JP2872393B2 - Edge detection method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an edge detection method, and more particularly, to a method for stabilizing an edge pixel sufficient to grasp an outline of a target even in an image having a gradual change in luminance. The present invention relates to a novel improvement to enable the automatic detection.

[Conventional technology]

FIG. 4 is a block diagram of a conventional edge detection device,
In the figure, (1) is a luminance signal of an original image, (2) is a gradient calculation circuit, (3) is a gradient calculation value from the gradient calculation circuit (2), (4) is a local maximum point detection circuit, and (5) is It is an edge pixel obtained from the maximum point detection circuit (4).

The conventional edge detecting device is configured as described above, and the gradient calculating circuit (2) calculates a luminance gradient from the luminance signal (1) of the original image as shown in, for example, Expression (1).

g = [{L (i, j) −L (i−1, j)} ² + {L (i, j) −L (i, j−1)} ² ] ^1/2 (1) where L (i, j) is the luminance of the pixel (i, j).

In the maximum point detecting circuit (4), a pixel having a maximum luminance gradient is selected as an edge pixel (5).

[Problems to be solved by the invention]

In the conventional edge detection device as described above, since the maximum point of the luminance gradient is detected as an edge pixel, in the case of a pixel having a gradual change in luminance, the luminance gradient has a relatively constant value in the luminance gradient portion, and as a result, There has been a problem that it is difficult to stably detect enough edge pixels to grasp the outline of the target.

The present invention has been made to solve such a problem, and in particular, even in the case of an image having a gradual change in luminance, an edge capable of stably detecting an edge pixel sufficient to grasp a contour of a target. The aim is to obtain a detection method.

[Means for solving the problem]

The edge detection method according to the present invention includes at least a lattice pattern to which a luminance signal of an original image is input and connected to a scanning circuit, a scanning circuit connected to the lattice pattern, and a luminance step detection connected to the lattice pattern. Circuit and
Both end pixel selection circuit connected to the luminance step detection circuit, an edge search path selection circuit connected to the both end pixel selection circuit, a curvature calculation circuit connected to the edge search path selection circuit, and the curvature calculation circuit An edge pixel selection circuit connected to the edge pattern detection circuit detects the luminance of the central pixel located at the center of the lattice pattern and the luminance of peripheral pixels located at the periphery of the lattice pattern from the luminance signal of the original image input to the lattice pattern. When the absolute value of the difference between the luminance of the central pixel and the luminance of the peripheral pixel is equal to or greater than a certain value in the luminance step detection circuit, the peripheral pixel is detected, and both ends of the peripheral pixel are detected in the both ends pixel selection circuit. A pixel is selected, and the edge search path selection circuit is set so that the distance from a straight line orthogonal to a line connecting the centers of the pixels at both ends is short. A pixel on the edge search path is selected, the curvature calculation circuit calculates the curvature of the luminance signal for each pixel on the edge search path, and the edge pixel selection circuit calculates a change in the curvature for each pixel on the edge search path. An edge pixel is selected from the pixels on the edge search path based on the edge search path, and the scanning circuit scans the grid pattern.

(Operation)

In the edge detection method according to the present invention, the grid pattern and the luminance step detection circuit find a discontinuity in luminance, and the both ends pixel selection circuit and the edge search path selection circuit determine the edge based on the positional relationship between the lattice pattern and the discontinuity in luminance. A search path is selected, and a curvature calculation circuit and an edge selection circuit select an edge pixel based on a change in curvature of the luminance on the edge search path.

〔Example〕

Hereinafter, preferred embodiments of the edge detection method according to the present invention will be described in detail with reference to the drawings.

FIGS. 1 to 3 show an edge detection method according to the present invention. FIG. 1 is a block diagram, FIG. 2 is a lattice pattern diagram, and FIG. 3 is a conceptual diagram showing luminance and curvature changes. .

In the figure, (1) is exactly the same as the conventional device, and is a luminance signal of an original image. (7) is a scanning circuit,
(8) is a scanning signal from the scanning circuit (7), (9) is a grid pattern to which the luminance signal (1) is input, and (10) is a center pixel and a peripheral pixel obtained by the grid pattern and those pixels. (11) is a luminance step detection circuit, (12) is a subset of peripheral pixels obtained from the luminance step detection circuit (11), (13) is a pixel selection circuit for both ends, (1)
4) Both end pixels obtained from the both end pixel selection circuit (13),
(15) is an edge search path selection circuit, (16) is an edge search path obtained from the edge search path selection circuit (15), (16)
a) is a smoothing circuit, (16b) is a smoothed luminance signal of a pixel on an edge search path, (17) is a curvature calculation circuit, (18)
Is the smoothed luminance signal and curvature at each pixel position on the edge search path, (19) is the edge pixel selection circuit, (20)
Are edge pixels obtained from the edge pixel selection circuit (19).

FIG. 2 is a diagram showing a lattice pattern in one embodiment of the present invention, wherein (21) is a central pixel detection window or a central pixel detected by the central pixel detection window, and (22) is a peripheral pixel detection window or peripheral pixel. Neighboring pixels detected by the detection window, (23) and (24) are both end pixels, (25) is an object,
(26) is a boundary between the object and the background or another object, (27) is a line segment connecting the centers of the both end pixels (23, 24), (28) is an edge search direction, (29) and (30) is the axis of the rectangular coordinate system defined for the grid pattern.

FIG. 3 is a conceptual diagram showing a pixel on an edge search path and a smoothed change in brightness and curvature of the pixel in one embodiment of the present invention, and (31) is an axis showing a pixel position t on the edge search path. Indicates that the variable t increases toward the right. (32) is the axis representing the smoothed luminance, (33)
Is the axis representing the curvature change, (34) is the smoothed brightness corresponding to the pixel position, (35) is the curvature change corresponding to the pixel position,
(36) is the start position of the edge search path, (37) is a smoothed luminance of minimum value + (maximum value−minimum value) × 0.8 or more when searching in the increasing direction of t, and the curvature increases from a decrease. This is the first pixel position that turns to.

Next, a method of actually performing edge detection in the edge detection device configured as described above will be described. First, the luminance of the center pixel (21) and the peripheral pixels (22) is detected by the grid pattern (9). If the absolute value of the difference between the luminance of the central pixel (21) and the luminance of the peripheral pixel (22) is equal to or greater than a certain value D, that pixel is selected. The subset of the peripheral pixels thus selected is a pixel between the two pixels (23) and (24) in FIG. 2, and as shown in FIG.
If 5) is located, both end pixels (23) and (24)
And five pixels in between. The both end pixel selection circuit (13) detects both end pixels (23) and (24) from a subset of the peripheral pixels. The edge search route selection circuit (15)
Selects an edge search path based on the positions of the both-end pixels (23) and (24) described above. Each pixel on the edge-search path is searched for an edge search orthogonal to a line (27) connecting the centers of both end pixels. It is selected to follow the direction (28). Next, the smoothing circuit (16a) removes a high-frequency noise component from a luminance signal of a pixel on the edge search path and outputs smoothed luminance. In the curvature calculation circuit (17), the curvature is settled from the smoothed luminance of each pixel on the edge search path, which is performed according to a method described later. Further, the edge pixel selection circuit (19) sequentially examines the smoothed luminance of each pixel on the edge search path, and finds a pixel in a range having a value of minimum value + (maximum value−minimum value) × 0.8 or more. , The first pixel whose curvature changes from decreasing to increasing is detected as an edge pixel. The scanning of the grid pattern shown in FIG. 2 is performed by the scanning circuit (7). By scanning the grid pattern for the entire screen and repeating the above procedure, edge detection of the entire screen can be performed.

Here, an edge search path determination method will be described. Now, the coordinate series of the edge path starting from each pixel included in the subset of the peripheral pixels is set as {μ (t), ν (t)}, and is called an edge search path. The centers of both pixels are represented by (μ ₁ , ν ₁ ) and (μ ₂ , ν ₂ ).

The edge search path includes two points (μ ₁ , ν ₁ ) and (μ ₂ , ν ₂ )
Are determined along a straight line (28) orthogonal to a line segment (27) connecting A line segment (2) connecting two points (μ ₁ , ν ₁ ) and (μ ₂ , ν ₂ )
The intersection of 7) and the straight line (28) is hereinafter referred to as intersection H. When the center pixel is on the low signal level side (for example, the background side) as in the case of FIG. 2, the search direction on the edge search path is parallel to the direction from the center pixel to the intersection H, and the center pixel is high. When it is on the signal level side (for example, on the target side), it is set in parallel with the direction from the intersection H toward the center pixel.
Next, a method of determining the edge search direction will be described, and an edge search path calculation formula will be described.

<Edge Searching Direction> Assuming that the inclination of a perpendicular drawn to a line segment (27) connecting two points (μ ₁ , ν ₁ ) and (μ ₂ , ν ₂ ) from the center pixel is m, m = −Δμ / Δν (1a ) Given by

At this time, the quadrant where the intersection H exists with respect to the (μ, ν) coordinate system is Given by Here, the definition of the quadrant is μ> 0 and ν> 0
Is the first quadrant, and the second, third, and fourth quadrants are shown clockwise.

Next, conditions for uniquely determining the quadrant where the intersection point H exists are given as follows.

 First, the following quantities are defined.

(1) If m ≧ 0, the intersection H is in the first quadrant if μ ⁺ + ν ⁺ ≧ 0 (5), and if μ ⁺ + ν ⁺ <0 (6), the intersection H is in the third quadrant. It is in.

