JPH0461393B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to an image signal processing device for processing image signals obtained from an image object.

[Technical background of the invention and its problems]

In the field of pattern recognition, image processing, image measurement, etc., image signal processing devices that convert image signals obtained from image objects into electrical signals and process these image signals with electronic computers are widely used. . The image object includes visible patterns and invisible patterns, and an image signal is extracted from the image object by a method such as acoustoelectric conversion using ultrasonic waves or photoelectric conversion using visible light or infrared rays.

For example, a method using photoelectric conversion using visible light, as used in television cameras and video cameras, involves irradiating light onto an object to be imaged and emitting reflected light according to the reflectance distribution on the object's surface. It is input as an image signal and converted into an electrical signal by a photoelectric conversion element. This image signal is then processed and reproduced as an image on a television receiver screen or the like.

Further, this image signal is used as basic data for measuring parameters such as the size, length, shape, positional relationship, texture, etc. of an object in the measurement field. Regarding this, an example will be described in which the distance to the subject is measured using a camera.

When capturing an image of an object with a camera, the distance between the eye and the lens is adjusted by the focusing action of the human eye. Once the distance between the focused eye and the lens is determined, the distance between the object and the eye can be measured using optical geometry calculations. At this time, since the approximate value of the distance between the object and the eye is known, the approximate distance between the eye and the lens that can be focused can be determined and the image can be focused and sharpened. Additionally, you can determine the exact distance between your eyes and objects. This means that a parameter called distance can be measured by adjusting the sharpness of the image.

Conventional automatic focusing devices perform predetermined calculations on each image signal extracted from an image of an object under each parameter value, and determine whether the relative deviation between the two calculation values is maximum or minimum. This method automatically adjusts parameter values and performs focus matching.

The automatic focusing of the camera will be explained in detail below. In automatic focusing of a camera, the focus position p, which is the distance between the object and the lens, is used as a parameter, and when the image corresponding to the focus position p is G(p), the following image sharpness Z(p) is defined.

Z(p)= Σ ^x,y (∂/∂xG(p)(x,y)) ^2The above formula integrates the square of each pixel value of the image that is first differentiated in the x direction for all x and y pixels. (integration).

When performing automatic focusing of the camera, image sharpness Z(p1) and Z(p2) are calculated for the two extracted focus positions p1 and p2, respectively. Z(p1), Z corresponding to these two parameter values p1, p2
(p2), and move the focus position p so as to maximize Z(p). In this way, the true focus position pt
Detect.

In this way, conventional devices automatically move the lens mechanically and change the aperture of the camera to align the focus between the object and the lens, and then compare the degree of focus alignment of multiple captured images. While comparing the images, images with proper focus alignment were obtained. However, a certain parameter value obtained first (for example, the distance D between the object and the lens mentioned earlier)
Based on the image signal at It has not been possible to convert the signal into an image signal for appropriate parameter values, obtain a focused image signal, and automatically perform focus alignment. For this reason,
In order to obtain a sufficiently focused image of one object, it is necessary to change the parameter values and obtain images under each parameter value, which has the disadvantage that a large amount of time is spent on focusing. .

[Purpose of the invention]

The present invention reproduces an image in which focus alignment has been properly performed by numerically changing parameters that are conditions for image signal acquisition from an image signal acquired in one measurement from an image target. The present invention provides an image signal processing device that can convert images into signals.

[Summary of the invention]

The present invention converts an image signal obtained from an image target in one measurement into an original image signal of a quantized discrete electric signal and stores it, and converts this original image signal and the image signal to the image target. The original image signal is converted into a reproduced image signal by correlating it with a point spread function to which the parameters that are the conditions for acquisition are applied, the reproduced image signal is normalized, and the entropy of the information amount is calculated in the normalized signal. Find the image entropy by applying the concept of , change the parameter values that are the conditions for acquiring the image signal so that this image entropy is minimized, and convert the image signal into a reproduced image signal that applies the changed parameters. It is something.

[Embodiments of the invention]

Embodiments based on the present invention will be described with reference to the drawings. In this embodiment, a method will be described in which a reflected wave of ultra-temperature waves emitted to an object is introduced as an image signal. The image signal input method is an application of the concept of the synthetic aperture method used in the field of radio communications.

In this example, as shown in FIG. 2, the object 1 that is rotating at a constant speed in a circular orbit is the object to be imaged, and the image to be reproduced (reproduced image) is the image of the object 1 in a stationary state. It is a statue. This object 1 is installed at the tip of a vertical rod 3 connected to the circumference of a turntable 2, and has an unknown rotational speed x.
By rotating the turntable 2 at a constant speed of , the object 1 rotates in a circular orbit in space.

5 is an ultrasonic oscillator, which emits ultrasonic waves to the object 1 rotating in a circular orbit. Reference numeral 11 denotes an image signal input section, which is provided close to the ultrasonic oscillator 5, receives reflected waves of ultrasonic waves from the object 1 to be imaged, and converts them into image signals.

The ultrasonic oscillator 5 and the image signal input section 11 are integrally connected to a scale 6, which is an axis perpendicular to the plane of rotation of the object 1, every time the object 1 rotates once.
is moved downward by a distance D corresponding to one divided into n parts.

Also, when the object 1 moves on an arc on the side of the ultrasonic oscillator 5 that has two points of contact at both ends when a tangent is drawn from the ultrasonic source of the ultrasonic oscillator 5 to the circle drawn when the object 1 rotates, Ultrasonic waves are emitted at a radiation range angle formed by the two contacts and the ultrasound source for a predetermined time, and data collection is performed.

FIG. 1 is a block diagram showing the overall configuration of this embodiment. In the figure, an input circuit section 10 is composed of an image signal input section 11, an A/D converter 12, and an image memory 13.

The image signal input unit 11 performs a cross-correlation calculation between the received reflected wave and the ultrasound transmitted by the ultrasound oscillator 5 to obtain phase amplitude information data. This is output as an image signal. This image signal is a spatial frequency component that has been Fourier transformed, and the image directly targeted by this image signal is a hologram (diffraction image distribution image) of the object 1, not a reproduced image of the object 1.

The A/D converter 12 converts the image signal output from the image signal input section 11 into a quantized discrete signal. The image memory 13 stores the original image signal D(i,j) obtained by conversion by the A/D converter 12 in an original image signal memory area 13a. Note that the image memory 13 also has a reproduced image signal memory area 13b.

Reference numeral 14 denotes an image reproducing unit, which converts the original image signal into a reproduced image signal for reproducing an image.

Here, the principle of image conversion and reproduction will be explained.

The original image signal D (i, j) is obtained by measurement and quantized from the image object, and the reproduced image signal G ₁ (i, j) is the image to be reproduced representing the image of the object.
j), the following relationship generally holds between the two through a transfer function k(i, j) involved in the conversion.

D(i,j)=k(i,j)*G ₁ (i,j) Here, * indicates a convolution operation. The reproduction of the reproduced image signal G ₁ (i, j) is performed by the following arithmetic processing.

If the Fourier transform is represented by F and the inverse Fourier transform is represented by F ^-1 , the following formula can be derived from the above formula.

G ₁ (i,j) =F ^-1 [F(D(i,j)/F(k(i,j))] Here, / indicates a deconvolution operation, and this deconvolution and is the original image signal D(i,j) and the point spread function k in the frequency domain
This is a cross-correlation calculation of (i, j).

In this way, the reproduced image signal G ₁ (i, j) is obtained.

The transfer function k(i,j) is defined by the characteristics of the measurement system, and includes a parameter x when acquiring an image signal from an image object. Therefore,
In this example, the transfer function k (i, j) is a point spread function k including an unknown parameter x.
Take (i, j, k). Then, from the relationship in the above equation, the unknown parameter x is applied to the point spread function k (i, j, k), and based on the original image signal D (i, j, k) obtained by measurement, The reproduced image signal G ₁ (i, j, k) can be reproduced. Note that in this embodiment, this original image signal is acquired as a spatial frequency component that has already been Fourier transformed.

