JPH0658687B2 - Edge enhancement device - Google Patents

Description

TECHNICAL FIELD The present invention relates to an apparatus for performing edge enhancement processing on a grayscale image of an object to extract a contour. It should be noted that such an apparatus is suitable for use in a robot vision system that recognizes the shape of an object from the contour of the object and measures the position.

[Conventional technology]

As a method of extracting an edge from a grayscale image, a method using various differential operators (Sobel, Roberts, Robinson operators, etc.) has been conventionally known.

[Problems to be solved by the invention]

However, even with such a method, there is a problem that an edge having a small differential value and noise due to the texture of the background remain. Further, although noise removal and edge extraction are performed in the contour extraction processing and the recognition processing, they are not always sufficient. That is, there are problems such as remaining noise line segments and inability to extract some contour line segments.

Therefore, it is an object of the present invention to provide an edge enhancement apparatus capable of removing background noise and enhancing edge by correctly extracting even a small differential value if the edge is a part of the contour. To do.

[Means for solving problems]

Differentiate the grayscale image obtained by imaging and quantizing the object,
A calculating means for calculating the tilt angle of the edge and its strength, a calculating means for dividing the effective screen into a plurality of small areas, and calculating a histogram of the tilt angle for each small area, and two kinds of dividing the effective screen. With a plurality of middle areas consisting of a predetermined number of small areas, calculating means for calculating the tilt angle histogram for each middle area, and detecting the peak value from the tilt angle histogram of each middle area, Rotating the middle region in the direction corresponding to the inclination angle peak value and the determination means for determining the range,
Calculating means for calculating the projection distribution in the direction orthogonal to the edge, deciding means for deciding the area where the edge emphasizing process is to be executed from the edge projection distribution, and the edges in the vicinity of the emphasizing area are emphasized. And a processing means for making it zero.

[Action]

A grayscale image of the object is input and a differential operation is performed to calculate an edge tilt angle and edge strength. After that, the screen is 2N ×
It is divided into 2N small areas and divided into A and B areas.
A kind of medium area (2 × 2 small area) is divided, and an angle histogram is calculated for each division. Edges having the same inclination angle existing in each middle region are detected from the peak value of the angle histogram. Furthermore, by rotating the middle area by the extracted tilt angle and obtaining the projection distribution in the direction orthogonal to the edge direction within that area, the object edge and the noise edge are separated, and only the object edge is emphasized. To process. Further, by performing this intensity processing for the A division and the B division, even if the edge in the peripheral portion of the middle region is judged to be noise and is not emphasized in one division, it can be emphasized in the other division. To do.

The following is a list of items.

(1) Divide the effective screen into 2N × 2N small areas and 2 × 2
The edge tilt angle histogram is calculated in the small area, that is, in the middle area, and a typical edge tilt angle existing in the middle area is detected.

(2) Rotate the middle area in the direction corresponding to the typical tilt angle in the middle area, calculate the projection distribution in the direction orthogonal to the edge, and the area where the edge of the same tilt angle exists (emphasized area). To decide.

(3) The emphasis processing is performed on the edge in the edge emphasis area obtained for the representative edge of each medium area.

(4) By performing the above processing not only for A division but also for B division, emphasis processing is uniformly performed for all edges.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is an object, 2 is an imaging device such as a television camera,
3 is an analog / digital (A / D) converter, 4 is an image memory, 5 is a differential operation unit, 6 is a differential image memory, 7 is an edge inclination operation unit, 8 is an edge intensity operation unit, 9 is an edge inclination memory, Reference numeral 10 is an edge strength memory, and 11 is an edge emphasis calculation unit.

