JP2596613B2 - Threshold value determination device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a method of binarizing a grayscale image in a binarization process which is a typical process for separating an object in an image from a background. And an apparatus for determining an optimum threshold value.

[Related Art] Conventionally, a method based on a density histogram has been generally used for determining such a threshold value. However, according to this method, there is a problem that an appropriate threshold value cannot be determined for an image having a small object or an image having a low contrast. After calculating an evaluation function indicating the quality of the binarization result based on the feature amount measured from the above, a method of determining an optimum threshold value by judging from a change state of the evaluation function with respect to the threshold value, It has already been proposed by the present inventors (Japanese Patent Application No. 1-121719, filed on May 16, 1999).

Since the present invention relates to the improvement of this, first,
The configuration of the above-mentioned prior application invention will be described. For example, from each binary image when binarized while changing the threshold T,
As the feature quantity, the sum A (T) of adjacent numbers (the number of '1' points adjacent to this point in the vicinity of 8 when a certain point in the binary image is '1') and '1' The total number of points, ie the area S
(T) is obtained, and as an evaluation function, R (T) = A (T) / S
The calculation of the average number of neighbors defined by (T) and the determination of the optimum threshold value using the calculated number can be realized, for example, by the configuration shown in FIG.

In the figure, (6) is a threshold value processing unit for converting a grayscale image into a binary image for each threshold value T, (7)
Is a feature extraction unit that measures the sum A (T) and area S (T) of the number of neighbors from the binary image output from the threshold processing unit (6), and (5) is a feature extraction unit (7). ), The average number of neighbors R (T), which is an evaluation function, is calculated based on the feature amounts A (T) and S (T) obtained from This is a threshold value determining unit that determines an optimal threshold value based on the state.

 Next, the operation will be described.

First, the lower limit value T ₁ and the upper limit value T ₂ of the threshold to be changed are determined. Normally, the minimum and maximum values of the density of the grayscale image may be set to T ₁ and T ₂ , respectively. Then, in the section [T ₁ , T ₂ ], the following processes (I) and (II) are performed while changing the threshold T, and then the optimum threshold is determined in (III).

(I) The threshold value processing section (6) binarizes the grayscale image with respect to each threshold value T to obtain a binary image including points having a value of '0' or '1'.

(II) The feature extraction unit (7) calculates the sum A (T) of the number of neighbors from the binary image which is the output result of the threshold processing unit (6).
And the area S (T). Now, assuming that the number of points of '1' where the number of neighbors is i is a _i (T), (i = 0,..., 8),
A (T) and S (T) are represented as follows.

(III) In the threshold value determining section (5), the feature extracting section (7)
The average number of neighbors defined by R (T) = A (T) / S (T) is calculated based on the sum A (T) of the number of neighbors and the area S (T) obtained in the above.

The optimum threshold value is considered to be a threshold value at which the target area (connected component of '1') on the binary image is extracted as a “coarse (not jagged)” area, and the average number of neighbors is considered. R (T) is an evaluation function that indicates the goodness of unity of the area. In other words, it is considered that the goodness of the unity of the area means that the ratio of the internal points is high and the ratio of the boundary points is low. As shown in FIG. 6, the number of neighbors at the interior point is 8, and at the boundary point, the number of neighbors takes a value of 0 to 7 according to the shape of the neighborhood, so that the ratio of the interior points is high (the The better the unity, the higher the average number of neighbors. In general, when the object and the background in the image have a density distribution as shown in FIG. 7, the change in the average number of neighbors R (T) with respect to the threshold value T is as shown in FIG. Is determined as the optimum threshold.

[Problem to be Solved by the Invention] Since the conventional device is configured as described above, for all the threshold values to be changed, the threshold value is actually binarized and the feature amount (sum and area of the number of neighbors) is calculated. Since the measurement has to be performed, a processing time proportional to the number of thresholds to be changed is required. Further, in order to reduce the treatment time due to a request for real-time processing or the like, it is necessary to arrange in parallel the same number of threshold values to be changed (for example, the worst 256 in a gray scale image expressed in 256 levels). However, there is a problem that the circuit becomes large-scale.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and has a processing amount independent of the number of threshold values to be changed. The aim is to obtain a device that can measure quantities.

