JPH05115005A - Method and device for deciding threshold level - Google Patents

Abstract

PURPOSE:To reduce the processing time by decreasing number of times of scanning with respect to the decision of a threshold level for binarizing a gradation picture and to reduce the circuit scale. CONSTITUTION:The device is provided with a local data extract section 1 extracting a density G0 of a noted point and densities G1,...,G4 of four points selected not including two points pairs symmetrical to the noted points among the eight neighbouring points, and with five minimum value selection sections 3 each selects and outputs a value Fn (n:0,...,4) which is smaller between the density G0 of the noted point and each of the five densities G0,...,G4, and also with five histogram count section 4 counting simultaneously the incidence occurrence of the output Fn of each selection section 3, and with a threshold level decision section 5 obtaining a total sum A(T) of adjacent numbers and an area S(T) based on a histogram obtained simultaneously each Fn from each histogram count section 4, calculating an average adjacent number R(T) and deciding an optimum threshold level based on a change state of the average adjacent number R(T) with respect to the threshold level Y.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used for binarizing a grayscale image in a binarization process which is a typical process for separating an object in an image from a background in a visual sensor or the like. And an apparatus for determining the optimum threshold value of

[0002]

2. Description of the Related Art Conventionally, with respect to this kind of threshold value determination, a method based on a density histogram representing a distribution of the appearance frequency of the density of each point of an image has been generally used. For example, a mode method for determining the threshold value at the position of the valley portion of the density histogram, or when the density histogram is divided into two classes, the threshold value is determined so that the sum of the variances of the density in each class is minimized. There is a discriminant analysis method.

In addition to the method based on such a density histogram, an evaluation function representing the quality of the binarization result is obtained based on the feature amount measured from each binary image when binarizing while changing the threshold value. After calculating, there is a method of determining the optimum threshold value by judging from the change state of the evaluation function with respect to the threshold value. This method has a problem that the amount of calculation increases in proportion to the number of threshold values to be changed. The present inventor has already proposed in Japanese Patent Application No. 2-202802 a "threshold value determining device" for solving the above problem.

Since the present invention relates to an improvement of this, the above-mentioned prior invention will be described first. The optimum threshold value is considered to be a value that is extracted as a region where "1" connected components are "collected (the boundaries are not jagged)" on the binary image. Now, regarding the “1” point of the binary image, if the number of “1” points in the 8 neighborhoods is defined as the number of adjacencies, the number of adjacencies is 8 at the internal points of the region, as shown in FIG. It takes a value of 0 to 7 at the boundary point. Therefore, if the total number of adjacencies is calculated for the entire binary image and this value is divided by the total number (area) of "1" points to define the average number of adjacencies, the better the area cohesion, the more the area occupies the area. Since the ratio of the internal points is high and the ratio of the boundary points is low, the average number of neighbors shows a high value.

Here, the sum of the numbers of adjacencies in the binary image with respect to the threshold value T is A (T), and the area of the point of "1" is S (T).
When expressed by, the average number of neighbors R (T) is

[0006]

[Equation 1]

It is represented by Furthermore, the number of neighbors is i (i =
If the number of 3 × 3 patterns of 0, ..., 8) is represented by ai, the sum A (T) and the area S (T) of the number of adjacencies can be expressed as follows.

[0008]

[Equation 2]

[0009]

[Equation 3]

In general, when the object and the background in the image have the density distribution as shown in FIG. 9, the change of the average number of neighbors R (T) with respect to the threshold value T is as shown in FIG. In this case, if the value that maximizes the average number of neighbors is used as the threshold value, the object is extracted as a grouped area, and a good binarization result is obtained.

FIG. 5 shows an embodiment of the invention of the prior application. In the figure, reference numeral 1 is a local data extraction unit for extracting the density G0 of the point of interest and the densities G1, ..., G8 of the points in the vicinity of the eight from the grayscale image, and 2 is output from the local data extraction unit 1 9
A selection signal n (n = n) from among the individual densities G0, ..., G8
0, ..., 8), one of which is selected by the local data selection unit 3, and 3 is an output Gn (n = 0,
, 8) and the density G0 of the point of interest output from the local data extraction unit 1 are compared, and the smaller value Fn (n = 0,
, 8) for outputting a minimum value selection unit, 4 is a histogram counting unit for counting the appearance frequency of the output Fn of the minimum value selection unit 3, and 5 is obtained by the histogram counting unit 4.
2 from the 9 types of histograms for n
After calculating the sum A (T) and the area S (T) of the number of neighbors in the binarized image and calculating the average number of neighbors R (T),
It is a threshold value determining unit that determines an optimum threshold value based on the change state of the average number of neighbors R (T) with respect to the threshold value T.

