JPH0143348B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (1) Technical Field of the Invention The present invention relates to an image processing method, and particularly to an improvement of an image processing method that is effective in extracting extreme points of a grayscale image.

(2) Background of the technology In general, images obtained from TV cameras, etc.
It is believed that the parts where the brightness changes suddenly are important points corresponding to the edge points of the object's surface, and extracting these points as a series of connections is useful for pattern recognition of the object. This is an important process.

(3) Prior art and problems Based on the above-mentioned idea, the applicant had previously proposed the following image processing method (Patent Application No. 1983).
−104714). For example, this can be done by subdividing the multilevel grayscale image obtained through preprocessing into pixel units.
After comparing the gray levels of the pixel of interest and the pixels adjacent to it in the horizontal and vertical directions, if the pixel of interest has a high level among the three pixels lined up in either direction, the pixel of interest is extracted as the extreme point. The idea is to do so. According to this method, by finding the above-mentioned extreme points as a series of connections, it is possible to recognize the outline of an object.

However, in such image processing methods, for example, when recognizing the contours of overlapping objects or when recognizing the contour of an object with corners, the pixels corresponding to the branching points of the contours are divided into two. Therefore, if the respective contour lines were to extend horizontally and vertically in the vicinity of the pixel, there would be a situation where the pixel corresponding to the branch point of the contour could not be extracted as an extremity. As a result, the outline of the object is interrupted, and recognition of the object becomes inaccurate accordingly.

(4) Purpose of the Invention The present invention has been made based on the above-mentioned viewpoints, and its purpose is to eliminate discontinuities in the contours of objects, thereby making it possible to accurately recognize objects. An object of the present invention is to provide an image processing method.

(5) Structure of the present invention The basic structure of the present invention is to divide an image obtained from a receiver or the like into a plurality of pixels arranged in a matrix, and to and an image processing method in which a pixel having a larger difference in gray level than any two pixels lined up in the left and right diagonal directions is extracted as an extreme point.

(6) Embodiments of the invention Hereinafter, the present invention will be described in detail based on embodiments shown in the accompanying drawings.

First, the basic principle of the extreme point extraction process according to the present invention will be explained based on FIG. Now, to simplify the explanation, the density distribution in the one-dimensional direction of the original image is
As shown in Figure a, in the case of P, find the extreme point R. First, as shown in Fig. 1b, an image pattern obtained by shifting the original image P by one pixel △ to the left is set as P _L , and as shown in Fig. 1c, this left-shifted image pattern P _L and the original image are A portion Q _L of which the density of the original image P is greater than or equal to that of the original image P (hereinafter referred to as P≧P _L ) is extracted by comparing the original image P with The image pattern shifted by one pixel △ is P _R , and Fig. 1 e
As shown in FIG. 2, this right-shifted image pattern P _R and the original image P are compared to extract a portion Q _R in which the density of the original image P is greater than or equal to that of the original image P (hereinafter, this will be referred to as P≧P _R ). Next, since the sought extreme point R is considered to be at least larger _than or at the same level as the pixels on both sides, as _shown in FIG. By extracting _L Q _R ), we can obtain the extreme point R. However, when P=P _L =P _R , the pole R does not exist and is therefore excluded.

Now, if we subdivide a two-dimensional image into a matrix, that is, into pixel units aligned in the horizontal and vertical directions, we can use the basic principle described above to divide the pixel of interest in the two-dimensional image into horizontal, vertical, and left-aligned pixels. It is possible to individually determine whether the point is the extreme point in the corner direction (upward left direction with respect to the horizontal direction) or the right diagonal direction (upward rightward direction with respect to the horizontal direction). In this case, in order to determine whether the pixel of interest E is a pole in each direction, the second
As shown in the figure, a window W is set up by a pixel of interest E and eight pixels A, B, C, D, F, G, H, and I surrounding it.

FIG. 3 shows a specific algorithm for extracting extreme points using the window W described above.

