JPS60118979A - Peak extracting processing device - Google Patents

Abstract

PURPOSE:To extract only an effective local peak point by scanning a data of two-dimension histogram from a window, extracting a maximum point being a peak canditate point and selecting the maximum point of a prescribed level or over. CONSTITUTION:A shift buffer 3 gives a timewise delay to a data of the two- dimension histogram read sequentially from a memory 1 corresponding to a window 2 and prepares a data group adjacent on the two-dimension space. A maximum point extracting circuit 4 is provided with a comparator section 6, which compares the data at the center of the window applied from the shift buffer 3 and the data therearound, and when the central data is larger than any data therearound, a signal representing the maximum point is outputted via an AND circuit 7-1. A threshold circuit 5 compares a predetermined threshold value with the central data of the noted window at present, and when the said data value is larger than the threshold value, a selection signal is outputted. The selection signal selects the maximum point by the AND circuit 7-2.

Description

Detailed Description of the Invention (5) Technical Field of the Invention The present invention relates to a peak extraction processing device, particularly to histogram processing in statistical processing in image processing. This is a dispute regarding a peak extraction processing device that extracts peak points.

iB) Background and problems of the technology In order to recognize two objects, such as in the eyes of a robot, from the density data of images obtained with a television camera,
*Hoxgh as an effective method to extract line drawings
conversion is known.

FIG. 1 is a diagram for explaining Holtgh transformation, and FIG. 2 shows an example of a two-dimensional histogram.

When recognizing objects using robot eyes, etc.

First, from image data of the object, various well-known techniques are used.

gr1, which extracts one contour in which a point on the contour line is "1" and all other points are "0", represents a sequence of points on such a contour. When extracting straight lines from these points 1. For example, by Hough transformation.

The process of mapping straight lines to points is performed as follows.

By mapping a straight line to a point, you can see the edges of the straight line,
Extraction with less influence of fluctuation becomes possible.

Ikutsukano point (zi, vi), (i=1,2...,
N) is.

Assuming they are on the same straight line.

vi= Coz(+6(1<=L 2+ −N −−(
1). Consider these equations with (cLO, bO) as variables.

(α, 6) Expressed on a plane.

ho ”-Zt (Lo + yi ・・・・・・(2
). Although all these straight lines are different, +<Zi, u
i) are on the same straight line, so they are (c
L, 6) It passes through two points (C0, bO) on the plane. Utilizing this fact, from the point sequence (re+''), (==t, 2...N) extracted by preprocessing.

As shown in FIG. 1(b), a histogram regarding the above equation (1) is created on the (α, h) plane, and the point at which the count value is maximum is found.

FIG. 2 is an example of a three-dimensional representation of the two-dimensional histogram of the Hosgh transformation obtained as described above. In this conversion, if there is only one straight line to be extracted, the peak can be extracted relatively easily, but in general there are multiple straight lines, and the S/N is poor and the peak is not very steep. Therefore, it is not easy to extract a meaningful peak point that is valid as a straight line.

This is not limited to the above-mentioned example of straight line extraction using Hough transformation, but also in general digital statistical processing.

There is a need for a technique for extracting reliable local peak points from two-dimensional histograms.

(C1 Object and structure of the invention The present invention aims to solve the above points, and 2.
It is an object of the present invention to provide a device that extracts only local peak points from a dimensional histogram with a good S/N ratio. Therefore, the peak extraction processing device of the present invention is a peak extraction processing device that extracts locally high-level peak points of a two-dimensional histogram.The data of the two-dimensional histogram is scanned by an nxm window, and peak candidates are A local maximum point extraction circuit that extracts a local maximum point that is a point, and a threshold circuit that compares the value of the local maximum point extracted by the local maximum point extraction circuit with a predetermined threshold value and selects a local maximum point that is at a predetermined level or higher. It is characterized by extracting locally meaningful peak points.

Further, the second aspect of the present invention provides a peak extraction processing device for extracting locally high-level peak points of a two-dimensional histogram, which scans data of the two-dimensional histogram using an nXm window and filters the scanned data. with a filter circuit.

n′xm from the data modified by the filter circuit
A local maximum point extraction circuit that extracts local maximum points as peak candidate points using a window of It is characterized by being equipped with a threshold circuit and extracting locally significant peak points. Embodiments will be described below with reference to the drawings.

(D) Embodiment of the Invention FIG. 3 is an explanatory diagram of a window adopted in the present invention, FIG. 4 is a configuration of an embodiment of the present invention, and FIG. 5 is a detailed circuit diagram of the embodiment shown in FIG.