Next, the operation will be described. First, the local data extraction unit 1 uses nine latches 1 as shown in FIG. 6, for example.
It is composed of one and two shift registers 12, and extracts the density G0 of the point of interest and the densities G1 to G8 of the points in the vicinity of eight from the grayscale image to be binarized. Next, the local data selection unit 2 selects a selection signal n (n = 0, ..., From the nine density data G0, ..., G8 output from the local data extraction unit 1.
Select one data Gn according to 8). Next, the minimum value selection unit 3 compares the output Gn of the local data selection unit 2 with the density G0 of the point of interest sent from the local data extraction unit 1 and outputs the smaller value Fn. Now Fn to G0
, ..., G8,

[0013]

[Equation 4]

[0014] By changing the selection signal n from 0 to 8 and scanning the grayscale image 9 times, Fn is output from the minimum value selection unit 3, and the histogram counting unit 4 counts the frequency of appearance of the nine types of histograms. Is obtained.

In the histogram of the result Fn obtained by performing the minimum value selection process, the number of pixels having the density value k is hn (k).
(K = 0, ..., N-1; n = 0, ..., 8). Here, N is the number of density levels representing a grayscale image. Therefore,

[0016]

[Equation 5]

When the number of pixels is expressed by Hn (T),

[0018]

[Equation 6]

It becomes Here, the output Fn (n = 1, ..., 8) of the minimum value selection process is the density G0 of the point of interest and the density Gn (n = 1, ..., 8) of one of the eight neighboring points.
, Which is the smaller value of Hn (T) (n = 1, ...,
In 8), when the grayscale image is binarized with the threshold value T, the target point is "1", the points corresponding to Gn in the 8 neighborhoods are "1", and the other seven points are "0". 3 or 3 that becomes "or" 1 "
Represents the number of patterns. Hn (T) (n = 1, ...,
Each of 8) includes the number of 3 × 3 patterns in which the number of adjacencies is 1 to 8 in the binary image with respect to the threshold value T, and the sum of all of these results in the number of adjacencies i (i =
1, ..., 8) The total number of 3 × 3 pattern numbers ai added i times. That is, the sum A (T) of the number of adjacencies in the binary image with respect to the threshold T is

[0020]

[Equation 7]

[0021] Further, the area S (T) in the binary image with respect to the threshold value T is the number of pixels itself equal to or greater than the threshold value T in the histogram of F0 = G0.

[0022]

[Equation 8]

[0023] From the above, the sum A (T) of the numbers of adjacencies and the area S (T) are calculated as the result Fn (n = 0,
,, 8), which is obtained from nine types of histograms,
The average number of neighbors R (T) can be calculated based on these.

By the way, in the above example, Fn (n = 0,
, 8), the grayscale image is scanned and counted 9 times while changing the selection signal n from 0 to 8, but the minimum value selecting unit 3 and the histogram counting unit 4 are used as shown in FIG. 9 are arranged in parallel, and (G0, G0), and
By giving (G1, G0), ..., (G8, G0), nine kinds of histograms for Fn (n = 0, ..., 8) can be obtained by scanning the grayscale image only once. You can also In this case, the local data selection unit 2 becomes unnecessary. Needless to say, the minimum value selection unit 3 having two inputs (G0, G0) can be omitted by directly inputting G0 to the histogram counting unit 4.

[0025]

Since the conventional threshold value determining apparatus is constructed as described above, the histogram of the output Fn (n = 0, ..., 8) of the minimum value selecting section is converted into the selection signal n. It is necessary to scan and count the grayscale image 9 times while changing from 0 to 8, which requires processing time. Also, Fn
The nine types of histograms for (n = 0, ..., 8) are
Although it is possible to obtain the grayscale image by scanning the grayscale image only once, it is necessary to arrange nine minimum value selection units and nine histogram counting units in parallel, which causes a problem such as an increase in circuit scale. It was

The present invention has been made in order to solve the above problems, and provides a threshold value determining method and apparatus capable of reducing the number of scans to shorten the processing time and the circuit scale. The purpose is to get.