The image of the object 1 is picked up by the TV camera 2 and its video signal A
The / D converter 3 stores the digital value in the image memory 4. The image memory 4 stores the density value f (x, y) of the object image.
Is stored as, for example, 8-bit data. The coordinates (x, y) of the image are expressed by the X, Y coordinate system as shown in FIG. The density value f (x, y) of the pixel of interest (x, y) is 0 ≦ f (x,
y) ≦ 255. This image A is differentiated by the differential operation section 5, and the result is stored in the image memory 6. There are two directions of differentiation, the X direction and the Y direction, and the first-order derivatives thereof are expressed as Δ _x f and Δ _y f, respectively. Here, the Sobel differential operator as shown in FIG. 3 is used. Expressing the x-direction differential D ₁ shown in FIG. 9 (a) and the y-direction differential D ₂ shown in FIG. 9 (b) by equations, Δ _x f = {f (x + 1, y-1) + 2 · f (X + 1, y) + f (x + 1, y + 1)}-{f (x-1, y-1) + 2 · f (x-1, y) + f (x-1, y + 1) } (1a) Δ _y f = {f (x + 1, y + 1) + 2 · f (x, y + 1) + f (x-1, y + 1)}-{f (x + 1, y-1) + 2 · f (x, y-1) + f (x-1, y-1)} (1b).

The differential value (Δ _x f, Δ _y f) will be described with reference to FIG. As is clear from the figure, the pixel of interest Pe (x, y)
The inclination angle θ of the edge vector V and the intensity w are expressed by the following expressions.

Here, the unit vector (u, of the differential value (Δ _x f, Δ _y f)
From v), θ is obtained using a table of tilt angles. If θ is to be obtained with an accuracy of 1 degree, (u, v)
It is enough to have 360 values of. The unit vector is as follows.

The edge tilt angle calculation unit 7 calculates the tilt angle θ from the unit vector (u, v) using an angle table, and stores the result in the edge tilt angle memory 9. The edge strength calculator 8
The strength w is calculated by the equation (3) and stored in the edge strength memory 10. The edge emphasis calculation unit 11 accesses the edge inclination angle memory 9 and the intensity memory 10 to perform edge emphasis processing.

Now, the effective screen of the object 1 is set to 512 (horizontal) × 512 (vertical) pixels, and in this case, as shown in FIG.
Of the small area R _s . Each small area (i, j) has (1 ≦
i ≦ 2N, 1 ≦ j ≦ 2N). Next, each of the small areas R _s is divided into 2 × 2 middle areas. The middle region divided by the solid line in FIG. 5 is shown in FIG. 6 (a), which will be referred to as A division. The area divided by the broken line is the small area R _S. FIG. 6B shows a small region of 2N × 2N, which is obtained by dividing the (2N−2) × (2N−2) region excluding the outer small region into the middle region. The division shown in FIG. 6B will be referred to as B division. Next, a histogram of the edge inclination angle θ is obtained for each small area. Further, when the θ histogram Hθ for each of the middle areas of the A division and the B division is calculated, as shown in FIG. 7, the peak value (P
θ ₁ , Pθ ₂ etc.) exist. Then, the peak value of the inclination angle histogram is obtained for each middle region. Only one peak is shown as Pθ in FIG. Then, the inclination angle θ _P corresponding to the peak value Pθ of the θ histogram is obtained, and further (θ _PL , θ _PU ) corresponding to the half value Pθ / 2 is obtained to obtain the half value width Wθ. FIG. 9 shows the intensity histogram H _W for the same middle area as the tilt angle histogram. The w histogram in this middle region cannot be distinguished from noise because there is no information on the location of the edge. Therefore, as shown in FIG. 10, the target middle region R _C is rotated by the inclination angle θ _P of the edge, and the projection distribution S _{W of the} edge in the edge direction C _e and the vertical direction Y ′ is obtained. If, from the projection distribution of edge of the theta _P direction, it can be seen a region R _e where edge contours are present. The projection distribution _{S W} of theta _P shown in FIG. 11. The position X _P is found for the peak value P _X of the projection distribution, and the emphasized region [X _PL , X _PU ] can be determined from P _XB determined from P _X. The edge intensity w is set to the value w _s only for the edge in the emphasized area R _e in the rotated middle area R _{R when} the edge inclination angle θ is in the range of [θ _PL , θ _PU ], and otherwise. The edge strength w is set to "0". This completes the edge enhancement of the contour in a specific direction.