The image reproduction section 14 performs the following processing.

Original image signal memory area 13a of image memory 13
The original image signal D(i,j) stored in is extracted. The point spread function k created by applying the parameter value x is Fourier transformed. This Fourier-transformed point spread function k and the original image signal D are deconvoluted to obtain G ₁ (i, j, k). This reproduced image is an image based on the reflectance distribution of the object 1 in a stationary state. And this reproduced image signal
G ₁ (i, j, k) is stored in the reproduced image signal memory area 13b of the image memory 13.

The image regularization unit 15 reproduces the reproduced image signal G ₁ (i, j,
k) is normalized.

The image entropy measurement unit 16 is the image normalization unit 15
The image entropy A(x) is measured from the signal G ₂ (i, j, k) normalized by .

Image sharpness display section 17 shows image entropy A(x)
Image sharpness Z(x) is calculated from and output. The image optimization unit 18 determines the parameter x _nax that maximizes the image sharpness Z(x).

Then, the point spread function to which this parameter value x=x _nax is applied is subjected to Fourier transformation. This Fourier-transformed point spread function and the original image signal D(i,j)
Optimal reproduction image signal by deconvolution
G ₁ (i, j, x=x _nax ) is obtained and the image memory 13
The reproduced image signal is stored in the reproduced image signal memory area 13b.

Next, the operation of this embodiment will be explained.

In FIG. 2, object 1 is constantly rotating at an unknown rotational speed x. This rotation speed is a parameter that is a condition when acquiring an image signal from the object 1.

First, the ultrasonic oscillator 5 and the image signal input section 11 are placed at the top of the scale 6. The ultrasonic oscillator 5 emits ultrasonic waves for a predetermined period of time while the object 1 rotates once. A reflected wave reflected from the object 1 is input as an image signal to an image signal input section 11. During one rotation, m pieces of data are collected, and m pieces of pixel data on the first row of the hologram surface is collected.
In this way, an image signal serving as information on the pixels in the first row on the hologram surface is obtained. This signal input method is an application of the concept of the synthetic aperture method used in the field of radio communications.

Next, the ultrasonic oscillator 5 moves downward by a moving distance D divided by n in the height direction. Then, in the same manner as in the first row of the hologram surface, image signals serving as information for each pixel in the second row are acquired. In this way, by acquiring an image signal every time the ultrasonic oscillator 5 and the image signal input section 11 are moved downward n-1 times, n
Information on n×m pixels on the hologram surface in rows and m columns is acquired.

The image signal thus obtained is converted into a continuous electrical signal by the image signal input section 11. This electrical signal is converted by the A/D converter 12 into a discrete electrical signal quantized with 8-bit gradations at an appropriate sampling period determined by the processing speed of the entire system. This quantized discrete signal is stored in the original image signal memory area 13a of the image memory 13 as information for each pixel on the hologram surface of n rows and m columns. In this way, the image of the object 1 is obtained as information on each n×m pixel of the hologram plane of n rows and m columns, which has been subjected to Fourier transformation. This stored signal is referred to as an original image signal D(i,j).
Here, D(i,j) and two-dimensional array elements i,j have 8-bit gradation, 0≦D(i,j)≦255, 1
≦i≦n, 1≦j≦m, and i, j, and D (i, j) are integers. This original image signal D (i, j) is subjected to a correlation calculation with a Fourier-transformed point spread function created by applying an appropriate value x1 to the parameter x, which is a condition when acquiring an image signal in the image reproduction unit 11. By doing so, the reproduced image signal G ₁ (i, j, x=
x1). And this reproduced image signal G ₁
(i, j, x=x1) is stored in the reproduced image signal memory area 13b of the image memory 13. This reproduced image signal G ₁ (i, j, x=x1) constitutes a reproduced image of n rows and m columns.