[0027]

According to the threshold value determining method of the present invention, the density G0 of the point of interest from the grayscale image and its
Of the neighboring points, the densities G1, ..., G4 of four points selected so as not to include two points symmetrical with respect to the target point are extracted, and the densities G0, ..., G4 of the five points extracted are extracted. Gray image 1
Select one for each scan and select the selected density G
Of n (n = 0, ..., 4) and the density G0 of the point of interest, the smaller value Fn is selected, and the frequency of appearance of each Fn obtained by scanning the grayscale image five times is counted to determine five types. Generate a histogram, calculate the sum A (T) and the area S (T) of the number of neighbors based on the generated histogram, and calculate the average number of neighbors R (T).
After calculating, the optimum threshold value is determined based on the change state of the average number of neighbors R (T) with respect to the threshold value T.

Further, the threshold value determining method according to another invention of the present application does not include the density G0 of the target point from the grayscale image and two points of the eight neighboring points which are symmetrical with respect to the target point. , G4 at the four points selected in step S4 are extracted, and the smaller value Fn (n = 0, ...) Of the respective concentrations G0 ,. ，
4) is selected, the appearance frequency of each Fn obtained by scanning the grayscale image once is counted, and five types of histograms are generated. Based on the generated histograms, the sum A (T) and the area of the adjacent numbers are calculated. After calculating S (T) and calculating the average number of neighbors R (T), the optimum threshold value is determined based on the change state of the average number of neighbors R (T) with respect to the threshold value T.

On the other hand, the threshold value determining apparatus according to the present invention is selected so as not to include the density G0 of the point of interest from the grayscale image and two points of the eight neighboring points which are symmetrical with respect to the point of interest. Depending on the selection signal n (n = 0, ..., 4) from the extraction unit for extracting the densities G1, ..., G4 at the four points and the densities GO, ..., G4 at the five points output from the extraction unit. A first selecting section for selecting and outputting one, and a second selecting section for selecting and outputting a smaller value Fn of the output Gn of the first selecting section and the density G0 of the target point, A histogram counter that counts the frequency of appearance of each output Fn of the second selector, and a sum A (T) and an area S (T) of the number of neighbors based on the histogram obtained for each Fn from the histogram counter. After calculating the average number of neighbors R (T), the average number of neighbors R (T) with respect to the threshold T is changed to It is obtained by a threshold value determination unit that determines an optimal threshold Hazuki.

Also, a threshold value determining device according to another invention of the present application is the density G0 of five points which is the output of the extracting unit ,.
A selection unit that selects and outputs the smaller value Fn (n = 0, ..., 4) of each of G4 and the density G0 of the point of interest, and the appearance frequency of each output Fn of the selection unit is simultaneously counted. The present invention is provided with five histogram counting units which are provided to the threshold value determining unit.

[0031]

In the threshold value determining method of the present invention, the density G0 of the point of interest from the grayscale image and the four points selected so as not to include two points symmetric with respect to the point of interest among the eight neighboring points. By extracting the densities G1, ..., G4, the optimum threshold value can be determined from the five types of histograms, so that the grayscale image can be scanned five times.

A threshold value determining method according to another aspect of the present invention is the same as the above, in which the concentrations of five points GO, ..., G4 are extracted.
, And the density G0 of the point of interest, whichever is smaller, Fn (n = 0, ..., 4) is selected, and the appearance frequency of each Fn obtained by scanning the grayscale image once is counted to obtain 5 Since the types of histograms are generated, it is possible to scan the grayscale image only once, and the circuit scale in that case can be reduced.

On the other hand, the extraction unit of the threshold value determining device according to the present invention determines the density G0 of the point of interest from the grayscale image and its density.
From the neighboring points, the densities G1, ..., G4 of four points selected so as not to include two points symmetrical with respect to the target point are extracted,
The selection section of the five points of concentration GO, which is sent from the extraction section,
, G4 is selected according to the selection signal n (n = 0, ..., 4), and its output Gn is used as one input of the second selection section.