First, to explain the case of determining whether or not a pixel of interest E of a two-dimensional image is a polar point in the horizontal direction, the density levels of the pixel of interest E and pixels D and F on both sides of this pixel of interest E are compared. ,E≧D
(The density of pixel E is greater than or equal to the density of pixel D, and the same applies hereafter), then S _L = 1, otherwise S _L = 0, and if E≧F, S _R = 1,
Otherwise, let S _R = 0, where E=D=F
Then, since the gray level does not change, S _N =0,
Otherwise, S _N =1. Then, the index S indicating whether or not the pixel of interest E is a pole in the horizontal direction is obtained by the AND operation of S _L , S _R and S _N , that is, S = S _L ×S _R ×S _N. , S=1 indicates that the pixel E is a pole, and S=0 indicates that the pixel E is not a pole.

Next, to explain the case of determining whether or not the pixel of interest E of a two-dimensional image is a pole in the vertical direction, the pixel of interest E, the pixels B above and below this pixel of interest E, and the pixels B above and below this pixel E and Compare the shading level with H, and if E≧B, T _L = 1, otherwise, T _L = 0, and if E≧H, T _R = 1,
Otherwise, let T _R =0, where E=B=
If H, T _N =0, otherwise T _N =1. Then, the index T indicating whether or not the pixel of interest E is a pole in the vertical direction is obtained by the AND operation of T _L , T _R and T _N , that is, T = T _L ×T _R ×T _N , and T
= 1 indicates that pixel E is the pole, and T =
When it is 0, it indicates that the pixel E is not at the extreme point.

Also, a case will be explained in which it is determined whether or not the two-dimensional pixel E is a pole in the right diagonal direction (the same applies to the left diagonal direction, so it will be indicated in brackets below). , substantially in the same way as described above, a pixel of interest E and a pixel C adjacent to this pixel of interest E in the right diagonal direction (left diagonal direction).
and G (A and I), and compare the shading levels with E
If ≧C (E≧A), then U _L =1 (V _L =1), otherwise, U _L =0 (V _L =0), or E≧G (E
≧1), then U _L =1 (V _L =1), otherwise
Let U _L = 0 (V _L = 0), but E = C = G (E
=A=I), then U _N =0 (V _N =0), otherwise, U _N =1 (V _N =1). Then, the index U(V) indicating whether or not the pixel of interest E is a pole in the right diagonal direction (left diagonal direction) is U _L , U _R and
The conjunction operation of U _N (V _L , V _R and V _N ), i.e. U = U _L
×U _R ×U _N (V=V _L ×V _R ×V _N ),
When U(V)=1 or 0, it indicates that the pixel E is a pole or not a pole.

Furthermore, an index X indicating whether or not the pixel of interest E is an extreme point in any direction on the screen is S,
OR operation of T, U and V, i.e. X=S+T+U
+V, and when X=1, it indicates that pixel E is a pole in either direction, and X=
When it is 0, it indicates that the pixel E is not a pole in any direction, and the value when X = 1 is set to high level, and the value when X = 0 is set to low level to extract the pole. It is. and,
When the window W is moved on the screen using such an algorithm, all the extreme points on the two-dimensional image will be extracted.

An example in which the above-described image processing method is applied to an object recognition device will be shown below.

The basic configuration of this object recognition device, as shown in FIG.
A converter 2, a spatial filter 3 that performs first-order or second-order differential calculations on the digital image data signal from the AD converter 2, and stores data signals necessary for polar point extraction processing among the digital image data signals from the spatial filter 3. Store and
A data shift circuit 4 that shifts a data signal in synchronization with the horizontal scanning signal of the TV camera 1; and a data processing circuit 5 that performs calculations for polar point extraction processing based on the data signal stored in the data shift circuit 4. Consists of.

Now, as shown in FIG. 5, if a circular object 7a and a triangular object 7b are placed on top of each other on the table 6 and these objects 7a and 7b are photographed by the TV camera 1, the TV camera 1, a grayscale image signal is output in synchronization with the horizontal scanning signal of the TV camera 1, and a multilevel grayscale image is obtained subdivided into pixel units. And the above TV
The output signal from the camera 1 is converted to, for example, an 8-bit digital image data signal via an AD converter 2. In this case, as shown in FIG. 6, if we pay particular attention to the divergent parts of the contours of the objects 7a and 7b (the part surrounded by the dashed line) in the photographed image of the TV camera 1, we can see that the digital shading based on the digital image data signal is The image appears as shown in FIG. 7, with each data value indicating the gray level at each pixel. Then, the output signal from the AD converter 2 passes through the spatial filter 3 and is converted into a digital image data signal with enhanced gray level. In this case, the digital gray image is as follows:
For example, it is expressed as shown in FIG. Furthermore, the 8th
Each data value in the figure indicates the gray level at each pixel.