The reproduced image signals G ₁ (i, j, x=x1) are each normalized into a two-dimensional distribution probability density function by the image normalization unit 15. This is the reproduced image signal G ₁ (i, j, x
= x1) Each signal G ₁ (i, j, x
= x1). Letting this value be the reproduced image normalized signal G ₂ (i, j, x=x1), the following equation is obtained.

The reproduced image normalized signal obtained in this way
G ₂ (i, j, x=x1) is 0≦G ₂ (i, j, x=
The probability distribution is x1)≦1.

This reproduced image normalized signal G ₂ (i, j, x=x1)
From this, the image entropy measuring unit 16 calculates the image entropy A (x=x1). This image entropy A (x = x1) is a value that represents the degree of diffusion of a probability density function with image coordinates as a variable on the reproduced screen (image information diffusion degree), for example, the degree of blurring or blurring of the image. be. This probability diffusivity is evaluated by applying the concept of entropy of information. This image entropy A (x = x1) is the sum of the values obtained by multiplying each reproduced image normalized signal G ₂ (i, j, x = x1) by the logarithm of this signal value for all signals, and the sign is It is obtained by inverting , and becomes as follows.

A (x=x1)=- _o Σ ⁱ⁼¹ _n Σ ⁱ⁼¹ G ₂ (i, j, x=x1) logG ₂ (i, j, x=x1) However, G ₂ (i, j, x= When x1)=0,
Set logG ₂ (i, j, x=x1)≦0. here,
The numerical range of the image entropy A(x) found in the reproduction screen of n rows and m columns is determined by the discretization condition of the image signal, and in this case, 0≦A(x)≦lognm
becomes. This means that the image entropy A(x) is minimum, that is, the probability is concentrated in one pixel in the playback screen,
The reproduced image normalized signal G ₂ (iO,
When jO,x) is 1, the image entropy A(x) becomes 0 according to the above equation. Furthermore, when the image entropy A(x) is maximum, that is, the probability is uniformly distributed over the entire playback screen as 1/(n×m),
The image entropy A(x) is lognm.

The image entropy A obtained in this way
(x=x1) is converted into image sharpness Z (x=x1) by the image sharpness display section 17 and output. This image sharpness Z(x) is a value that indicates the degree of sharpness of the image, that is, the degree of adjustment of focus alignment, and is the opposite of the degree of blurring of the image, which indicates the degree of blurring of the image that is proportional to the image entropy A(x). It is a proportional value. This image sharpness Z (x = x1) is the maximum value of image entropy A (x = x1) determined by the discretization conditions.
It can be expressed as the following formula using lognm.

Z(x=x1)=1/lognm ・{lognm−A(x=x1)} As solved from the above equation, image sharpness Z(x=
x1) is a value normalized within the range of 0 and 1.
When the image sharpness Z(x) is maximum, a sharp image is obtained in which focus alignment is properly performed. The image optimization unit 18 determines the optimal parameter x _nax that maximizes the image sharpness Z(x). Image sharpness Z(x) is a function of parameter x, which is a condition when acquiring an image signal. When this parameter value is x _nax , image sharpness Z(x
= x _nax ) is maximized. From this, the parameter x value is successively changed so that the value obtained by subtracting the image sharpness Z(x) by the parameter x becomes 0 or approximates 0.

ΔZ(x)/ΔX=0 Then, the parameter value when the above equation holds or approximately holds, that is, |Δz/Δx|<ε; ε is sufficiently small, is output to the image reproduction unit 14 as the optimal parameter value x _nax . . The image reproduction unit 14 performs Fourier transform on the point spread function to which this parameter value x=x _nax is applied. Then, the original image signal D(i,j) stored in the original image signal memory area 13a of the image memory 13 is read out, and the read original image signal D(i,j) and the point spread function are decombined. The optimum reproduced image signal G ₁ (i,
j, x = _nax ). Then, this optimum reproduced image signal G ₁ (i, j, x= _nax ) is stored in the reproduced image signal memory area 13b of the image memory 13.