In the threshold value determining device according to another invention, five (or four) units corresponding to the second selection unit are arranged in parallel as selection units, and (G0, G0), (G1, G0), ..., (G4,
G0) is provided and the frequency of appearance of each output Fn is simultaneously counted and provided to the threshold value determining unit by five histogram counting units, so that the grayscale image need only be scanned once, The selection unit is not necessary, and the circuit scale of the selection unit and the histogram counting unit can be reduced to about half of the conventional one.

[0035]

EXAMPLES Example 1. An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of one embodiment. In the figure, 1 includes a density G0 of a point of interest from a grayscale image and two points in the vicinity of 8 of them, which are symmetrical with respect to the point of interest. 4 points of density G1, ..., G
A local data extraction unit 2 for extracting 4 outputs 1 from the five densities G0, ..., G4 output from the local data extraction unit 1 according to a selection signal n (n = 0, ..., 4). A local data selection unit (first selection unit) 3 for selecting one of the output points Gn (n = 0, ..., 4) of the local data selection unit 2 and the point of interest output from the local data extraction unit 1 A minimum value selection unit (second selection unit) 4 that outputs the smaller value Fn (n = 0, ..., 4) by comparing with the density G0, 4 is the frequency of appearance of the output Fn of the minimum value selection unit 3. Histogram counting section 5 for counting
From the 5 types of histograms for n, the threshold value is 2
After calculating the sum A (T) and the area S (T) of the number of neighbors in the binarized image and calculating the average number of neighbors R (T),
It is a threshold value determining unit that determines an optimum threshold value based on the change state of the average number of neighbors R (T) with respect to the threshold value T.

Next, the operation will be described. First, the local data extraction unit 1 uses, for example, five latches 1 as shown in FIG.
It consists of one and one shift register 12, and the density G0 of the point of interest from the grayscale image to be binarized and the density of the eight neighboring points do not include two points which are symmetrical with respect to the point of interest. The four densities G1, ..., G4 selected for are extracted. Next, the local data selection unit 2 selects one of the five density data G0, ..., G4 output from the local data extraction unit 1 according to the selection signal n (n = 0, ..., 4). Select Gn. Next, the minimum value selection unit 3 compares the output Gn of the local data selection unit 2 with the density G0 of the point of interest sent from the local data extraction unit 1 and outputs the smaller value Fn. Now, when Fn is expressed using G0, ..., G4,

[0037]

[Equation 9]

[0038] By changing the selection signal n from 0 to 4 and scanning the grayscale image 5 times, each Fn is output from the minimum value selection unit 3, and the histogram counting unit 4 counts the frequency of appearance thereof to obtain 5 types of Fn. A histogram is obtained.

Now, as shown in FIG. 3A, considering a local region 13 of 3 × 3 and a local region 14 shifted by one pixel to the right of the local region 13, the densities (G0, G1) at two points in the local region 13 are considered. And the densities (G5, G0) at the two points in the local region 14 are the same. Therefore, F1 = min (G0, G0,
G1) and F5 = min (G5, G0) of the local area 14 have the same value. In other words, F1 and F5 in the whole image
Will have the same histogram.

Further, as shown in FIG. 3B, considering the local region 13 of 3 × 3 and the local region 15 with one pixel to the right and one pixel below, the densities of two points in the local region 13 are considered. (G0,
G2) and the density (G6, G0) at the two points of the local region 15 are
Each is the same. Therefore, F2 = m of the local area 13
in (G2, G0) and F6 = min (G
6 and G0) have the same value. That is, the histograms of F2 and F6 are the same in the whole image.

Similarly, the histograms of F3 and F7,
The histograms of F4 and F8 are the same. From the above, F
The number of pixels with density value k in the histogram of n hn
(k) is

[0042]

[Equation 10]

The following relationship holds. In addition, the number of pixels Hn (T) that satisfies equation (5)

[0044]

[Equation 11]

The relationship of is established. Therefore, when the equation (11) is applied to the equation (7), the sum A (T) of the numbers of adjacencies is

[0046]

[Equation 12]

Is represented by That is, the sum A of the numbers of adjacencies
(T) and area S (T) are the result of the minimum value selection process Fn (n =
0, ..., 4) are obtained from five types of histograms, and the average number of neighbors R (T) can be calculated based on these.

