JPH04205271A - Method and device for determining optimum binarized threshold value - Google Patents

Abstract

PURPOSE:To allow the stable extraction of an object to be detected by analyzing the degree of smoothness of the binary images binarized by respective threshold values, i.e., the ratio of straight parts, thereby determining the optimum binarized threshold value. CONSTITUTION:The threshold values are successively changed and are subjected to binarization processing. A picture element state analyzing section 1 divides the picture elements which attain the values above the binarized threshold values to three kinds; the flat points which are the picture elements to form the edges of the assembly of the picture elements, the internal points which are the internally existing picture elements and the other picture elements including the isolated points at which all the ambient picture elements are the picture elements to attain the values below the binarized threshold values. The analyzing section accumulates the respective appearance frequencies and records the frequencies by each threshold value. A threshold value determining section 106 determines the optimum binarized threshold value from the relation between the appearance frequency of the flat points over the entire part of the picture elements exclusive of the internal points and the binarized threshold values used for binarization processing and the relation between the appearance frequencies of the isolated points and the points on lines and the binarized threshold values. The binarized threshold value for extracting the object to be detected is obtd. in this way.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a method for determining an optimal binarization threshold in industrial image processing, and in particular, to determine the optimal binarization threshold for a segment whose target shape is unknown. The present invention relates to an optimal binarization threshold determining method and device capable of automatically determining a value.

[Conventional technology]

Conventional methods, for example, as described in JP-A-2-24787, binarize an image using multiple thresholds and calculate the number of black pixels, the total number of limbus, etc. from the binary pixels at each threshold. Accumulating feature amounts and changing predetermined parameters obtained from these,
Area of object to be detected.

The optimal binarization threshold was determined from graphical information specific to the object to be detected, such as its shape.

[Problem to be solved by the invention]

The above-mentioned conventional technology requires graphical information specific to the object to be detected in order to perform optimal binarization, and does not take into consideration the case where the shape of the object to be detected is unknown. Therefore, when performing image preprocessing for autonomous robots, etc. in artificial environments such as indoors where the shape, area, and number of objects to be detected change sequentially, it is necessary to optimally extract objects from images. It is difficult to uniquely determine the binarization threshold.

The purpose of the present invention is to provide a binarization threshold for stably extracting detection objects from an image, regardless of differences in graphical properties such as shape and area of detection objects, or changes in the number of objects. An object of the present invention is to provide an optimal binarization threshold determination method and apparatus that enables determination of the threshold value.

[Means to solve the problem]

In order to achieve the above object, the present invention converts the input image into
Binarization processing is performed by sequentially changing the threshold value, and the pixel state of each pixel of the obtained binary image is analyzed, and the appearance frequency of each type of analyzed pixel state is calculated for each binary image. At the same time, the cumulative result is recorded for each threshold value, and based on each data in the cumulative result record,
Alternatively, this can be achieved by determining a plurality of optimal binarization thresholds.

In addition, in the present invention, when accumulating the frequency of appearance by type of the analyzed pixel states described above, among the target pixels, those that take a value equal to or higher than the binarization threshold used in the binarization process are (a) A flat point, which is a pixel forming the edge of a collection of pixels that takes a value equal to or higher than the binarization threshold, (b) An internal point, which is a pixel located inside the collection of pixels, (
c) Divide the pixel into three types, including an isolated point, in which all pixels surrounding the pixel take a value less than the binarization threshold, and accumulate the appearance frequency of each pixel, and further,
The cumulative results are recorded for each threshold value.

[Effect]

For example, in image preprocessing by autonomous robots, etc. in artificial environments such as indoors, or in image preprocessing for automatic image learning, a binarization threshold is used to automatically extract unknown detection targets. A decision method is needed. In such cases, the object is an artificial shape, and its edges are often composed of straight parts.

As mentioned above, when the edge of the detection target is mainly composed of straight lines, among the pixels on the target image that have a value above the threshold, that pixel is The proportion of flat points increases when binarized with an appropriate threshold. This becomes an index indicating the straightness of the edge of each pattern in the image.

On the other hand, in the transition state to optimal binarization, the proportion of isolated points, points on lines, and uneven portions of the pattern increases.

In the present invention, as in the above configuration, flat points and internal points are used.

The appearance frequencies of each of the three types of other pixels including isolated points are recorded for each binarization threshold used in the binarization process.

Therefore, based on these data, the relationship between the appearance frequency of flat points for all pixels other than internal points among pixels that take a value above the threshold value and the binarization threshold used in the binarization process is calculated. The optimal binarization threshold value can be determined by searching for the relationship between the appearance frequency of isolated points, points on a line, etc., and the binarization threshold value.