The image sharpness Z (x=x _nax ) corresponding to this optimal reproduced image signal G ₁ (i, j, x= _nax ) is determined by the image normalization section 15 , image entropy measurement section 16 , and image sharpness display section 17 . It is sought after.

When this image sharpness Z (x=x _nax ) is supplied to the image optimization unit 18, the image optimization unit 18 outputs the optimal parameter value x _nax and the image sharpness Z (x=x _nax ), and Optimal reproduced image signal G ₁ (i, j,
x=x _nax ) is read out from the image memory 13 and output.

As explained above, in one measurement, image signals inputted as reflections of ultrasonic waves from an object are exchanged and stored as original image signals constituting a two-dimensional hologram surface. A reproduced image signal is constructed by applying parameter values that are conditions for acquiring an image signal to this original image signal. Then, the optimum reproduced image signal is obtained by changing the parameter values of the reproduced image signal. Therefore, by numerically changing parameter values using the original image signal and analyzing the resulting reproduced image signal, it is possible to easily measure the values of unknown parameters that are the conditions for image signal acquisition. . Moreover, at this time, an optimal reproduced image signal can also be obtained.

In one embodiment, image entropy is converted to image clarity, which is a value proportional to the degree of focus matching, and parameters that are conditions for image signal acquisition are changed so that this image clarity is maximized. Ta. However, in this invention, if the parameters are changed so that the image entropy, which is inversely proportional to the degree of focus matching, is minimized, the optimal reproduced image signal is obtained by applying the optimal parameters that minimize the image entropy. can be obtained. In other words, if it is not necessary to convert the image entropy into a standardized value proportional to the degree of focus alignment, the image sharpness display section is removed and the image optimization section sets the parameters so that the image entropy is minimized. Just change it.

Further, in one embodiment, an image signal obtained from an object is stored as an original image signal constituting a two-dimensional hologram surface. However, in the present invention, 1
It is also possible to store the original image signal as a subsequent hologram.

Further, in one embodiment, an A/D converter converts the image signal into a discrete signal quantized with 8-bit gradation. However, in the present invention, quantization may be performed at an appropriate gradation depending on the level of the image signal.

Furthermore, in the present invention, as long as the degree of focus alignment of an image can be measured, the same operation as in the embodiment can be achieved even if the image normalization operation and the definition of image entropy are slightly modified. In one embodiment, each reproduced image signal is divided by the total sum of all reproduced image signals as an operation to normalize the reproduced image signal. However, in the present invention, it is only necessary to know the diffusion state of each reproduced image signal with respect to all reproduced image signals. For example, the minimum signal value is found among the reproduced image signals, and this minimum reproduced image signal is subtracted from each reproduced image signal. The reproduced image normalized signal can also be obtained by dividing this subtracted value by the total sum of the subtracted values over the entire reproduced screen.
When image entropy is calculated using this reproduced image normalized signal, signal values less than the minimum reproduced image signal value are removed, so the range in which image entropy can change can be expanded.

Further, in one embodiment, the image sharpness Z is normalized using the maximum value lognm determined by the discretization conditions of the image entropy A.

However, in the present invention, the formula for calculating the image sharpness Z may be modified depending on the degree of focus matching of the image to be determined. For example, the obtained image sharpness Z can be transformed into a psychological image sharpness _Z1 that suits human senses. This uses the image sharpness Z as a variable of a function representing human sensation to calculate the psychological image sharpness Z ₁ using the following formula.

Z ₁ =−logZ Even if the result obtained in this manner is output, the object of the present invention can be achieved.

Furthermore, in one embodiment, one of the parameters serving as a condition for acquiring an image signal was the rotation speed of the object. However, the present invention can be applied even when there are a plurality of parameters. This applies these multiple parameters x1, x2, x3, ..., xh to the original image signal to generate a reproduced image signal G ₁ (x1, x2, x3, ..., xh)
Convert to Here, multiple parameters are
(x1, x2, x3, ..., xh) Expressed as ^t .