Example 2. By the way, in the above embodiment, F
The case where the histogram of n (n = 0, ..., 4) is counted by scanning the grayscale image 5 times while changing the selection signal n from 0 to 4 has been described. However, as shown in FIG. 3 and 5 histogram counting units 4 are arranged in parallel, and (G0, G0), (G1) are input to the respective minimum value selecting units 3 which constitute the selecting unit 30 according to another invention of the present application.
, G0), ..., (G4, G0), five kinds of histograms for Fn (n = 0, ..., 4) can be obtained by scanning the grayscale image only once. .. In this case, the local data selection unit 2 becomes unnecessary. Needless to say, the minimum value selection unit 3 having two inputs (G0, G0) can be omitted by directly inputting G0 to the histogram counting unit 4.

[0049]

As described above, according to the threshold value determining method and the apparatus therefor of the present invention, the density G0 of the point of interest from the grayscale image and the points of the eight neighboring points are symmetrical with respect to the point of interest. Density of four points G1, ..., G4 selected not to include two points
, G4 are selected from the extracted five densities G0, ..., G4, one for each scan of the grayscale image, and the selected densities Gn (n = 0, ..., 4) and the point of interest are selected. Concentration G
The smaller value Fn of 0 is selected, the appearance frequency of each Fn obtained by scanning the grayscale image five times is counted, five types of histograms are generated, and the optimum threshold value is determined based on this histogram. Since this is done, it is not necessary to scan the grayscale image 9 times as in the prior art, and only 5 scans are required, and the processing time can be shortened.

Further, according to the threshold value determining method and the apparatus therefor according to another invention of the present application, the extracted 5
Each of the density of points GO, ..., G4 and the density of the point of interest G0
Of these, the smaller value Fn (n = 0, ..., 4) is selected, and the appearance frequency of each Fn obtained by scanning the grayscale image once is counted to generate five types of histograms. There is an effect that it is possible to scan the grayscale image only once, and it is not necessary to arrange nine minimum value selecting units and nine histogram counting units in parallel as in the conventional case, and the circuit scale can be greatly reduced.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of the present invention.

FIG. 2 is a configuration diagram showing an embodiment of a local data extraction unit in the present invention.

FIG. 3 is an explanatory diagram showing the identity of minimum value selection processing.

FIG. 4 is a configuration diagram showing an embodiment of another invention.

FIG. 5 is a configuration diagram showing a conventional example.

FIG. 6 is a configuration diagram showing a local data extraction unit in a conventional example.

FIG. 7 is a configuration diagram showing another conventional example.

FIG. 8 is an explanatory diagram of the number of neighbors.

FIG. 9 is an explanatory diagram showing density distributions of an object and a background in an image.

FIG. 10 is an explanatory diagram showing how the average number of neighbors changes with respect to a threshold value.

[Explanation of symbols]

 1 Local Data Extraction Section 2 Local Data Selection Section (First Selection Section) 3 Minimum Value Selection Section (Second Selection Section) 4 Histogram Count Section 5 Threshold Determination Section 30 Selection Section

[Procedure amendment]

[Submission date] August 24, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0046

[Correction method] Change

[Correction content]

[0046]

[Equation 12]

Claims (4)

[Claims] 1. Regarding threshold value determination when binarizing a grayscale image, when a certain point of the binary image with respect to the threshold value T is "1", it is adjacent to this point of interest in the vicinity of eight of the points. The total sum A (T) of adjacent points that indicates the number of “1” points and “1”
Based on the area S (T) indicating the total number of points
(T) = A (T) / S (T) is obtained, and the optimum threshold is determined based on the change state of the average number of neighbors R (T) with respect to the threshold T. From the image, the density G0 of the point of interest and the densities G1 of the four points selected so as not to include two points symmetrical with the point of interest among the eight neighboring points G1 ,.
G4 is extracted, and one is selected for each scanning of the gray-scale image from among the extracted five points of density GO, ..., G4, and the selected density Gn (n = 0, ..., 4) is noted. The smaller value Fn of the point densities G0 is selected, the appearance frequency of each Fn obtained by scanning the grayscale image five times is counted, and five types of histograms are generated. Based on the generated histogram, After calculating the total number of adjacencies A (T) and the area S (T) of the "1" point to calculate the average number of adjacencies R (T),
A method for determining a threshold value, characterized in that an optimum threshold value is determined based on a change state of the average number of neighbors R (T) with respect to the threshold value T. 2. Regarding threshold value determination when binarizing a grayscale image, when a certain point of the binary image with respect to the threshold value T is "1", it is adjacent to this point of interest in the vicinity of eight of the points. The total sum A (T) of adjacent points that indicates the number of “1” points and “1”
Based on the area S (T) indicating the total number of points
(T) = A (T) / S (T) is obtained, and the optimum threshold is determined based on the change state of the average number of neighbors R (T) with respect to the threshold T. From the image, the density G0 of the point of interest and the densities G1 of the four points selected so as not to include two points symmetrical with the point of interest among the eight neighboring points G1 ,.
, G4 is extracted, and the smaller value Fn (n = 0, ..., 4) of each of the five extracted densities G0, ..., G4 and the density G0 of the target point is selected, and the grayscale image is set to 1 The frequency of appearance of each Fn obtained by scanning twice is counted to generate five types of histograms, and based on the generated histograms, the sum A (T) of the numbers of adjacencies and the area S (T ) Is calculated and the average number of neighbors R (T) is calculated, and then an optimal threshold value is determined based on the change state of the average number of neighbors R (T) with respect to the threshold value T. .. 3. Regarding threshold value determination when binarizing a grayscale image, when a certain point of the binary image with respect to the threshold value T is "1", it is adjacent to this point of interest in its 8 vicinity. The total sum A (T) of adjacent points that indicates the number of “1” points and “1”
Based on the area S (T) indicating the total number of points
(T) = A (T) / S (T), which is a threshold value determining device that determines an optimum threshold value based on the change state of the average number of neighbors R (T) with respect to the threshold value T. From the image, the density G0 of the point of interest and the densities G1 of the four points selected so as not to include two points symmetrical with the point of interest among the eight neighboring points G1 ,.
An extraction unit for extracting G4 and the output from the extraction unit 5
A selection signal n (n = 0, from among the density of points GO, ..., G4)
, 4), which selects and outputs one, and the smaller value Fn of the output Gn of the first selector and the density G0 of the point of interest is selected and output. A second selection unit, a histogram counting unit that counts the frequency of appearance of each output Fn of the second selection unit, and a sum A (of the adjacency numbers based on the histogram obtained from the histogram counting unit for each Fn. T) and the area S (T) of the point “1” are calculated to calculate the average number of neighbors R (T), and then the optimal threshold value is calculated based on the change state of the average number of neighbors R (T) with respect to the threshold value T. And a threshold value determining unit that determines the threshold value. 4. Regarding threshold value determination when binarizing a grayscale image, when a certain point of the binary image with respect to the threshold value T is "1", it is adjacent to this point of interest in its 8 vicinity. The total sum A (T) of adjacent points that indicates the number of “1” points and “1”
Based on the area S (T) indicating the total number of points
(T) = A (T) / S (T), which is a threshold value determining device that determines an optimum threshold value based on the change state of the average number of neighbors R (T) with respect to the threshold value T. From the image, the density G0 of the point of interest and the densities G1 of the four points selected so as not to include two points symmetrical with the point of interest among the eight neighboring points G1 ,.
An extraction unit for extracting G4 and the output from the extraction unit 5
Each of the point densities G0, ..., G4 and the point of interest density G0
A selection unit that selects and outputs the smaller value Fn (n = 0, ..., 4), and five histogram counting units that simultaneously count the frequency of appearance of each output Fn of the selection unit; Based on the histograms simultaneously obtained for each Fn from each histogram counting unit, the sum A of the adjacent numbers is obtained.
Calculate the area S (T) of (T) and the point of "1", and calculate the average number of adjacencies R
After calculating (T), the average number of neighbors R with respect to the threshold value T
A threshold value determining device, comprising: a threshold value determining section for determining an optimum threshold value based on a change state of (T).