JPH0740304B2 - Image feature extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial field of application] The present invention is an image in which a feature amount is extracted from an image in the visual recognition field of recognizing, inspecting, measuring, etc. a target object based on the feature amount obtained from image information. The present invention relates to a feature extraction device, and more particularly to an image feature extraction device suitable for calculating a moment feature of an image read by raster scanning.

[Conventional technology]

In the conventional visual recognition field, there is an important moment feature as a shape feature of a target object in the feature amount obtained from image information. This moment feature is described in, for example, Yuji Kiuchi, "The Story of Image Recognition" pp.50-51, published by Nikkan Kogyo Shimbun, and "Introduction to Computer Image Processing", edited by Hideyuki Tamura, pp.86, published by Soken Publishing. There is. Although the expression method and the name of this moment feature are slightly different depending on the literature, the moment feature M of the (p + q) th order of the image (x, y)
_pq is defined by the following equation.

However, p, q = 0,1,2, ...

FIG. 11 illustrates a block diagram of a conventional image feature extraction device that calculates a general moment feature. In FIG. 11, 10
Is a power calculator, 20 is a multiplier, 100A is a coordinate calculator, 20
0A is an accumulator. As the conventional apparatus main components, x, type y coordinates x ^p, exponentiation unit 10, and x ^p · y ^q consisting multiplier 20 to obtain the coordinates calculator 100A obtains the y ^q
And an accumulator 200A that calculates the sum of x ^p and y ^q when (x, y) = 1.

As described above, the conventional image feature extraction apparatus for computing the moment feature requires an exponentiation calculator and a multiplier that require a large amount of hardware and a long processing time.

[Problems to be solved by the invention]

In the above-mentioned conventional technique, it is difficult to realize an image feature extraction device with compact and high-speed hardware, because a hardware amount is huge to calculate the moment feature and a multiplier or the like which requires a great deal of processing time is required. There was a problem.

An object of the present invention is to provide a compact image feature extraction apparatus capable of computing a moment feature with high-speed hardware, which does not require a power calculator and a multiplier which require a huge amount of hardware and require a great deal of processing time.

[Means for solving problems]

The above-mentioned object is, firstly, that the following three secondary moments are practically important in the moment feature of equation (1): Secondly, focusing on the fact that the target image (x, y) is generally of the raster scan type, the following relational expression holds then: x _i = x _i-1 +1 (6) y _j = y _j-1 +1 (7) Therefore, for x ² , y ² and xy, the following arithmetic expression is given, and a _i = x _i ² = a _i-1 + 2x _i-1 +1 (8) b _j = y _j ² = B _j-1 + 2y _j-1 +1 (9) c _ij = x _i y _j = c _i-1 , _j + y _j (10) These (8), (9), (10) formula It calculates the x ^2, y ^2, xy used, and (3), (4), (5) using the accumulator by equation (x, y) = x ² for 1, y ^2, xy
This is achieved by the image feature extraction device that is adapted to extract the cumulative moments of the second moments M ₂₀ , M ₀₂ , and M ₁₁ .

[Action]

In the image feature extraction device, the x, y coordinate input or the like is selected based on the x, y coordinate data of the image (x, y) read by raster scanning, and the adder is used (8), (9), Based on equation (10), x ² , y ² , xy, etc. are obtained, and by using an accumulator, Σx ² , Σy ² for (x, y) = 1 based on equations (3), (4) and (5). Since the second moments M ₂₀ , M ₀₂ , and M ₁₁ can be extracted by obtaining Σxy, the hardware can be realized with a simple configuration of an adder and the like without the need for a power calculation and a multiplier.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS. 1 to 10.

First, FIG. 7 to FIG. 9 show an embodiment of an image feature extracting apparatus according to the present invention.
x ^2, y ^3, an illustration of xy arithmetic principles. 7 to 9
A method for performing the calculation of x ² , y ² and xy, which are the basic terms of the moment feature of the raster scan type image (x, y), with a simple configuration such as an adder will be described with reference to the drawings.

First, up to the second order, the moment feature represented by the above equation (1) is often used in practice. That is, p + q2 (2) Of these, the moment feature when p and q are the following values is particularly important.

(A) Moment feature when p = 2 and q = 0 M ₂₀ (second moment around the y-axis), (B) Moment feature when p = 0, q = 2 M ₀₂ (second moment about x axis), (C) Moment feature when p = 1 and q = 1 M ₁₁ (conjugate moment around the origin), Secondly, assuming that the target image (x, y) is of the raster scanning type, the x, y coordinates at that time necessarily satisfy the following equation.

x _i = x _i-1 +1 (6) y _j = y _j-1 +1 (7) Therefore, x ² , y ² , xy in equations (3), (4) and (5)
Can be transformed as follows using equations (6) and (7).

For (a) x ² , according to the equation (6), a _i = x _i ² = (x _i-1 +1) ² = x ² _i-1 + 2x _i-1 +1 = a _i-1 + 2x _i-1 +1 ( 8) However, i = 0 to n, n is the maximum value of the pixels in the x direction of the image. FIG. 7 shows a _{0 to} a _n based on the equation (8).

(B) For y ² , b _j = y _j ² = (y _j-1 +1) ² = y ² _j-1 + 2y _j-1 +1 = b ² _j-1 + 2y ² _{j-1 according to the} equation (7). +1 (9) where i = 0 to m, m is the maximum value of pixels in the y direction of the image. B _{0 to} b _m based on the equation (9) are shown in FIG.

For (c) xy, according to equation (6), c _xy = x _i y _j = (x _i-1 +1) y _j = x _i-1 y _j + y _j = c _i-1 , _j + y _j (10) FIG. 9 shows c _{00 to} c _nm based on the equation (10).

By the above (8), (9) and (10), x ² , y ² , xy which are the basic terms of the second-order moment feature M ₂₀ , M ₀₂ , M ₁₁ of the raster scan type image (x, y) Since there are no terms for exponentiation and multiplication in the arithmetic expression of, it is possible to obtain it by a simple operation of only addition. Note that the second term of the equations (8) and (9) can be doubled by a 1-bit shift, so that the calculation is unnecessary.

10 (a) and 10 (b) show second-order moment features M ₂₀ , M ₀₂ ,
It is an explanatory view of the example of a function of M ₁₁ . In FIGS. 10 (a) and 10 (b), when the length L and width W of an inclined object such as the spanner S as shown in FIG. First, the tilt angle θ of the object S is obtained, and secondly, by rotating by the tilt angle θ of the tilted object S using the affine transformation method, normalization is performed as shown in FIG. 10 (b). Thirdly, the projections S ₁ and S ₂ of the object S in FIG. 10 (a) in the x and y directions are taken, and the projection width at that time is obtained, that is, the projection width in the y direction is the spanner S.
Is L, and the projected width in the x direction is the width W of the spanner S. For such a tilted object S, the tilt angle θ of the object
Is required to be detected, and the inclination angle θ of this object can be obtained by solving the following equation using the second-order moment features M ₂₀ , M ₀₂ , and M ₁₁ .

In such a case, the moment features M ₂₀ , M ₀₂ , M ₁₁ are important.

Next, FIG. 1 shows the first embodiment of the image feature extraction apparatus according to the present invention, and the second moment M ₂₀ about the y-axis of the equation (3).
It is a block diagram of the image feature extraction apparatus which extracts. In FIG. 1, 30 is a shifter, 40 is an adder, 50 is a register, 60 is an adder, 70 is a register, 80 is an AND gate, 100 is a coordinate calculator, and 200 is an accumulator. This device is the x of the image (x, y)
Cumulative calculator 100 that calculates the basic term x ² of the moment feature M ₂₀ from the coordinates and the sum of x ² when (x, y) = 1 to obtain the final secondary moment M ₂₀ The coordinate calculator 100 for calculating x ² receives the x coordinate of the image (x, y) output corresponding to the raster scan and shifts it by 1 bit to the left to create 2x. , An adder that adds the created 2x and the calculation result x ² and 1 one pixel before
40 and a register 50 that stores the result x ^{2 of the} previous pixel
And the accumulator 200 has x ² and 1 when (x, y) = 1
An adder 60 for adding the accumulation result? X ² for the preceding pixel, a register 70 which stores one pixel before the accumulated result? X ^2, consists AND gate 80. To create a storage clock, 300 is a system bus is there.

