histogram by the output of the histogram generating means and sending out a corrected interclass dispersion value; and maximum value detecting means for determining the binarization level for the image signal from the maximum value of the output of the dividing means. An image processing device comprising: