JPS6280769A - Internal distortion method for unevenly spaced image distortion considering aperture characteristics - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to image correction processing, and is particularly concerned with blurring caused by the imaging system (deterioration due to aperture characteristics), misalignment of the imaging system (detection element, etc.), or continuous correction due to vibration. The present invention relates to an interpolation method for unevenly spaced image distortion that takes into account aperture characteristics suitable for correcting images that have distortions with uneven pixel spacing.

[Background of the invention]

The latest sensor, H installed on SPOTI 7 stars
R V (}Iigh Resolution Vi
FIG. 2 shows the configuration of the sensor. CCD
(Charge Transfer Element) The number of elements in the sensor is approximately 1000 elements, and if the resolution of one element is 15m, the imaging width of one CCD is as narrow as 15km, so the installed sensor is as shown in Figure 2. Four CCDs are connected to a prism 10 which has been processed with a half mirror to obtain an imaging width of 60 km. The reflected light 20 from the ground surface is separated by a half mirror and is incident on a light receiving surface (element section) 50 of the CCD 40 .

The bonding surfaces of multiple CCDs are connected to one CC as shown in Figure 3.
In D, the elements 60 are arranged at equal intervals, but it is not possible to arrange the CCDs at completely equal intervals (mechanical alignment of 10 μm or less is required. Also, there may be problems due to shocks during launch, etc.). High possibility of matching) CCDI (3) and CC
Overlap of elements as between D (2}.70 (90)
- This is called overlapping -1 or CCD2 and C
Missing elements like between CD3 - called underlap -
, both of which produce a distortion called non-uniformly spaced distortion.

Unequal interval distortion is distortion that does not occur in conventional MSS (multispectral scanners).

Similar problems exist in medical images as well as satellite images.

Figure 9 shows XgCT (Computed Tomogr
aphy) device. X-rays emitted from the radiation source 300 pass through the subject 310 and enter the detector.

A projected image 340 is obtained.

The detector has a certain aperture angle due to the collimator to improve the resolution. The size of one detector is about 0.21 and the aperture is about 8°.

The detection section is composed of a total of 600 detectors in which several to several tens of detectors are connected in series. The detector array within one unit can be equally spaced and of equal size, but there is a problem.

Since the detector 1320 and the detector 11330 are connected by soldering or the like, the detector arrangement becomes discontinuous (underlap). (Similarly between detectors m, rv, and v) For this reason, discontinuous distortion between the detectors is included in the projection image data, and a ``ring artifact'' occurs in the reconstructed image, resulting in deterioration of image quality.

In addition to the above-mentioned distortions, image users desire higher quality images that are corrected for blurring and the like.

As shown in FIG. 4, this corresponds to the target scene I(x)14
Data of 0 indicates the optical system 150 such as collimator lens and mirror.
, the detector 160, the analog electronic circuit 170 such as an amplifier, and the sampling electronic circuit 180 are degraded, and the observed image sample data P(x) 200 is blurred.

As shown in FIG. 5, if the imaging system is an ideal δ (delta) function, the observed image sample data P(x) is.

It can be expressed as However, the shooting system is H (=Ho-Ho・H^
・Hs) (assuming that spatial invariance is preserved), the actual observed image sample data P' (x) P' (x) = Σ I (jΔX) - H ( X)-δ
(x-jΔX)j=-■ = Σ I(jΔx)・h(x+JΔx)
(2) j=-ω where: Convolution h(x, jΔx)=H*δ(x-jΔx)
It can be expressed as (3).

The impulse response of H in equation (3), that is, the observed data of the imaging system H when a point light source-like signal is input, and h is
P S F (Point 5pread Func
tion). PSF indicates the deterioration factor of the system. Therefore, in order to correct the deterioration of the imaging system, find the PSF [1
! Restoration processing 190 using a filter with a characteristic of H÷ for I data P(x) 200 may be performed. The restoration process is
Since PSF is called aperture characteristic, it is also called aperture characteristic correction.

In conventional image processing, the opening time correction is performed on the corrected image as shown in "Image Processing and Analysis" (Kyoritsu Shuppan) p. 192, and interpolation is already performed using an existing function. , it cannot be said that optimal aperture characteristic correction has been performed.

Furthermore, since spatial invariance is not preserved in unevenly spaced image distortion, it is difficult to apply to this method a PSF obtained on the assumption that spatial invariance is preserved.

To correct uneven image distortion by considering the aperture characteristics, w
After passing the measured image data through an aperture characteristic compensation filter, the pixel intensity at the point of interest may be determined using an unequal interval interpolation function. However, as mentioned above, the aperture characteristic compensation filter is calculated from the PSF based on spatial invariance, so this way of thinking cannot be applied to both cases where spatial invariance is not preserved in the scanning amount and has nonuniformly spaced distortion. It cannot be applied as is.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide a method for interpolating image blur and irregularly spaced image distortion caused by the imaging system at high speed and with high precision.