In addition, in the present invention, the external information necessary for determining the optimal binarization threshold is only image information and does not include graphic information, so even if the area, shape, etc. of the detection target are unknown,
An optimal binarization threshold can be determined.

〔Example〕

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

FIG. 1 is an overall configuration diagram showing an embodiment of a device used for determining an optimal binarization threshold according to the present invention.

In the figure, 101 is an image input section, 102 is a binarization processing section, 103 is a pixel state analysis section, 104 is a plurality of accumulation sections,
105 is an accumulation result recording section, and 106 is an optimal binarization threshold determining section.

The overall processing operation in this embodiment using the apparatus configured as above will be described below.

FIG. 2 is a flowchart showing the overall processing. First, in step 201, an initial threshold value TH is set. In the next step 202, the binarization processing unit 102 binarizes the pixels using the set threshold value TH. In the next step 203, in the pixel state analysis unit 103,
A 3×3 block is set on the binary image binarized in the previous step, and the pixel state of the block center pixel is analyzed based on the shape pattern of the block. Here, 3×3
Regarding the center pixel of the block, the number of connections and the curvature coefficient can be defined as shown in FIG. 3, as described, for example, on page 10 of "Digital Image Processing [2] (written by Toriwaki, published by Shokodo)". Using this number of connections and curvature coefficient, the state of the central pixel point of the 3×3 block can be defined as shown in FIG.

An example of each state is shown in FIG.

Next, in step 204, the accumulator 104 accumulates the frequency of appearance for each type of pixel state.

In the next step 205, it is determined whether the pixel state analysis has been performed for all pixels, and the processes of steps 203 and 204 are repeated until the pixel state analysis has been completed for all pixels. When the pixel state analysis has been completed for all pixels, in step 206, the cumulative results accumulated in the accumulating unit 104 are recorded in the cumulative result recording unit 105 corresponding to the set threshold value TH. In the next step 207, it is determined whether or not the predetermined threshold value change has ended, and if it has not ended, the threshold value TH is changed in step 208,
The processing from steps 202 to 206 is repeated. In the next step 209, the optimal binarization threshold determining unit 106 determines the threshold based on the data of the appearance frequency of each pixel state at each threshold recorded in the cumulative result recording unit 105 through the above processing. One or more optimal binarization thresholds are determined by analyzing each data for changes in value.

The basic processing operations have been shown above, and the pixel state analysis unit 1
In 03, when analyzing a pixel state, it is possible to provide a pixel state dictionary table showing the pixel state of the target pixel corresponding to the shape pattern of one partial region. FIG. 6 shows an example of a pixel state dictionary table. In the figure, reference numeral 601 indicates target device X. Centered around
This figure shows a 3×3 block scanned over a binary image. In the following description, a pixel that takes a value equal to or greater than a threshold value will be referred to as a white pixel, and a pixel that takes a value less than the threshold value will be referred to as a black pixel. Here, if xo is a white pixel, then For each pixel from X to xe around , 1 bit corresponds to 1 pixel, and if it is a white pixel, 1. By setting O if it is a black pixel, the shape pattern of all 3x3 blocks is converted to 8 bits (code: 0) as shown at 602.
~255). For each shape pattern, the pixel state is calculated in advance from the number of connections and the curvature coefficient, and the correspondence between the coded shape pattern and the pixel state is set in a pixel state dictionary table indicated by reference numeral 603. Note that if xo is a black pixel, the data is excluded from the optimal binarization threshold determination process, as will be described later.

The pixel state dictionary table 603 is set as a preprocessing for the process shown in FIG.
The shape pattern of the partial area to be scanned is encoded by
By referring to the pixel state dictionary table 603, it is possible to uniquely determine the pixel state of the target pixel.

Here, hardware that executes optimal binarization threshold determination according to this embodiment will be explained. FIG. 7 shows a block diagram of the hardware.

FIG. 7(a) is a block diagram showing the overall configuration, and shows the system bus 701. CPU702. It consists of a memory 703 and an analysis block 704 that analyzes an input image from an image input device 705. FIG. 7(b) is a block diagram showing the hardware for the analysis block 704 when using the pixel state dictionary table. Below, Figure 7(b)
The analysis block will be explained below.

The grayscale image input from the image input device 705 is binarized in a binarization processing unit 706 based on a threshold value, and the output binary image signal is input to a shift register 707 . Then, three bits of the signal are taken out from each shift register to form a 3x3 block, and inputted to the 3x3 bit register 708. Each signal (0, 1) of the peripheral bits of the block is stored in the 5 pixel state dictionary table. As in the case of creation, it is assigned to each bit of the 8-bit register and input into the pixel state dictionary table 603.

The pixel state dictionary table 603 outputs a coded pixel state signal in response to the input pattern signal.