(2) When m <0 (μ ⁺ > 0) {(μ ⁺ = 0)} (ν ⁺ <0)}
If (7), the intersection point H is in the second quadrant, and (μ ⁺ <0) {(μ ⁺ = 0)} (ν ⁺ ≧ 0)}
(8) Then, the intersection point H is in the fourth quadrant.

A flag that gives the edge search direction is determined as follows based on the condition that the signal level of the quadrant where the intersection H exists and the center pixel is on the high signal level side or on the low signal level side.

When intersection H is in the first quadrant When in the second quadrant When in the third quadrant When in the fourth quadrant Here, ΔLμ, ν = Lμ, ν-L0,0 (15), where Lμ, ν is the luminance of the peripheral pixels detected by the luminance step detection circuit (11), and L0,0 is the luminance of the central pixel. It is.

<Edge Search Path> First, a function given by Expression (16) is introduced.

Also, referring to equation (2), A = max (| Δμ |, | Δν |) / min (| Δμ |, | Δν |) (17) B = | Δμ | − | Δν | (18) deep.

Next, each pixel included in the subset of peripheral pixels which is the output of the luminance step detection circuit (11) is used as a starting point ｛μ (1)
ν (1))}, each search path {μ (t), ν
(T)) is given by the following equation.

(1) When B ≧ 0 Here, t = 1, 2,..., T are sufficiently large positive integers.

(2) When B <0 Now, the pixel sequence ｛μ (t), ν determined as described above
(T))｝ (t = 1,2,... T) to obtain pixels t = 1,2,.
Calculate the curvature κ (t) and the curvature change Δκ (t) at T as follows.

κ (t) = L ₁ (t) / [1+ ｛L ₁ (t)｝ ² ] ^3/2 (25) Δκ (t) = κ (t + 1) + κ (t) (26) where L ₁ ( t) and L ₁ (t) are the primary difference and 2 respectively of the smoothed luminance signal L ₁ (t) at the pixel t.
The next difference is shown.

L ₁ (t) = L ₁ (t) −L ₁ (t−1) (27) L ₁ (t) = L ₁ (t) −L ₁ (t−1) (28) Incidentally, FIG. The smoothed luminance (34) curve shape shown corresponds to a case where a high-frequency noise component is removed from a general step edge.

〔The invention's effect〕

As described above, the present invention uses a grid pattern and a brightness step detection circuit to find a luminance discontinuity point, and based on the positional relationship between the grid pattern and the luminance discontinuity point, a pixel selection circuit at both ends and an edge search path selection circuit. Is used to select an edge search path, and a curvature calculation circuit and an edge pixel selection circuit are used to select an edge pixel based on a change in curvature on the edge search path of luminance.
Even when the luminance changes slowly, there is an effect that it is possible to stably detect enough edge pixels to grasp the outline of the target.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of an edge detecting apparatus to which the edge detecting method according to the present invention is applied, FIG. 2 is a diagram showing a grid pattern in one embodiment of the present invention, and FIG. FIG. 4 is a conceptual diagram showing pixels on an edge search path and smoothed luminance and curvature changes of the pixels in the embodiment. FIG. 4 is a block diagram of a conventional edge detection device. In the figure, (9) is a grid pattern, (7) is a scanning circuit, (11) is a luminance step detection circuit, (13) is a pixel selection circuit at both ends, (15) is an edge search path selection circuit, and (17) is curvature. The calculation circuit, (19) is an edge pixel selection circuit. In the drawings, the same reference numerals indicate the same or corresponding parts.

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int. Cl. ⁶ , DB name) G06T 7/00 JOIS

Claims (1)

(57) [Claims] 1. A grid pattern to which at least a brightness signal of an original image is input and connected to a scanning circuit, a brightness step detection circuit connected to the grid pattern, and both-end pixel selection connected to the brightness step detection circuit A circuit, an edge search path selection circuit connected to the both ends pixel selection circuit, a curvature calculation circuit connected to the edge search path selection circuit, and an edge including an edge pixel selection circuit connected to the curvature calculation circuit In the detection device, the luminance of the central pixel located at the center of the lattice pattern and the luminance of peripheral pixels located at the periphery of the lattice pattern are detected from the luminance signal of the original image input to the lattice pattern, and the luminance step detection circuit detects the luminance of the central pixel. When the absolute value of the difference between the luminance of the peripheral pixel and the luminance of the peripheral pixel is equal to or greater than a certain value, the peripheral pixel is detected, and the pixel at both ends is selected. In the edge search path selection circuit, a pixel on an edge search path is selected such that a distance from a straight line orthogonal to a line segment connecting the centers of the both ends pixels is short in the edge search path selection circuit.
The curvature calculation circuit calculates the curvature of the luminance signal for each pixel on the edge search path, and the edge pixel selection circuit calculates a curvature of the pixel on the edge search path based on a change in the curvature for each pixel on the edge search path. An edge detection method, wherein an edge pixel is selected, and the grid pattern is scanned by the scanning circuit.