In the figure, 1 is a memory in which two-dimensional histogram data is stored, 2 is a window for scanning the data in the memory 1, and 3 is a shift bumper.

4 is a local maximum point extraction circuit, 5 is a threshold circuit, and 7 is an AND circuit for realizing the function of window 2.

8 is a gate circuit, 10 and 11 are shift registers, 1
2 to 20 are flip-flops, 21 is a threshold generation unit,
22 to 30 represent comparators.

In our case, by scanning the histogram space.

In order to extract local maximum points that are peak candidates, a window as shown in FIG. 3 is used. The window 2 shown in FIG. 3(b) is for extracting 3×3 data elements, and the window 2' shown in FIG. 3(b) is for extracting 5×5 data elements.

When using window 2, the data in the center is.

If it is larger than any of the surrounding data α+b+'+'+/+g+h+', it is determined to be a local maximum value. This window 2
By sequentially moving the two-dimensional histogram data in the memory 1 in the horizontal and vertical directions, several maximum points will be obtained.

The same applies to the case where the window 2' shown in FIG. 3(b) is used, and when the data value is larger than any of the data a to 21% to V, the data sacrifice is considered to be the maximum value. Although the size of the window may be arbitrarily determined according to the data characteristics of the two-dimensional histogram, in the following example, a case will be described in which a 3×3 window 2 is used to simplify the explanation.

In FIG. 4, the shift buffer 31i applies a time delay corresponding to the window 2 to the two-dimensional histogram data sequentially read out from the memory 1.
It prepares data groups that are adjacent in a dimensional space. The maximum point extraction circuit 4 includes a comparison section 6, which compares the data at the center of the window supplied from the shift buffer 3 with the data around it.

When the central data is larger than any surrounding data, a signal indicating the local maximum point is outputted via the AND circuit 7-1. However, the maximum point obtained in this way does not always represent a valid line drawing in the histogram converted to Ho5g as shown in FIG. 2, for example. Therefore,
A threshold circuit 5 is provided to select only local maximum points above a certain predetermined level. That is, the threshold circuit 5 compares a predetermined threshold with data at the center of the currently focused window, and outputs a 1 selection signal when the data value is larger than the threshold. With this selection signal,
The maximum point is selected by the AND circuit 7-2, and the extreme value serving as the peak point is outputted via the gate circuit 8. In addition,
The threshold value may be changed depending on, for example, the complexity of the orchid.

The specific circuit configuration of the illustrated embodiment in FIG.
It is as shown in the figure. In memory l, for example, 25
6x256 data is stored.

One piece of data is sequentially read out from the address with the smallest address every unit time, and is supplied to the shift register 10 and the flip-flop 18. Note that the shift register 10.
The number of flip-flops 18 corresponding to the bit width of data is prepared, and the same is true for the other flip-flops. The output of the shift register 10 is read from the memory 1, and then 25
The data is delayed by 6 units of time and is supplied to shift register 11 and flip-flop 15, and shift register 11 further provides a delay of 256 units of time. Adjacent flip-flops 12-20 are.

Each holds data with a time difference of one unit time.

Therefore, each flip-flop 12-20 has data α-i corresponding to window 2 for each unit time.

Comparators 22-29 receive data α-d of flip-flops 12-15 and 17-20, respectively.

It compares f-i with data e of the flip-flop 16, and outputs logic rlJ if data e is larger. On the other hand, the data ε of the flip-flop 16 is.

The comparator 30 compares the threshold value th generated by the threshold r direct generation unit 21 in magnitude, and the comparator 30 outputs logic "1" when the data ε is large. When the outputs of the comparators 22 to 30 all become logic "1", the AND circuit 7 outputs the extreme value.

In this example, the extreme value circuit and the threshold value circuit are arranged in parallel, but they can also be arranged in series, and the order does not matter.

Next, a second embodiment of the present invention will be described.

FIG. A16 shows an example of a histogram for explaining the present invention in detail, an OFF diagram shows a configuration of an embodiment of the second invention, and FIG. 8 shows an example of a filter circuit shown in the OFF diagram.

In the figure, numerals 1 to 5 and 7 correspond to those in FIG.

40 is a filter circuit, 41 is a shift buffer, 42 is a multiplication circuit, 43 is an addition circuit, and 44ζ is a division circuit.

50 to 58 are multipliers, 60 to 67 are adders, 7
0 and 71 represent coefficient generators.

For example, in the histogram space, it is not often the case that the peaks forming the maximum value are smooth; in fact, as shown in FIG. 6, for example, there are many irregularities due to noise and the like. in this case.

With only the method shown in Figure 4, it is possible to
L, P2. P3 and a plurality of local maximum values may be detected. Also, for example, when the top of the mountain is blurred, as in the case of peak P4 in Figure 16.

It may not be detected as a local maximum value. Therefore.

In the case of the present invention, - Prior to extracting the local maximum points, filtering is performed on the data of the two-dimensional histogram.

The off-illustrated filter circuit 40 is a circuit that performs filter processing on data sequentially read out from the memory 1. For example, the filter circuit 40 is a shift buffer 41.

It consists of a multiplication circuit 42, an addition circuit 43, and a division by 1 circuit 44. The shift buffer 41 functions similarly to the shift buffer 3 described above.

Data is prepared in a 3×3 window 2. Filtering is performed using predetermined filter coefficients z1 to z1 corresponding to data a-i of 3×3 window 2.
z9 is multiplied by 1 by the 1 multiplication circuit 42, and data α to i are multiplied by the primary addition circuit 43.

This is carried out by adding all of them together and performing weight correction by the division circuit 44 as necessary. That is, when filter coefficients z1 to x9 are given.

Is the data e in the center of window 2 based on a linear equation?
will be corrected.

, s=(αx1+ bw2+-=+ex5+-+1z9
)/(n+z2+-+z9) As a low-pass filter for reducing noise and smoothing peaks, a filter having the following filter coefficient is used, for example.

However, when zH-s2+...+x9=0, the denominator is 1.
shall be.

On the other hand, in order to emphasize the peak point and form a steeper mountain, a second Laplacian filter, for example, is used.

Either one of the low-pass filter and the Laplacian filter may be used, or both may be used. Further, the filter coefficients may be changed as appropriate. A clear histogram can be obtained by the filter circuit 40. This output is supplied to shift buffer 3, but shift buffer 3. Maximum point extraction circuit 4. Threshold circuit 5
Since the extraction of peak points by is the same as that shown in FIG. 4, further explanation will be omitted.

Multiplication circuit 42 of filter circuit 40. The adder circuit 43 and the divider circuit 44 are constructed as shown in FIG.
0 to 58 are 2 data α to i, respectively.

Filter coefficients x1 to x9 supplied from the coefficient generator 70
Multiply by Adders 60 - 67 add the outputs of 1 multipliers 50 - 58 and output the result to divide-by-2 circuit 44 . The division circuit 44 divides the addition result by the coefficient of the coefficient generator 71.

In the above embodiment, the window size is set to 3×3, but other sizes can be used in the same manner. Also, here is the circuit for extracting the above peak.

From the point of view of speed, a 71-doware circuit is preferable.

1 by means of software circuitry, if necessary.
Substitutable.

fEl As described in detail, according to the present invention, it is possible to extract only effective local peak points from a two-dimensional histogram.1 For example, when applied to object recognition by a robot's eyes, etc. The recognition rate can be improved.

[Brief explanation of the drawing]

At1 diagram is a diagram for explaining Hoxgh transformation.
The figure is an example of a two-dimensional histogram, Figure 13 is an explanatory diagram of a window related to the present invention, Figure 4 is the configuration of an embodiment of the present invention, Figure 215 is a detailed circuit diagram of the illustrated embodiment, Figure 6 is The figure is an example of a histogram for explaining the present invention in detail, the OFF figure is the configuration of an embodiment of the present invention of N12, and the O8 figure is 2
FIG. 17 shows an example of the illustrated filter circuit. In the figure, 1 is a memory, 2 is a window, 3 is a shift buffer, 4 is a maximum point extraction circuit, and 5 is a threshold circuit. 40 represents a filter circuit. Patent applicant Fujitsu Ltd. Representative Patent Attorney Mori 1) Hiroshi (1 other person) 2 layer t↑ Yield 1, b-

Claims (2)

[Claims] (1) In a peak extraction processing device that extracts locally high-level peak points of a two-dimensional histogram, the data of the two-dimensional histogram is scanned by an nxm window, and local maximum points that are peak candidate points are extracted using the window. Peak extraction characterized by comprising: a maximum point extraction circuit; and a threshold circuit that compares the value of the maximum point extracted by the maximum point extraction circuit with a predetermined threshold value and selects a maximum point that is at a predetermined level or higher. Processing equipment. (2) A peak extraction processing device that extracts locally high-level peak points of a two-dimensional histogram, including a filter circuit that scans data of the two-dimensional histogram using an nxm window and filters the scanned data; n' from the data modified by the circuit
A local maximum point extraction circuit that extracts local maximum points that are peak candidate points using a window of A peak extraction processing device comprising a threshold circuit for selection.