At this time, the center bit of the 3×3 block is input to the data accumulating unit 709 as an enable signal, and the data accumulating unit 709
In step 09, only when the enable signal is 1, the output from the pixel state dictionary table 603 is accumulated for each type of state signal.

By using hardware as described above, it is possible to analyze the state of each pixel of an image input from an image input device in real time.

Next, the cumulative result recording unit 10 according to the change in the binarization threshold
The optimum binarization threshold value is determined by analyzing the changes in the cumulative data within 5. A method for determining the optimal binarization threshold in the optimal binarization threshold determining section 106 will be described.

As described above, the grayscale image input from the image input section 101 is binarized in the binarization processing section 102 to obtain a binary image. Then, for each pixel of this binary image, the pixel state analysis unit 103 scans 3×3 blocks, and determines the pixel state of each pixel by (a) flat point, (b) internal point, (c) a.

White pixel points other than b, uneven parts within the edges of each completely binarized pattern, isolated points, points on lines, etc.
They are divided into three groups and accumulated in the accumulating unit 104, and the accumulated results are recorded in the accumulated result recording unit 105 corresponding to each threshold value. The above process is repeated by sequentially changing the threshold value. As a result, the cumulative result recording unit 105 records the appearance frequency of each pixel state with respect to the change in the threshold value.

Now, if the total numbers of flat points belonging to (a) above and white pixel points other than a and b belonging to above (c) are A and C, respectively, then A/(A+C), that is, all the points other than internal points The ratio of flat points to white pixel points is an index representing the straightness of the edge of each binarized pattern. That is, C is a value that increases when the threshold value is a transient value, while A is a value that increases when binarized with an appropriate threshold value.

Therefore, in the optimal binarization threshold determining unit 106, the ratio of flat points to the change in threshold value, that is, A/(A0c), is calculated, one or more of its maximum points are calculated, and the maximum point is The corresponding threshold is determined as the optimal binarization threshold for optimally detecting one or more types of patterns from the image.

FIG. 8 shows an example of how each of the above parameters changes with respect to a change in the threshold value based on the image 801. FIG. 8(a) is an example of the ratio of the 1,000 points to the change in the threshold value. In this example, thresholds Tl, T2 . T3
The percentage of flat points in (Tl<T2<T3), A/(A+
C) becomes maximum. That is, the threshold values Tl, T2 . T
There is an optimal threshold at [T1. Each section of T2), (T2°T3), and (:T3, 255) corresponds to each area within the image 801.

Furthermore, by adding conditions to the method for determining the optimal binarization threshold described above, it is possible to obtain a more accurate optimal binarization threshold. This will be explained next.

The above C is the total number of uneven parts on the edges, isolated points, points on lines, etc. in each binarized pattern, and as shown in Figure 8(b), the threshold value is is a value that increases when is a transient value.

Therefore, when determining the optimal binarization threshold in the optimal binarization threshold determination unit 106, when the appearance frequency of C is below a certain value, the optimal binarization is performed in the threshold of the corresponding area. Add the condition that a threshold exists. By doing this, the optimal binarization threshold determination unit 1
The one or more threshold values calculated in step 06 are used to optimally detect one or more types of patterns from the image.
A more accurate optimal binarization threshold can be determined.

Furthermore, if the total number of internal points belonging to (b) in the pixel state of each pixel described above is B, then B is the inner part of each binarized pattern, and B is relative to the rate of threshold change. The rate of change in appearance frequency, ie, dB/dTH, represents the speed of binarization of the binarized portion. The speed of this binarization becomes smaller when binarizing with an appropriate threshold value because the increase in the binarized part temporarily stagnates, and conversely increases when the threshold value is a transient value. Become. Therefore, when determining the optimal binarization threshold in the optimal binarization threshold determining section 106, if the binarization speed of the binarization part is small with respect to the change in the threshold, the corresponding threshold value A condition is given that the optimal binarization threshold value exists in the region. By doing this, as in the case of the previous condition, one or more threshold values calculated by the optimal binarization threshold determination unit 106 can be used to optimally detect one or more types of patterns from an image. can be determined as a more accurate optimal binarization threshold.

〔Effect of the invention〕

According to the present invention, the optimal binarization threshold value is determined by analyzing the degree of smoothness, that is, the proportion of straight line parts, for a binary image binarized using each threshold value. It is possible to determine a binarization threshold for stably extracting a detected object from an image regardless of differences in graphical properties such as area and shape.

Furthermore, even if there are a plurality of types of objects to be detected, it is possible to determine the binarization threshold value for optimally extracting each object to be detected.

Furthermore, by allocating bits to the 3×3 block shape pattern as shown in Figure 6 and providing a pixel state dictionary table for classifying the patterns, it is no longer possible to calculate the number of connections and curvature coefficient for each pixel, unlike the conventional technology. This saves time and effort, and is effective in speeding up the determination of the binarization threshold.