FIG. 2 is an operation explanatory diagram showing operation timings of signals of respective parts in FIG. The operation of FIG. 1 will be described with reference to FIG.
Now, when an image (x, y) in which 4 pixels out of 6 pixels in the horizontal direction by raster scanning as shown in FIG. 2 are effective, only the effective pixel area of the image (x, y) is “ 1 "line enable signal a and clock CLK signal b
Based on the above, the x coordinate signal c is input to the coordinate calculator 100 of FIG. 1, doubled by 1 bit left shift in the shifter 30, and then 2x is input to the A input of the adder 40. On the other hand, the B input of one of the adders 40 receives x ² of the operation result of one pixel before stored in the register 950, and the carry C input always receives 1; Y = A + B + C corresponding to is calculated. The calculation result Y is stored in the register 50 operated by the clock CLK signal b, and at the same time, it is sent to the accumulator 200 as the x ² signal e output from the coordinate calculator 100. At this time, the register 50 stores the image (x,
The line enable signal a, which is “1” only in the effective pixel area of y), is always cleared line by line by the line clear LN1 signal d which is delayed by half a pixel by the CLK signal b.

In the next accumulator 200, x ² transmitted from the coordinate calculator 100
The signal e is input to the A input of the adder 60, and the accumulated result up to one pixel before stored in the register 60 is input to one of the B inputs.
x ² is input, Y = A + B corresponding to the equation (3) is calculated by these inputs, and the calculation result Y is AND gate 80
Register 70 operated by the stored clock signal h of the output of
To be stored as a Σx ² signal g. At this time, the storage condition is controlled by the storage clock signal h output from the AND gate 80. That is, CLK of three signals input to the AND gate 80
The signal b is "1" and the line clear LN1 signal d is "1".
And when the image '(x, y) signal obtained by delaying the image (x, y) by half a pixel by the CLK signal b is "1", the stored clock signal h becomes "1" and the value is sent to the register 70. Calculation result Σ x ²
Is allowed to be stored. Such an operation is executed for one plane of the image (x, y) until the end of the raster scanning, and finally the secondary moment M ₂₀ about the y axis in the equation (3) is stored in the register 70. The register 70 is initialized to zero by the initial signal i only once at the start of raster scanning. The second moment M ₂₀ thus obtained can be read from a CPU (not shown) via the system bus 300 by the register 70 of the accumulator 200.

According to this embodiment, generation of x ² which is a basic term necessary for calculating the second moment M ₂₀ about the y-axis can be realized by a simple adder configuration without a multiplier, and the second As shown in the figure, since the calculation of each pixel is completed within one machine cycle, the processing speed becomes high.

FIG. 3 is a block diagram of an image feature extracting device for extracting a second moment M ₀₂ about the x axis in the equation (4) showing a second embodiment of the image feature extracting device according to the present invention. Further, FIG. 4 is an operation explanatory diagram showing operation timings of signals of respective parts in FIG. In FIGS. 3 and 4, the same reference numerals or symbols as in FIGS. 1 and 2 indicate the same or corresponding portions.

1 and 2 in FIGS. 3 and 4, the y-coordinate signal j is input to the shifter 30 of the coordinate calculator 100 instead of the x-coordinate of the image (x, y). Secondly, the operation clock signal of the register 50 of the coordinate calculator 100 is the line enable signal a, and the clear signal of the register 50 is also the initial signal i. Only by changing the coordinate input signal and the control signal, the y ² signal k of the calculation result y ² corresponding to the equation (9) is generated from the output of the register 50 of the coordinate calculator 100 having the same configuration, and the accumulator 200 In the register 70 of x, x as the final accumulation result of the Σy ² signal l of the accumulation result Σy ² corresponding to the equation (4)
The second moment M ₀₂ about the axis is stored, and the system bus 3
It can be read via 00.

According to this embodiment, as in FIG. 1, a multiplier and the like are not required to generate y ² which is a basic term necessary for calculating the second moment M ₀₂ about the x-axis, and a compact and high-speed hardware is provided. In addition to being realized by software, it has the same architecture as that shown in FIG. 1, so hardware can be shared and further compactification is possible.

FIG. 5 is a block diagram of an image feature extraction device for extracting the conjugate moment M ₁₁ about the origin of the equation (5) showing the third embodiment of the image feature extraction device according to the present invention. Further, FIG. 6 is an operation explanatory diagram showing operation timings of signals of respective parts in FIG. 5 and 6, the same reference numerals or signals as in FIGS. 1 to 4 indicate the same or corresponding portions.

