JPS61201373A - Processing method for image signal - Google Patents

Abstract

PURPOSE:To reduce the quantity of calculation for an autocorrelation function, etc., in a local area by projecting the light-shade degree of picture elements in the local area on segments which have various predetermined angles and are on the same plane with the picture elements. CONSTITUTION:The picture elements in a small area (s) in an image to be processed are projected on a straight line L which has an angle theta to an X axis. Namely, values of picture elements on a straight line l which has an angle theta+pi/2 and is at distance (t) from a center (o) are summed up and stored in an address (t,theta) of an image memory repeatedly while (t) is varied from -R to +R and theta is varied from (o) to pi. When the contents of the address (t,theta) are denoted as A(t,theta), a two-dimensional spatial frequency power spectrum P(k,theta) regarding the index (t) is obtained by squaring the absolute value of a linear discrete Fourier transform F(k,theta) regarding the index (t) of the A(t,theta).

Description

[Detailed description of the invention] (Industrial application field) The present invention relates to an image signal processing method.

(Prior art and its problems) In image processing, various statistics such as mean and one variance are locally calculated and used for slam recognition, texture chair discrimination, etc. (For example, Toriwaki et al. Yokoi: Image processing algorithms, Information Processing, vol. 21, p.
613). Computing the local spatial frequency power spectrum of an image, or the local autocorrelation function, is also an analysis of the texture chair, or pattern Ii! ! ! m
It is considered to be a useful process when carrying out t, but
In that case, the amount of calculation would be enormous, making it difficult to put it into practical use.

(Objective of the Invention) The object of the present invention is to eliminate such conventional problems in calculating the local spatial frequency power spectrum or local autocorrelation function of an image, and to reduce the amount of calculation. An object of the present invention is to provide a method for processing image signals that can be processed at high speed.

(Structure of the Invention) The image signal processing method according to the present invention is a method for calculating a local autocorrelation function or a spatial frequency power spectrum of one image. , and project the brightness of pixels in that region onto a line segment that is coplanar with one image with various predetermined angles, and for each of these projected images, calculate its autocorrelation function or This is achieved by performing a process of determining the spatial frequency power spectrum with respect to the coordinates on the line segment.

(Example) Fig. 1 is a block diagram for implementing the processing method of one embodiment of the present invention, Fig. 2 is a plan view showing an example of an image subjected to the processing of Fig. 1, and Fig. 3 is a block diagram for implementing the processing method of an embodiment of the present invention. A block diagram that implements the processing method of the embodiment! ! FIG. 4 is a plan view showing an example of an image to which the processing of FIG. 3 is applied.

As shown in FIG. 1, in this embodiment, a scanning unit 1 comprising a television camera or an image scanner scans a pattern as an #L image and converts it into an image information signal. The brightness of this image information signal is converted into numerical values for each pixel by the A/D converter 2 and stored in the image memory 3. In the processing method according to this embodiment, the image information signal stored in the image memo IJ I 3 is first subjected to the following conversion using the processor 14.

First, a pixel t'' is extracted from a predetermined area in the image to be processed.The shape and size of this area are arbitrary, but
Here, for simplicity, a case of a circular area with a radius R will be described. The small head region S indicated by a circle in FIG. 2 is the region to be processed from now on.

Further, the frame G of the image, the Y axis, and the Y axis are determined as shown in FIG. The pixels in this small area S are projected onto a straight line forming an angle θ with the Y axis. That is, at the angle π/2+θ,
Add the values of pixels located directly on the ml whose distance from the center of the small head area S under consideration is t, and calculate the value at (ttθ) in the image memory I5 shown in Fig. 1. Store at the specified address.

Such projection is performed by varying t in the range -R to R and θ in the range 0 to π at predetermined sampling intervals. This sampling interval determines the resolution of the image.

The above-described conversion is similar to the measurement of projection data in so-called XWCT. In various types of CT such as X-ray CT', a narrowly focused beam is scanned and irradiated onto the target, and a detector detects the transmitted 1! k is measured to find the amount of attenuation of this XIM. At this time, the amount of attenuation of Xl@ is proportional to the integral value of @ degrees of the object along a straight line on the beam. Therefore, by performing these five scans while changing the angle, if the brightness of the pixel in this process is regarded as the density of the object in the case of X-ray CT, then the ( t
+θ) expression is obtained as projection data in the case of X-ray CT.

As is well known, in ICT, the density distribution of the original object can be reproduced by computer processing the projection data obtained in this way. In exactly the same way, (1,
θ) From the representation, it is possible to reproduce a one-dimensional image within the resolution range determined by the sampling interval. Therefore, the above-mentioned conversion in this process will not cause any image information to be missing.

By performing such a transformation by the processor 14, the amount of processing required to later calculate the local autocorrelation function or spatial frequency power spectrum of the image can be significantly reduced.

Next, the processor 6 performs the following processing on the thus obtained data in the image memory I5.

Now, assume the value t-A (t#θ) at address (1, θ) in the image memory IfS.

For example, when it is desired to obtain a two-dimensional spatial frequency power spectrum, the following processing is performed on A(t, θ).

F(k, θ)=fA(t, θ)eX)'(-i2y
rkt)...(1) Represents a vector. The processing in equation (1) can be executed by a fast Fourier transform algorithm, and in that case, further reduction in processing time can be expected.

Alternatively, when it is desired to obtain a local autocorrelation function of one image, the following processing is performed on A(t, θ).

C(d,“)=,i,A(suA(t-d)where,(:
:(d, θ) is an autocorrelation function with respect to the subscript t of A(t 2 , θ).

P(k, θ) is a two-arrow spatial frequency power spectrum of a one-dimensional image expressed in t-polar coordinates. Also,
C(d, θ)ti is not the local autocorrelation function of the image itself, but it is expressed as (1, θ) after undergoing exactly the same transformation as the one just applied to the original image. equal. These things can be proved analytically as follows.

Now, suppose that the value of the pixel at the position r=CX, Y) is expressed by a(r). Then, the above A ('' *θ) is.

A (t, θ)=Σa(r)δ(t−rcos(ψr−
θ))−+8)? Here r = l r l , ψ, = arg(r), δ
It is given by (X)=. In this case, the subscript t of A(t, #)
The autocorrelation function C(d, θ) for is.

C(d, θ)ΣA(t, θ)A(t-d, θ)δ(t
−d−r2cos(ψ72−θ))=Σ4 a<7. ＞
a<rz)δ(-d+rlcos (F,,-〇)?□
, r2 −r2cos (ψ7□−θ)) cos (ψ?1−7□・−10nn=1C(Δ)δ(d〜1Δ1 cos (ψj−θ))
Here, c(Δ)==Σa(r) a (r+Δ) is calculated.

As can be seen by comparing with Equation (3), this is nothing but the autocorrelation function of the image expressed as 0(Δ) e (t*θ).

)'(k, θ), according to the Wiener-Khinchin theorem.

The spatial frequency power spectrum p(k, θ) with respect to t of A(t, θ) is equal to the autocorrelation function with respect to t of A(t, θ) (:: the Fourier transform with respect to d of (d, θ). Therefore p(k, θ) is )' (k, θ)=ΣC(
d, θ) is given by eXP (−d to −i2π). Transforming this, exp(-i2π to -d) = ΣC(Δ)eXp(-i 2xk-Δ) where k
2(k cos θ, k sin θ), which is −
11P beam According to the Wiener-Kinchin theorem, ll1
It is equal to the two-dimensional spatial frequency power spectrum P(k) of the JI image a(r). Therefore.

P(k, θ) is nothing but a two-dimensional spatial frequency power spectrum P(k)'t-1 expressed in polar coordinates.

For the same reason that it is possible to reconstruct the original image from the (1, θ) representation of one image as described above,
(From H(d, 0), the local autocorrelation function of the original image can be reconstructed. In that sense, C contains m information equivalent to the local autocorrelation function of the original image. You can say that it is.

Now, when trying to calculate the two-dimensional spatial frequency power spectrum or autocorrelation function of one image, if the number of pixels in one image is ~N2, ~N4 calculations (multiplying 4) are required (using fast Fourier transform). (This number will be reduced further.) However, in the processing method according to the present embodiment, Fourier transform or autocorrelation function calculation, which requires multiplication, is difficult.

Since it is only necessary to perform the calculation for the one-dimensional subscript (1), the amount of calculation is reduced to -N3.

This is the reason why the processing method according to this embodiment can achieve high-speed processing.

The processing result (P or C) thus obtained is sent to the next determination unit 7, where it is used for processing such as slam recognition or texture discrimination.

FIG. 3 shows a block diagram of another embodiment in which the processing according to the present invention is implemented in a multiprocessor configuration and applied to a slam recognition device.

The image alert signal of the bang inputted from the imaging scanning section 1 is
The brightness level is converted into a number for each pixel by the A/D converter 2 and stored in the image memory 13. In the processing according to this embodiment, this image is divided into several small regions Sl to 8i that overlap with each other. FIG. 4 shows an example of division of C.

In FIG. 4, the small regions 81 to Si have a circular shape, but this shape is arbitrary.

In another embodiment shown in FIG. 3, each divided small area S
Processing as described above for 1 to SI is executed in parallel by dedicated processors 41%i and 6l-xi, respectively. The subscript that distinguishes each small area Sl to si is represented by i, and the processing result t-P(
k, θ; i). p(k, θ; i) calculated by each processor is:

It is sent to the next slam judgment unit 8. In this slam determination unit 8, this three-dimensional slam P(k.

With respect to θ;

As is well known, the spatial frequency power spectrum of a batan remains unchanged even when the batan is translated by t, so the p(
The three-dimensional pattern of k, θ; i) is very stable against displacement of the button on the order of the size of the small area Si. Therefore, by the method shown in FIG. 3, it is possible to realize an image signal processing method that can quickly execute the slam recognition R that is resistant to positional deviation of the putter.

(Effects of the Invention) As explained above, the present invention projects the brightness of pixels onto line segments at various angles on the image plane, and processes the projection data to localize the autocorrelation of one image. This method has the advantage that a function or a local spatial frequency spectrum ll, - can be obtained with a much smaller amount of calculation than conventional methods.

When calculating the autocorrelation function, if the number of pixels is ~N2, the conventional method requires ~N4 calculations (multiplications)'1, whereas the method according to the present invention requires ~N3 calculations. (If you use fast Fourier transform to obtain the spatial frequency spectrum, this reduction in calculation amount will be less than in this evaluation, but it will still require a significant reduction in calculation amount.) realizable).

[Brief explanation of drawings]

FIG. 1 is a block diagram for implementing the processing method of one embodiment of the present invention, FIG. 2 is a plan view showing an example of an image subjected to the processing of FIG. 1, and FIG. FIG. 4 is a block diagram for implementing the example processing method, and FIG. 4 is a plan view showing the processing shown in FIG. 3. Roset'X, 7-...-judgment unit, 8...
...Bang judgment unit.

Claims (1)

[Claims] For the image signal, extract a part of the area in the image, project the brightness of the pixels in that area onto line segments that are on the same plane as the image and have various predetermined angles, An image signal processing method characterized by determining an autocorrelation function or a spatial frequency power spectrum of each of these projected images with respect to the coordinates on the line segment.