[Brief explanation of the drawing]

Fig. 1 is an overall configuration diagram of an apparatus according to an embodiment of the present invention, Fig. 2 is a flowchart showing the overall processing in the embodiment, Fig. 3 is a diagram showing the number of connections and curvature coefficients, and Fig. 4 is a diagram showing the number of connections and curvature coefficients. FIG. 5 is a diagram showing classification of pixel states of value images; FIG. 5 is a diagram showing some examples of pixel states of binary images; FIG. 6 is a diagram showing an example of a pixel state dictionary table used in this embodiment; FIG. 7 is a block diagram showing the hardware configuration of this embodiment, and FIG. 8 is a diagram showing an example of changes in each parameter with respect to changes in the threshold value. Explanation of symbols 101... Image input section 102... Binarization processing section 103... Pixel state analysis section 104... Accumulation section 105... Accumulation result recording section 106... Optimal binarization threshold Value determination unit 603...
Pixel status dictionary table 701...System bus 702...CPU 703...
・Memory 704 ・Analysis block 705・
・Image input device 706・Binarization processing unit 707・・Shift register 708・・3×3 bit register 709・・Data accumulation department Junnosuke Nakamura, patent attorney Attorney Nc(χo) Nc(z. )=E(Naga-Nagaoka, Hiro,)*tj. ”441aR(Zo) R=1 2Nc
fhfg++ String εS Figure 3 Figure 4 Figure 5 6o2 Figure 6 (Q) 7o4

Claims (1)

[Claims] 1. Binarize the input image by sequentially changing the threshold value, analyze the pixel state of each pixel of the obtained binary image, and , accumulates the frequency of appearance of each type of the analyzed pixel state, records the cumulative result for each threshold value, and sets one or more optimal binarization thresholds based on each data in the cumulative result record. An optimal binarization threshold determining method characterized by determining a value. 2. In the optimal binarization threshold determination method according to claim 1, the analysis of the pixel state is performed by scanning a binary image in a partial region of an arbitrary size, and determining the pixel state of the target pixel from the shape pattern of the partial region. A method for determining an optimal binarization threshold, characterized in that it is a method for determining an optimal binarization threshold. 3. In the optimal binarization threshold determination method according to claim 2, a pixel state dictionary table is provided for uniquely determining the pixel state of the target pixel with respect to the shape pattern of the partial region when analyzing the pixel state. An optimal binarization threshold determination method characterized by the following. 4. In the optimal binarization threshold determination method according to any one of claims 1 to 3, the cumulative appearance frequency of each type of analyzed pixel state is determined by the number of pixels used for the binarization process among the target pixels. For those that take values above the binarization threshold,
A flat point, which is a pixel forming the edge of a collection of pixels that takes a value equal to or higher than the binarization threshold, an interior point, which is a pixel located inside the collection of pixels, and all pixels surrounding the pixel have the above binary value. 1. A method for determining an optimal binarization threshold, characterized in that the pixels are divided into three types including isolated points, which are pixels that take a value less than the threshold, and the frequency of appearance of each pixel is accumulated. 5. In the optimum binarization threshold determination method according to claim 4, the optimum binarization threshold is determined based on all changes in the frequency of appearance of pixel states with respect to changes in the threshold value, excluding internal points. 1. A method for determining an optimal binarization threshold, characterized in that one or more threshold values at which the ratio of the appearance frequency of flat points to the appearance frequency becomes maximum are determined as the optimal binarization threshold. 6. In the optimum binarization threshold determination method according to claim 5, when the appearance frequency of isolated points, etc. is below a certain value when determining the optimum binarization threshold, the area corresponding to the appearance frequency An optimal binarization threshold determination method characterized by providing a condition that an optimal binarization threshold exists among the thresholds. 7. In the optimum binarization threshold determination method according to claim 5, when determining the optimum binarization threshold, the degree of binarization progress in the binarization portion with respect to a change in the threshold value is lower than a predetermined value. 1. A method for determining an optimal binarization threshold, characterized in that, when the threshold is small, the optimal binarization threshold exists among the thresholds of the corresponding region. 8. An image input means for inputting an image, a binarization processing means for binarizing the image obtained from the image input means by sequentially changing a threshold value, and a pixel conversion means for each pixel of the obtained binary image. a pixel state analysis means for analyzing the state; a plurality of accumulation units for accumulating the frequency of appearance by type of pixel states analyzed by the pixel state analysis means for each binary image; It is characterized by comprising a plurality of cumulative result recording sections that record separately, and an optimal binarization threshold determining means that determines one or more optimal binarization thresholds based on the data in the cumulative result recording section. Optimal binarization threshold determination device.