As shown in FIGS. 4 and 9, if the number of pixels arranged in the horizontal direction of the captured image screen of the TV camera 1 is assumed to be N, the data shift circuit 4 has a function of 2N+3.
A register that stores data corresponding to each pixel, for example, an 8-bit data width register.
It consists of R _e1 to R _e9 and shift registers S ₁ and S ₂ , which also have an 8-bit data width and store data corresponding to N-3 pixels.
The data stored in R _e1 to R _e9 correspond to pixels I to A shown in window W in FIG. 2. That is, as shown in FIG.
When the photographed _image screen of the TV camera 1 is sequentially scanned from above to below by horizontal scanning lines _{h 1} , h ₂ , h ₃ . ₃
At the stage when the scanning of the first three pixels is completed, the data shift circuit 4 stores 2N-3 pixel data, and the register
Data corresponding to pixels g33 (the pixel located in the third row and third column on the photographed image screen, the same applies hereinafter), g32, g31, g23...g11 is stored in R _e1 to R _e9 , respectively. Each piece of data is used as data for pixels I to A of window W. When the scanning of pixel g34 by scanning line _h3 is completed, the data corresponding to pixel g34 is newly stored in register R _e1 , and the data in each register is shifted as a whole. At this stage, pixels g34, g33, g32, g24, . . . are stored in registers R _e1 to R _e9 , respectively.
The data corresponding to g12 will be stored, and the window W will be moved one row to the right. Furthermore, as scanning by scanning line _h3 progresses, the window W shifts to the right in synchronization with the scanning signal, and when scanning by scanning line _h4 progresses, the window W moves down one line. After that, it moves sequentially from left to right. In this way, the window W moves sequentially as the scanning line m scans, and at the stage when the scanning line m has finished, the window W has moved over the entire range of the photographed image pixel by pixel. , As a result, all the data necessary for the pole point extraction process are read, and each data is sequentially sent to the data processing circuit 5.
sent to.

Further, the data processing circuit 5 is shown in FIGS. 4 and 11.
As shown in the figure, a horizontal comparison unit 5a that compares data between a pixel of interest E and pixels D and F adjacent to the pixel of interest E in the horizontal direction;
A vertical comparison unit 5b compares data between the pixel E and pixels B and H adjacent to the pixel E in the vertical direction, and a pixel C and G adjacent to the pixel E in the right diagonal direction. A right diagonal comparison unit 5c that compares data, a left diagonal comparison unit 5d that compares data between a pixel of interest E and pixels A and I adjacent to the pixel of interest E in the left diagonal direction, and a comparison result. Comparators C _p1 ... C _p8 , AND gates AN ₁ , AN ₂ , OR gates OR ₁ ... OR _9, NAND gates NA ₁ ... NA ₄ , and data of the target pixel E is set and the pixel of interest E is at the extreme point, the peak register PM outputs the data,
The circuit is configured to execute the algorithm shown in FIG. That is, comparator C _p1 compares pixels E and D
The comparator C p2 compares the data of pixels E and F and outputs "1" when E≧D, and the comparator C _p2 compares the data between pixels E and F and outputs "1" when E≧F. The comparator C _p3 compares the data of pixels E and B and outputs "1" when E≧B, and the comparator C _p4 compares the data of pixels E and H and outputs "1". It outputs "1" when E≧H, and comparator C _p5 compares the data of pixels E and C and determines that E≧
It outputs "1" when C, and the comparator
C _p6 compares the data of pixels E and G and outputs "1" when E≧G, and comparator C _p7 compares the data of pixels E and A and outputs "1" when E≧A. The comparator C p8 outputs "1", and the comparator C _p8 is connected to the pixel E
It compares the data of and I and outputs "1" when E≧I. Then, if the comparators C _p1 and C _p2 both output "1" and E=D=F is not the case, in other words, if the pixel E is the horizontal pole, the AND gate AN ₁ outputs "1". Similarly, if pixel E is the vertical pole, AND gate AN ₂ outputs "1" and pixel E
and gate if is the right diagonal pole
AN ₃ outputs "1", and if pixel E is the left diagonal pole, AND gate AN ₄ outputs "1"
Output. Therefore, the OR gate OR ₉ is the pixel E
When is a pole in either the horizontal direction, the vertical direction, or the left and right diagonal directions, "1" is output, and this output signal is used as the index X to extract the pole. Also at the same time or gate OR ₉
When outputs “1”, peak register PM
The data at the poles can be obtained as is. In this case, in the object recognition device in this example, the digital grayscale image data is shifted in synchronization with the horizontal scanning signal of the TV camera 1, so that it is possible to extract the extreme points in real time.