Then, from the reproduced image signal G ₁ (x→), the image sharpness Z
(x→) is determined, and the parameter X→ _nax that maximizes the image sharpness is determined using the following formula.
(Here, t indicates the transposed matrix.) gradZ (X → _nax ) = 0 → However, grad = (d/dx1, d/dx2, d/dx3, ..., d/dxh,), 0 → is zero Vector The parameter x → _nax obtained in this way is
It is output to the image reproduction section as the optimum parameter.
The image entropy obtained from the reproduced image signal to which this optimal parameter is applied becomes the minimum value.

Next, other embodiments of the present invention will be described. In one embodiment, reflected waves of ultrasonic waves from an object are introduced as images. However, there are other methods in which the phase amplitude information of the reflected wave from an optical object is introduced as an image signal, invisible pattern information by invisible light is introduced as an image signal, and an image signal is generated using an electric signal representing the object as a medium. It is also possible to consider a method of introducing light, and in this embodiment, a method of irradiating an object with light and introducing a reflected wave of the light as an image signal will be described.

This electrical signal is converted into a continuous electrical signal by a photoelectric conversion element at the image signal input section. This electrical signal is input to the image input unit to produce the three primary colors of light, namely red,
The image signal is decomposed into image signals for each primary color by bandpass filters corresponding to each frequency band of green and blue (hereinafter referred to as R, G, and B). Each image signal is stored in the image memory as an original image signal of a discrete electric signal quantized by an A/D converter in the same manner as in the embodiment. Then, the image reproducing unit converts each original image signal into a reproduced image signal by applying parameters serving as conditions for image signal acquisition to each original image signal. The R component image sharpness, the G component image sharpness, and the B component image sharpness are determined from each reproduced image signal in the same manner as in one embodiment. Each component image sharpness is the third
As shown in the figure, a three-dimensional (Euclidean) space is constructed. The color image sharpness Zcolor(x) is determined as one point in this three-dimensional space, and evaluated as the Euclidean distance between this point and the origin, or the square of the Euclidean distance. This color image sharpness Zcolor(x) is output to the image optimization section. Then, the image optimization section determines the optimal parameters that maximize the color image sharpness, and outputs the optimal reproduced image signal in the same manner as in the first embodiment.

〔Effect of the invention〕

The present invention stores an image signal acquired from an image target in one measurement, applies parameters that are conditions for image signal acquisition to this stored image signal, calculates image entropy from the obtained signal, and By numerically changing the parameters so that the image entropy is minimized, the image entropy is minimized from the image signal acquired in one measurement. That is, an image signal with proper focus matching can be obtained by numerically processing parameters.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing the configuration of an embodiment of the present invention, Fig. 2 is a diagram of a measuring device for explaining the principle of the embodiment of the invention, and Fig. 3 is the principle of another embodiment of the invention. It is a vector diagram for explaining. Explanation of symbols, 1...object, 2...turntable,
3... Rod, 5... Ultrasonic oscillator, 6... Scale, 10
...input circuit section, 11...image reproduction section, 12...A/D
Converter, 13...image memory, 14...image reproduction section,
15... Image normalization section, 16... Image entrance measurement section, 17... Image sharpness display section, 18... Image optimization section.

Claims (1)

[Claims] 1. An input circuit unit that converts an image signal obtained from an image object into an original image signal that is a quantized discrete signal and stores it, and applies parameters that are conditions for obtaining the image signal. an image reproducing section that calculates a correlation between the point spread function created by the method and the original image signal, converts the original image signal into a reproduced image signal, and stores the converted image signal in the input circuit section; and a reproduced image of the image reproducing section. an image normalization section that normalizes the signal; and an image that measures the image entropy from the sum of the values obtained by multiplying the signal value of the signal normalized by the image normalization section and the logarithm of this signal value. an entropy measurement unit; and an image optimization unit that changes the parameter value to find an optimal parameter value that minimizes the image entropy, and outputs this optimal value to the image reproduction unit, An image signal processing device that outputs a reproduced image signal that is converted from an original image signal using an optimal value of and minimizes image entropy.