[Means for Solving the Problems] A threshold value determining apparatus according to the present invention provides a pixel density of a point of interest in a grayscale image, and densities of eight neighboring pixels adjacent to the center pixel with the point of interest as a center pixel. And a local data extraction unit (1) for extracting
The order of density 1 to 8 of two neighboring pixels is determined in descending order,
A ranking unit (2) for assigning a rank 0 to the center pixel; and a smaller density obtained by comparing the density of the pixel of each rank determined by the ranking unit with the density of the center pixel. A minimum value selecting unit (3) for outputting a value, and a histogram counting unit for generating a histogram using a pixel density as a variable from the output value of each rank of the minimum value selecting unit for all points of interest in the grayscale image. (4) and the sum S (T) of the number of the central pixels having a pixel density of T or more from the total sum of the specific pixel density T or more of the histogram of the rank 0.
From the sum of the specific pixel densities T or more of the histograms having the ranks 1 to 8, the pixel density T
The sum A (T) of the number of neighboring pixels having a pixel density of T or more among the neighboring pixels corresponding to the central pixel described above is obtained,
By dividing the A (T) by the S (T), two of the neighboring pixels corresponding to the center pixel whose binary density is “1” when the grayscale image is binarized by the pixel density T The number of pixels whose value density is '1', that is, the average value R of the number of neighbors
(T), and a threshold value determining unit (5) for determining a threshold value for binary imaging based on a change state of the average number of adjacent pixels R (T) with respect to the density T.

[Operation] In the present invention, first, the local data extraction unit (1) extracts the pixel density of a point of interest in a grayscale image and the density of eight neighboring pixels adjacent to the center pixel using the point of interest as a center pixel. Is done. Then, the ranking section (2) assigns the order of density 1 to 8 in descending order to the eight neighboring pixels adjacent to the center pixel, and assigns the order 0 to the center pixel. Minimum value selector (3)
Compares the density of the pixel of each rank with the density of the central pixel and outputs the smaller density value. A histogram counting unit (4) creates a histogram using a pixel density as a variable from the output value of each rank of the minimum value selecting unit for all the attention points of the grayscale image.

Then, the threshold value determining unit (5) obtains a total sum S (T) of the number of the central pixels having the pixel density of T or more from the total sum of the specific pixel density T or more of the histogram of the rank 0 (this sum). , The sum S (T) of the number of the central pixels having a pixel density equal to or higher than T is obtained (this sum S (T) corresponds to the area), and the histograms having the ranks 1 to 8 are specified. From the sum total of the pixel densities T or more, a total sum A (T) of the number of neighboring pixels whose pixel density is T or more among the neighboring pixels corresponding to the central pixel whose pixel density is T or more is obtained. ) Is divided by the above S (T), so that the binary density of the neighboring pixels corresponding to the central pixel whose binary density is “1” when the grayscale image is binarized by the pixel density T is “1”. The number of 1 'pixels, that is, the average value R (T) of the number of adjacent pixels is obtained. Ri, the threshold decision unit (5)
In the calculation of the feature amount, attention will be paid only to the case where the binary density of the central pixel is “1”. Then, a threshold value for binary imaging is determined based on the state of change of the average adjacent number R (T) with respect to the density T.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the overall configuration of the embodiment.
In the figure, (1) shows the density G _{0 of the} point of interest from the grayscale image,
The local data extraction unit extracts the densities G ₁ ,..., G ₈ of the eight neighboring points.
The G ₀ and output Z _0, when order signal j is 1-8, concentration G ₁ of 8 near points, ..., outputs j-th value from the larger of the G ₈ Z _j (j = 1, ... , 8), and (3) compares the output Z _j (j = 0,..., 8) of the ranking unit (2) with the density G _{0 of the} target point, and determines the smaller value F _j A minimum value selection unit that outputs (j = 0,..., 8), and (4) is an output of the minimum value selection unit (3).
A histogram counter for counting the frequency of occurrence of F _j (j = 0,..., 8), and (5) a histogram counter (4)
From the nine types of histograms obtained for F _j (j = 0,..., 8), the total sum A (T) of the number of adjacent pixels and the area S ( T)
After calculating the average number of neighbors R (T), the threshold value determination unit determines an optimum threshold value by judging from a change state of the average number of neighbors R (T) with respect to the threshold value T.

 Next, the operation will be described.

First, as shown in FIG. 2, for example, the local data extraction unit (1) includes nine latches (110) and two shift registers (1).
20), and sequentially extracts the density G ₀ (140) of the point of interest and the density G ₁ (151) to G ₈ (158) of the eight neighboring points from the grayscale image (130) input as a signal. I will do it.

Next, as shown in FIG. 3, for example, as shown in FIG. 3, when the ranking signal j (240) is 1 to 8, the densities G ₁ (151) to G ₈ (158) of points near 8 are obtained. And a ranking circuit (210) that outputs the j-th value from the larger one, and a ranking signal j (24
(0) is 0, and a zero determination circuit (220). When the rank signal j (240) is 0, the density G ₀ (140) of interest.
And a selector (230) for selecting the output of the ranking circuit (210) when the ranking signal j (240) is 1 to 8, and the output of this selector (230) is the output of the ranking unit (2). Z _j (250).

Next, the minimum value selecting section (3) outputs the output Z _j (j = 0,..., 8) of the ranking section (2) and the density G of the target point sent from the local data extracting section (1). ₀ and compare the smaller value F _j
(J = 0,..., 8) is output. Therefore, by changing the rank signal j from 0 to 8, F _j (j = 0,..., 8) is output from the minimum value selecting unit (3), and its appearance frequency is counted by the histogram counting unit (4). By doing so, nine types of histograms are obtained. Note that Z ₀ is G ₀ itself, and F ₀ is the smaller value of Z ₀ and G ₀ , so F ₀ is also
G ₀ itself. That is, a histogram for F ₀ is the normal density histogram.

Now, the histogram for F _j (j = 0,..., 8), that is, the frequency at which F _j takes the density value i is represented by H _j (i), (i = 0,
…, N−1; j = 0,…, 8). Where N
Is the number of density levels (for example, 256) representing a grayscale image. Then, in the histogram H _j (i) of F _j , the total number SH _j (T) whose density is equal to or greater than the threshold value T is Is represented by This SH _j (T) represents the total number of points of interest whose adjacent number is j or more in a binary image obtained by binarizing a grayscale image with a threshold value T. This is because, F _j is because it is smaller value of the density G ₀ of the target point and j th concentration Z _j from the largest in 8 near the target point, the point of interest has a density value of more than F _j, 8 Even in the vicinity, it is guaranteed that there are j or more objects having a density value equal to or greater than F _j . Therefore, if binarization is performed using F _j as a threshold, the point of interest is “1”.
In addition, since j or more of the eight neighborhoods are binarized to '1', the number of neighbors in this point of interest is j or more.

As described above, by using the property that SH _j (T) represents the total number of points of interest whose adjacent number is j or more in a binary image obtained by binarizing a grayscale image with a threshold value T, SH ₀ (T) = a ₀ + a ₁ + a ₂ + a ₃ + a ₄ + a ₅ + a ₆ + a ₇ + a ₈ , where a _i is the number of attention points having the adjacent number i.
(4) SH ₁ (T) = a ₁ + a ₂ + a ₃ + a ₄ + a ₅ + a ₆ + a ₇ + a ₈ (5) SH ₂ (T) = a ₂ + a ₃ + a ₄ + a ₅ + a ₆ + a ₇ + a ₈ (6 _{) SH 3 (T) = a} 3 + a 4 + a 5 + a 6 + a 7 + a 8 (7) SH 4 (T) = a 4 + a 5 + a 6 + a 7 + a 8 (8) SH 5 (T) = a 5 + a _{6 + a 7 + a 8 (} 9) SH 6 (T) = a 6 + a 7 + a 8 (10) SH 7 (T) = a 7 + a 8 (11) SH 8 (T) = a 8 (12) is satisfied . When both sides of the above equations (5) to (12) are all added, And the right-hand sides of Equations (1) and (13) are the same. Holds.

In addition, since the right side of the equations (2) and (4) is the same, S (T) = SH ₀ (T) (15) holds. That is, the sum A (T) of the adjacent numbers and the area S
It can be seen that (T) is obtained from the histogram H _j (i) of F _j .

In the threshold value determining unit (5), nine types of histograms H _j (i), obtained from the histogram counting unit (4),
Based on (i = 0,..., N−1; J = O,..., 8), A (T) and S (T) are calculated by equations (14) and (15) while changing the threshold T. Then, the average adjacent number R (T) = A (T) / S (T) is calculated. Thereafter, an optimum threshold value is determined by judging from the state of change of the average number of neighbors R (T) with respect to the threshold value T.

In the above embodiment, the case where the histogram of F _j (j = 0,..., 8) is counted while changing the rank signal j from 0 to 8 has been described. However, as shown in FIG. (2) Nine parts surrounded by a dotted line consisting of a minimum value selecting unit (3) and a histogram counting unit (4) are arranged in parallel, and the ranking signals (0) to (8) are assigned to each ranking unit (2). By giving, nine types of histograms for F _j (j = 0,..., 8) can be obtained simultaneously. In this way, it is not necessary to provide parallel circuits of the number N of threshold values to be changed, and when N >> 9, real-time processing can be easily realized with a small-scale circuit.

[Effects of the Invention] As described above, according to the present invention, the ranking unit and the selection unit rank 0 to 8 the density values of the 3 × 3 area extracted from the grayscale screen by the extraction unit. The calculated value is output, and the appearance frequency is counted by the counting unit, and the total number A of adjacent numbers A (T) and the area S (T) are obtained from nine types of histograms to calculate the average number of adjacent numbers R (T). Therefore, in the conventional example, all the threshold values to be changed are actually binarized, and the total number A (T) of the number of neighbors and the area S
As in the case of calculating (T), processing time proportional to the number of thresholds to be changed (for example, 256 steps) is not required,
Nine calculations for obtaining nine types of histograms when the rank signal is changed from 0 to 8 are sufficient, and the processing time is greatly reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an entire configuration of an embodiment of the present invention, FIG. 2 is a detailed configuration diagram showing a local data extracting unit in the embodiment, and FIG. 3 is a detailed configuration diagram showing a ranking unit in the embodiment. FIG. 4 is a block diagram showing another embodiment of the present invention, FIG. 5 is a block diagram showing a conventional example, and FIG.
FIG. 7 is an explanatory diagram of the number of neighbors, and FIGS. 7 and 8 are explanatory diagrams showing how the average number of neighbors changes with respect to the threshold value. (1) is a local data extraction unit, (2) is a ranking unit,
(3) is a minimum value selector, (4) is a histogram counter,
(5) is a threshold value determination unit. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (1)

(57) [Claims] 1. A threshold value determining apparatus for determining a threshold value for converting a grayscale image into a binary image, comprising: a pixel density of a point of interest in the grayscale image; A local data extraction unit for extracting the density of eight neighboring pixels adjacent to each other, and a first rank of the density of eight neighboring pixels adjacent to the central pixel.
8 is determined in descending order, and a ranking unit that assigns rank 0 to the center pixel, and compares the density of the pixel of each rank determined by the ranking unit with the density of the center pixel. A minimum value selection unit that outputs a smaller density value, and a histogram that creates a histogram using a pixel density as a variable from the output value of each rank of the minimum value selection unit for all points of interest in the grayscale image. A counting unit for calculating a total sum S (T) of the number of the central pixels having a pixel density of T or more from a total sum of the specific pixel density T or more of the histogram of the rank 0, and From the sum total of the specific pixel densities T or more in the histogram, the sum A (T) of the number of neighboring pixels whose pixel densities are T or more among the neighboring pixels corresponding to the central pixel whose pixel density is T or more is calculated. ( ) Is divided by the S (T), 2 of the neighboring pixels are binary density corresponding to the center pixel is "1" when binarizing the grayscale image at a pixel density T
The number of pixels whose value density is “1”, that is, the average value R (T) of the number of neighbors is obtained, and the threshold value for binary imaging is determined based on the change state of the average number of neighbors R (T) with respect to the density T. A threshold value determining unit for determining the threshold value.