JPH04576A - Processor for image signal - Google Patents

Abstract

PURPOSE:To attain image forming processing capable of detecting a sharp edge against noise executing the differential operation of an input image signal along plural azimuth elements, finding out their absolute values and removing the absolute value based upon the noise. CONSTITUTION:Image information obtained from an image pickup part 11 such as a telecamera is numerized at the illumination I(X,Y) of each picture element through an A/D conversion part 12 and stored in a memory 13. Then a differential operation part 14 executes the differential operation of the image and finds out its absolute value as I(X,Y,theta). The theta is a parameter indicating the direction of differential operation. Out of the I(X,Y,theta), a component to be moderately changed about the angle theta is attenuated by a direction emphasizing part 15. When the output I(X,Y,theta) of the emphasizing part 15 is added to the succeeding image restoring part 16, a high frequency noise can be attenuated and an image (X,Y) emphasizing the edge part can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image signal processing device, and more particularly to an image signal processing device that emphasizes edges.

[Conventional technology]

Detecting edges in areas where brightness changes rapidly in an image plays an important role as the first step in an image recognition system.

Many methods are known as edge detection algorithms. It is well known that rapid changes in brightness can be captured by applying first-order differential operations to waveform signals, but since differential operations can be considered a type of high-frequency pass filter, this method can be used to detect Even high-frequency noise is detected. In order to avoid this, it is usually attempted to remove noise components by performing threshold processing after the differentiation operation. (Rosenfeld (A, Rosenfeld), Cutter (A, C,
Kak), supervised translation by Makoto Nagao, Digital Image Processing, Kindai Kagakusha, 1976).

It is also well known to detect edges by applying a second-order differential operation or a Laplacian operator to an image signal and detecting a zero point in the output.

[Problem to be solved by the invention]

Among the conventional edge detection algorithms mentioned above, those that perform first-order differential operation on waveform signals perform threshold processing to suppress noise detected other than edges.
A drawback is that it is difficult to determine an appropriate threshold value that does not depend on the input image.

In addition, edge detection algorithms that apply second-order differential operations or Laplacian operators to image signals to detect the zero point in the output are vulnerable to high-frequency noise because these operations can also be regarded as a type of high-frequency filter. There is a drawback.

SUMMARY OF THE INVENTION An object of the present invention is to improve the above-mentioned conventional drawbacks and to provide an image signal processing device that can detect edges resistant to noise.

[Means to solve the problem]

The device of the present invention is an image signal processing device that emphasizes and processes edges of parts where the brightness changes rapidly included in an image signal, and performs differential operations along a plurality of directions with respect to an input image signal. determine whether the absolute value is based on the edge or on noise, depending on whether the absolute value obtained for each point of the input image has orientation dependence. The image forming apparatus is configured to generate an image in which edges are emphasized by determining only the absolute value based on the edge.

Furthermore, the apparatus of the present invention is an image signal processing apparatus that emphasizes and processes the edges of parts where there is a rapid change in brightness included in an image signal, and which processes an input image signal by overlapping it into a small area including a plurality of pixels. For each of the small regions, a Fourier transform power spectrum is calculated by performing Fourier transform on only the components exceeding a critical spatial frequency set in advance based on the degree of gentleness of the edge to be emphasized. , based on the edge by removing noise components that are independent of the angle from the Fourier transform power spectrum that depends on the magnitude of the spatial frequency indicating the fineness of the change in the image signal and the angle indicating the direction of the changing component. Edges are extracted by performing a Fourier inverse transform on objects only, and then a threshold setting process is applied to correct data changes due to the Fourier transform and inverse transform. It has the configuration to generate.

[Effect]

In the first invention, differential calculation operations are performed on the input image along various directions at each point, and the absolute value of the output is calculated. As is well known, the differential calculation operation changes the image so that the edges of the image are emphasized, but since the differential calculation can be regarded as a type of high-frequency pass filter, - for boat fishing, the output contains high-frequency noise components. is also included. Next, we focused on the fact that the dependence of these values on the azimuth angle when calculating the differential operation is different between edges and noise, and calculated the edge output after removing noise that has no azimuth dependence at each point of the image. be done.

While true edges have local orientation, high-frequency noise is generally isohistorical. Therefore, as shown in FIG. 2(a), the pressure of the differential calculation on the edge 101 is large when the differential output 103 in a specific direction, for example, in the horizontal method, is applied to the edge 101, whereas the differential output 104 in the vertical direction takes a large value.
does not occur because the brightness remains dark, and there is a clear dependence on the azimuth angle.

On the other hand, in the case of noise, as shown in FIG. 2(b), isotropic noise 102 produces the same level of output in all directions, such as horizontal differential output 105 and vertical differential output 106. bring. Therefore, it is possible to easily determine whether the differential output is an edge-like output or a noise-like output based on the presence or absence of azimuth dependence. In other words, it is possible to exclude as noise anything that gives an approximately constant differential output, thereby realizing edge detection that is resistant to noise.

Next, in the second aspect of the present invention, the input image signal is divided into several small regions including a plurality of pixels, and for each small region, only the components having a critical spatial frequency Kc or higher are subjected to Fourier transformation. A transformed power spectrum is calculated. The above-mentioned critical spatial frequency Kc is set in advance depending on how gently, or in other words, how finely, the edge portion is to be extracted.

Edge parts of the image where the brightness changes rapidly have high spatial frequency components, whereas areas with constant brightness do not have high frequency components, so the Fourier transform power spectrum is zero only in the edge parts. takes a value that is not The Fourier transform power spectrum in this case can be considered as a function of the magnitude of the spatial frequency and the angle indicating the direction of the changing component, and the true edge is oriented and for a spatial frequency at a specific angle. In contrast, isotropic noise outputs a negative value regardless of the angle.

Based on these conditions, isotropic noise components are suppressed and only edge components are extracted, which is then subjected to Fourier transformation.
Threshold processing is performed to obtain an image with reduced noise and enhanced edges.

〔Example〕

Next, the present invention will be explained with reference to the drawings.

FIG. 1(a) is a block diagram of a first embodiment of the present invention.

The embodiment shown in FIG. 1 includes an imaging unit 11 that uses a television camera or an image scanner, an A/D conversion unit 12 that digitizes data, an image memory 13 that stores digital data, and a system that reads data from the image memory 13. a differential calculation unit 14 that performs differential calculations on image data and outputs edge components containing noise; and a directional unit that removes noise components from the output of the differential calculation unit 15 and emphasizes and outputs edge components that have directionality. Emphasizing section 15 and directionality emphasizing section 1
5 and an image restoring section 16 for restoring and outputting an image with enhanced edges based on the edges output from the image memory 13 and data read from the image memory 13.

Next, the operation of the first embodiment will be explained.

Image information input from an imaging unit 11 using a television camera or an image scanner is sent to an A/D converter 12.
The brightness I (X, Y) for each pixel is converted into numerical values and stored in the image memory 13.

In the processing in this embodiment, edges in the image are emphasized by performing the following processing using a processor or the like for this image information processing.

First, the differential calculation unit 14 performs differential calculation on the image according to a normal method and calculates its absolute value. The result is I(X
, Y, θ). Here, θ is a parameter indicating the direction in which the differential operation was performed. The factor calculated in this way takes a large value in a portion where the pixel value changes rapidly. I(X
, Y, θ) are sent to the directionality emphasizing section 15, which attenuates components that change slowly with respect to the angle θ. That is,
The output of the differential calculation for true edges takes a large value only in a specific direction θ, whereas the noise component without directionality has the same output in all directions, so directionality emphasis is used. The unit 15 examines the variability of the output of the differential calculation unit 14 using the angle θ as a parameter, determines negative pressure as noise and eliminates it, and attenuates the noise component without attenuating the output of the edge portion.

The output HX, Y, θ) of the directionality emphasizing unit 15 is converted to an image in which the high-frequency noise is attenuated and the edge portion is emphasized by adding the θ components at each point (X, Y) in the next image restoration unit 16. I(X, Y) is obtained. Image restoration unit 16
The processing is given by the following equation.

HX, Y) = ): I (X, Y, θ) θ Of course, if it is necessary to obtain the edge portion in each direction in subsequent processing, it is also possible to apply pressure as is.

FIG. 1(b) is a configuration diagram showing a second embodiment of the present invention.

The embodiment shown in FIG. 1(b) divides the image data read out from the image capturing section 21, A/D converting section 22, image memory 23, and image memory into small regions, and performs a Fourier A Fourier transform calculation section 24 that performs transformation to obtain a Fourier transform power spectrum, a directionality emphasis section 25 that emphasizes edges with respect to the output of the Fourier transform calculation section 24, and a directionality emphasis section 25 that applies the output of the Fourier transform calculation section 24 and the directionality emphasis section 26. and an image restoration unit 26 that generates an image using the image data.

Next, the operation of the embodiment shown in FIG. 2 will be explained.

Image information input from an imaging unit 21 using a television camera or an image scanner is sent to an A/D converter 22.
The brightness I (X, Y) for each pixel is converted into numerical values and stored in the image memory 23. In the second embodiment, the following processing is performed on this image information using a processor or the like.

Now, assume that the number of pixels of the input image is NxN. Now, let's define this input image as n2X n, consisting of the number of pixels n, x n.
Divide into 2 small areas. When dividing the image so that there is no overlap, N=n+x
n2, but it may be divided so that there is overlap with each other. Next, the Fourier transform calculation unit 24
However, for each of the small regions thus divided, the Fourier transform and Fourier transform power spectrum are calculated only for components having a predetermined critical spatial frequency Kc or higher. The value of Kc mentioned above is set depending on the purpose of how much gentle edges are to be emphasized.