In this way, the extreme points of the grayscale image are extracted, but when we actually extract the extremes from the digital grayscale image shown in Figure 8, we find that the points surrounded by circles and squares in the figure are the extremes. . In this case, the poles surrounded by circles are extracted as poles in the horizontal or vertical direction, whereas the poles surrounded by squares are not extracted as poles in the horizontal or vertical direction, but in the left or right diagonal direction. This is the first time that it has been extracted as an extreme point. The extreme point portion surrounded by the square mark above corresponds to, for example, the contour branching point of the overlapping objects 7a and 7b, and by extracting the extreme point of this portion, the problem that occurred in the previous image processing method Object 7a,
Discontinuation d at the contour branching point of 7b (see Figure 12)
This allows the contours of overlapping objects to be recognized more accurately.

In the above embodiment, the present invention is applied to the digital grayscale image that has passed through the spatial filter 3, but the present invention may also be applied to the digital grayscale image directly from the TV camera 1. Further, the specific means using the image processing method according to the present invention is not limited to the above embodiment, and it goes without saying that the design may be changed as appropriate.

(7) Effects of the Invention As explained above, according to the image processing method of the present invention, an image obtained from a receiver or the like is divided into a plurality of pixels arranged in a matrix, and each of the plurality of pixels is Among them, pixels with a larger difference in gray level than any two neighboring pixels arranged horizontally, vertically, and left/right diagonally are extracted as extreme points, so that the extraction of extreme points is reliable and it is possible to avoid overlapping. Even when recognizing the outline of objects, etc.
There is no possibility that the outline of the object will be interrupted, and the object can be recognized more accurately.

[Brief explanation of drawings]

Figures 1 a to f are explanatory diagrams showing the basic principle of extracting the poles of a one-dimensional image, Figure 2 is an explanatory diagram showing an example of a window used when extracting the poles of a two-dimensional image, and Figure 3 is an explanatory diagram showing the basic principle of extracting the poles of a two-dimensional image. An explanatory diagram showing an algorithm of the image processing method according to the present invention, FIG. 4 is a schematic diagram showing an example of an object recognition device to which the image processing method according to the present invention is applied, and FIG. 5 is a state of an object photographed by a TV camera. FIG. 6 is a diagram showing an image captured by a TV camera, FIG. 7 is a diagram showing an example of digital image data obtained by digitizing an image captured by a TV camera, and FIG. 8 is a diagram showing an example of digital image data obtained by digitizing an image captured by a TV camera. 9 is a circuit diagram showing a specific example of the data shift circuit shown in FIG. 4; FIG. 10 is an explanatory diagram showing an example of a scanning method on the captured image screen of a TV camera; FIG. 11 is a circuit diagram showing a specific example of the data processing circuit shown in FIG.
FIG. 2 is a reference explanatory diagram showing an image state obtained when polar points are extracted using a conventional image processing method. W...Window, A to I...Pixel, 1...
TV camera, 2...AD converter, 4...data shift circuit, 5...data processing circuit.

Claims (1)

[Claims] 1 An image obtained from a television receiver, etc. is divided into a plurality of pixels arranged in a matrix, and from among the plurality of pixels, any two of the adjacent pixels arranged in the horizontal, vertical, and left/right diagonal directions are An image processing method characterized by extracting pixels with a large difference in gray level as extreme points.