The difference from FIGS. 1 to 4 in FIGS. 5 and 6 is that the shifter 30 is not required in the coordinate calculator 100 for inputting the y coordinate signal j of the image (x, y). Secondly, zero is connected to the carry C input of the adder 40 of the coordinate calculator 100, and thirdly, the clear signal m of the register 50 is the sixth signal.
As shown in the figure, it becomes "0" for each of the first line and the next line, and is cleared at this time. Only by changing some of these configurations and control signals, the xy signal n of the operation result xy corresponding to the expression (10) is generated from the output of the register 50 of the coordinate calculator 100, and the register 70 of the accumulator 200 is generated. Σxy signal Q of cumulative result Σxy corresponding to equation (5)
Conjugate moment around the origin as the final cumulative result of
M ₁₁ is stored and can be read out via the system bus 300.

According to this embodiment, as in FIGS. 1 and 3, the invention of xy, which is a basic term required to calculate the conjugate moment M ₁₁ about the origin, does not require a multiplier or the like and is compact and high-speed. It can be realized by hardware.

In the embodiment shown in FIGS. 1 to 6, the coordinate calculator is used.
100 x, y coordinate input and adder 40 input data 2x, 2y, y
Similarly, only the control clock and the clear signal of the register 50 are made different. Therefore, by making it possible to change these programmableally, the hardware can be shared and the secondary moment M can be commonly used with a small amount of hardware.
It is also possible to perform calculations of ₂₀ , M ₀₂ , and M ₁₁ .

Further, by providing the above-mentioned hardware for calculating the second moments M ₂₀ , M ₀₂ , M ₁₁ respectively, the three second moments M ₂₀ , M ₀₂ , M ₁₁ can be extracted at the same time, and the recognition processing can be speeded up. It goes without saying that it can be realized.

In such a case, one of four coordinate values of x, y coordinates and 2x, 2y coordinates can be arbitrarily selected by the x, y coordinate data of the image (x, y) read by raster scanning. A multiplexer that does not exist can be provided, and the coordinate value selected by this multiplexer can be connected to the A input of the adder 40 of the coordinate calculator 100.

Further, by connecting the output of the register 50 of the coordinate calculator 100 to the image bus, it is possible to directly transfer to an image memory (not shown).

〔The invention's effect〕

According to the present invention, the first extraction of the moment features, particularly the second-order moment features M ₂₀ , M ₀₂ , M ₁₁ of the raster scan type image, which are important positions of the shape features of the object in visual recognition, is performed.
It can be realized by compact hardware with a simple configuration such as an adder that does not require a multiplier. Second, it can perform high-speed calculation in real time at the same speed as raster scanning (TV scanning). Second moment feature M ₂₀ , M
₀₂ , M ₁₁ can be extracted with a common architecture to facilitate the design and achieve the effect.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of an image feature extraction device according to the present invention, FIG. 2 is an operation explanatory diagram of FIG. 1, and FIG. 3 is a block diagram showing a second embodiment of the present invention. , Fig. 4 is the third
FIG. 5 is an operation explanatory diagram, FIG. 5 is a configuration diagram showing a third embodiment according to the present invention, FIG. 6 is an operation explanatory diagram of FIG. 5, and FIGS. 7, 8, and 9 are according to the present invention. Each showing an example
x ^2, y ^2, xy operational principle illustration, FIG. 10 (a), (b) is 2
FIG. ₁₁ is an explanatory diagram of a function example of the next moment features M ₂₀ , M ₀₂ , and M ₁₁ , and FIG. 11 is a configuration diagram illustrating a conventional image feature extraction device. 40 …… adder, 50 …… register, 60 …… adder, 70 ……
Register, 100 ... coordinate calculator, 200 ... accumulator.

Claims (2)

[Claims] 1. An image f (x, y) read by raster scanning
Connect the selected coordinate value to one input and the means such as a multiplexer that can arbitrarily select one of the four coordinate values of x, y coordinate and 2x, 2y coordinate by the x, y coordinate data of An adder for performing an addition operation by connecting the output of the register storing the addition operation result for the pixel at the position one before the pixel to the other input and selecting 1 or 0 for the carry input to perform the addition operation and the addition A coordinate calculator including the above-mentioned register for storing the calculation result, and the output value of the above-mentioned register are input, and cumulative calculation is performed in the processing object image area of the above-mentioned image f (x, y) to obtain the image f (x, y). An image feature extraction device comprising: an accumulator that extracts moment features. 2. The image feature extraction apparatus according to claim 1, wherein the output of the register of the coordinate calculator is connected to an image bus and can be directly transferred to an image memory.