The above emphasis processing is shown as a flow chart.
As shown in the figure.

(1) Input the image of the TV camera to the image memory 4 (see).

(2) The differential image of the original image is calculated by the differential operation and input to the differential image memory 6 (see).

(3) The edge tilt angle θ and the strength w are calculated from the differential image and stored in the edge tilt angle memory 9 and the edge strength memory 10 (see,).

(4) The edge slope angle histogram is calculated for each small area (see).

(5) A tilt angle histogram is obtained for each of the A and B medium areas (see).

(6) Detect some peak values of the inclination angle histogram for each middle area (see).

(7) In the middle region rotated corresponding to each peak value θ _P ,
The projection distribution of the intensity w is obtained for the edges within the half width of θ _P (see).

(8) The emphasis region corresponding to the peak value θ _P is determined from the projection distribution (see).

(9) Edges in the emphasized area obtained with respect to the peak value θ _P of the edge inclination angle are emphasized, and the other edge strengths are made zero (see).

With the above processing, noise edges can be removed and the entire surface can be uniformly emphasized.

〔The invention's effect〕

According to the present invention, the effective screen is divided into 2N × 2N small areas, the histogram of the edge inclination angle θ is calculated for each 2 × 2 small area, that is, the medium area, and the histogram is present in the middle area. Since we are trying to detect the direction of the contour,
Weak edge can be detected. Further, since the projected distribution of the edge near θ _P in the rotated middle area is calculated from the detected edge tilt angle θ _P to determine the area where the edge exists, the edge of the noise in the edge in the same direction is determined. Can be removed. Further, since the above is performed for both the middle areas of the A division and the B division, it is possible to uniformly edge-enhance the entire area.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram for explaining a relationship between an object image in an image pickup screen and differential processing, and FIG. 3 is a block diagram showing a differential operator. Fourth
The figure is an explanatory view for explaining the edge strength and its inclination,
FIG. 5 is an explanatory view for explaining an example of a small area dividing method of the effective screen, FIG. 6 is an explanatory view for explaining two kinds of middle area dividing methods, and FIG. 7 is a tilt angle in the middle area. FIG. 8 is a graph showing a histogram, FIG. 8 is an explanatory view for explaining the peak value and half width of the inclination angle histogram, FIG. 9 is a graph showing an intensity histogram, and FIG. 10 is a graph for explaining the intensity distribution in the middle region. An explanatory diagram, FIG. 11 is an explanatory diagram for explaining the intensity projection distribution, and FIG. 12 is a flowchart showing an edge emphasis processing flow. Explanation of symbols 1 ... Object, 2 ... TV camera, 3 ... A / D converter, 4 ... Image memory, 5 ... Differential calculation unit, 6 ... Differential image memory, 7 ... Edge slope calculation unit , 8 ... Edge strength calculation unit, 9 ... Edge tilt memory, 10 ... Edge strength memory, 11 ... Edge enhancement calculation unit.

Claims (1)

[Claims] 1. An arithmetic means for differentiating a grayscale image obtained by picking up and quantizing an object and calculating an inclination angle of an edge and its intensity, and an effective screen is divided into a plurality of small areas, and each small area is divided. Calculating means for calculating the histogram of the tilt angle for each of them, and calculation for dividing the effective screen into a plurality of middle areas consisting of a predetermined number of small areas in two types of division modes, and calculating the tilt angle histogram for each of the middle areas. Means and a determination means for detecting the peak value from the inclination angle histogram of each middle area and determining the range of the inclination angle, and rotating the middle area in the direction corresponding to the inclination angle peak value, in the direction orthogonal to the edge. A calculation means for calculating the projection distribution, a determination means for determining the area where the edge enhancement processing is to be performed based on the projection distribution of the edges, and a processing means for enhancing the edges in the vicinity of the enhancement area and zeroing the other edge strengths. , With An edge enhancing device characterized by being formed.