Assuming that the maximum spatial frequency is Kmax, in order to emphasize up to the edges that change over about m pixels, Kc is K.
It may be set to about max/+n.

Here, each small area is represented by a subscript (X, YXX, Y=1-n2), and one pixel in each small area is represented by a subscript (x, y).
The value of a pixel in one of the pixels is expressed as I(x, y:X, Y).

Then, the Fourier transform F and its power spectrum P described above are calculated by the following equation.

x exp(i 2π(x-Kx+y-Ky)/n2)
P(Kx, Ky:X, Y) = F(Kx, Ky:X,
Y) In a region where the brightness of the image is constant, the Fourier transform does not include high frequency components, so F(K
x, Ky :X, Y) and P(Kx, Ky :X, Y)
is present only in areas where brightness changes rapidly. next,
The power spectrum P(Kx, Ky
:X, Y) are sent to the directionality emphasizing section 25, and first, Kx,
It is converted into polar coordinate representation P(K, θ; X, Y) for Ky. Since a true edge has an orientation, a certain angle θ
It takes a large value for the spatial frequency with
Isotropic noise outputs a −-like output regardless of the angle θ.

Therefore, using the angle θ as a parameter, the polar coordinate display P(K,
It is possible to examine changes in the pressure of θ:X, Y) and determine that --like outputs are noise and eliminate them, thereby attenuating isotropic noise components. Furthermore, the directionality emphasizing unit 25 returns this output to the display regarding Kx and Ky, and thus obtains P'(Kx.

Ky:X, Y) is sent to the next image restoration section 26.

In the image restoration unit 26, this P'(Kx, Ky: X, Y)
The Fourier transform F (Kx, Ky : y:X, Y). For example, F'(Kx, Ky:X, Y) is calculated by the following equation.

F'(Kx, Ky:X, Y) = 0(P(Kx, K
If y:X,Y)=0)F'(Kx,Ky:X,Y)
=F(Kx, Ky:X, Y)x(P'(Kx, Ky:X
, Y)/P(Kx, Ky:X, Y))(P(Kx, Ky
:X, Y) is other than 0) Furthermore, the image restoration unit 26
' Perform threshold processing on (x, y: X, Y),
Replace negative values in inverse Fourier transform with 0. This is because in the above processing, it is generally not guaranteed that ■' will be a positive value. In the reconstructed image obtained in this way, the edge portions are emphasized while isohistorical high-frequency noise is suppressed.

〔Effect of the invention〕

As explained above, the present invention has the effect of enabling image generation processing that enables edge detection that is resistant to noise by removing noise components included in edge information of image data and performing image generation processing. be.

11.21... Imaging section, 12.22...A
/D converter, 13.23... Image memory, 14
... Differential operation section, 24 ... Fourier transform calculation section, 15°25 ... Directionality emphasis section, 16.
26... Image restoration section, 101... Edge, 102... Noise.

Agent: Patent Attorney Susumu Uchihara

[Brief explanation of drawings]

FIG. 1(a) is a block diagram of the first embodiment of the present invention, FIG. 1(b) is a block diagram of the second embodiment of the present invention, and FIG. 2(a)
2(b) is an explanatory diagram of edge detection in the embodiment of FIG. 1, and FIG. 2(b) is an explanatory diagram of noise detection in the embodiment of FIG.

Claims (1)

[Scope of Claims] 1. An image signal processing device that emphasizes and processes the edge of a portion of a sudden change in brightness contained in an image signal, which performs differentiation along a plurality of directions with respect to an input image signal. Calculate the absolute value by performing a calculation, and determine whether the absolute value obtained for each point of the input image is based on the edge or based on noise, depending on whether or not the absolute value has orientation dependence. An image signal processing device characterized in that the image signal processing device generates an image in which the edges are emphasized by determining whether the edges are the same and securing only the absolute values based on the edges. 2. An image signal processing device for emphasizing and processing the edge of a portion where there is a sudden change in brightness contained in an image signal, the input image signal being divided into small regions including a plurality of pixels with overlap allowed; For each of the small regions, a Fourier transform is performed on only the components exceeding a preset critical spatial frequency based on the degree of gentleness of the edge to be emphasized, and a Fourier transform power spectrum is calculated, and changes in the image signal are calculated. From the Fourier transform power spectrum, which depends on the magnitude of the spatial frequency indicating the fineness of the minute and the angle indicating the direction of the changing component, noise components that have no dependence on the angle are removed, and only those based on the edge are targeted. It is characterized by extracting edges by performing inverse Fourier transform, then applying threshold setting processing to correct data changes due to Fourier transform and inverse transform, and securing the edges as edges to generate an image with emphasized edges. Image signal processing device.