[Summary of the invention]

Considering the aperture characteristics of the imaging system, and correcting the nonuniform image distortion, after passing the observed image data through an aperture characteristic compensation filter, the pixel intensity of the point of interest is calculated using a nonuniform interpolation function. All you have to do is ask for. However, the aperture characteristic compensation filter is based on the Po1nt 5 pread Function (P
Since the calculation is based on SF), the above concept cannot be applied as is to images with non-uniform spacing in which spatial invariance is not preserved between detection elements or in the amount of scanning. For this reason, we invented the following method.

The concept of the method of the present invention consists of three steps.

Regarding notation: As shown in Figure 6, the interpolated value x(
xj) is a weighting function (interpolation function) found by appropriately weighting the intensities I (X) of multiple adjacent points
Assuming W, I(xj) can be expressed by the following equation, where u=XJ

Step 1: Unequally spaced sample data P(y 1) 26
Obtain evenly spaced sample data I umbrella (xi) 270 from 0 4 Evenly spaced sample data l5 (xi) 270 is obtained by convolution of nonuniformly spaced sample data P (yi) 260 and nonuniformly spaced interpolation function 200WNE(x) Seek.

Here, the unequal interval interpolation function WNE (x) 260 is a linear or cubic spline function that corresponds to the unequal interval quantity.

I-(xi)=ΣWNE(xt-ya)P(yi)
(5) Step 2: Equally spaced sample data I (
xi) 270 and the aperture characteristic compensation filter F(x) to obtain equally spaced sample data I (xi) in consideration of the aperture characteristics. (In FIG. 7, step 2 and step 3 are combined and the aperture characteristic compensation interpolation 280
It's Toshide. ) Here, the aperture characteristic compensation filter F (x) is based on the characteristics of the actually photographed image, for example, P S F (Point 5 prea
dFunction) and is designed as an inverse filter that compensates for the PSF.

Step 3: Estimated sample data I(xi) 290 is determined by convolution of uniformly spaced sample data I(xi) with aperture characteristic compensation and uniformly spaced interpolation function W CC (x).

The feature of the present invention is that the calculation speed is improved and the interpolation function table capacity is reduced by combining three steps into one step as described below.

Substitute equation (6) into equation (7).

Here, WA P (x I x a) = Σ WCC (xt-
xk) F (xlxt) Substituting equation (5) into equation (8): 1 Here, the estimated sample data ■290 is the first type of interpolation function W and ml ill! It can be obtained from the I sample data 260.

In order to obtain an estimated image of one point, it is necessary to prepare a table of 6740 interpolation coefficients and perform addition and multiplication 100 times if the calculation is performed for each step.

However, in this method, the number of calculations is 6612 and 11, respectively, which has the effect of significantly reducing the number of calculations.

(Computation speed is improved by 10 times.) [Embodiments of the invention] An embodiment of the present invention will be described below.

This embodiment is a processing system that simultaneously corrects the uneven image distortion occurring in the 5POT satellite HRV image described in FIG. 3 and the deterioration type in the imaging system.

FIG. 1 shows an overview of the HRV image shape distortion correction processing system.

First of all, there is an equal interval interpolation function 2 (actually a table value, the same applies hereafter), an unequal interval interpolation function 3 that does not take the aperture characteristics into account.
Using the aperture characteristic compensation interpolation function 4 and the aperture characteristic compensation interpolation function 4, an unequal interval interpolation function 5 that takes the aperture characteristics into consideration is determined in an interpolation function calculation process 1.

Wrap fitD is defined as shown in FIG. P-1゜P
Pixels o and Pl correspond to the rightmost three pixels of CCD2 in FIG. 3, and P4 e P 3+ Pg correspond to the leftmost three pixels of CCD3. That is, it is assumed that there is an underlap between CCD2 and CCD3 with a magnitude of l<D≦2. Therefore, in one scan, CCD2 and CC
When underlap occurs during D3 and shape distortion and blur correction is performed, interpolation processing becomes difficult because spatial invariance characteristics are not maintained. For this reason, (1
Using equation 1), an unequal interval interpolation function W that takes into account the aperture characteristics that allows interpolation of pixels for which spatial invariance is not maintained is determined as follows.

First, evenly spaced sample data E0 to I+, 160 are obtained from nonuniformly spaced sample data Po=Ps150 using a uniformly spaced interpolation function and an unevenly spaced interpolation function.

In other words, in interval 1 and interval 3, evenly spaced interpolation 180 interval 2
Then, non-uniform interpolation 190 is applied.

Io= Po (1
2) Iz=P (1
3) I2=P (1
4) I3=SUNEz(2)Po+WNEz(2)Pt
+WNEa (2) P2+ IIINE number (2) Pa+W
NEδ(2)Pa NEo(2)Pas (15)
Ia=WNEz(3)Po+WNE2(3)Pt+SN
Ea(3)P2-+4NEa(3)Pa-11NEg(
3) P4-41NEe (3) Pas (16) Ia=l
ilCCz(2-D)Pa+WCCz(2-D)Pa
+WCCa(2-D)Ps 10VCCa(2-D)Pg
(17) Next, considering the range of X, calculate the intensity value Q at point x 290 using the equally spaced sample data 270 and the aperture characteristic compensation interpolation function.

(i) O<x≦1 Q (x) =WAPt(x)Io+WAPz(x)
I t+WAPa(x) I z+APa(x) I
s (18) In equation (18), (12
) to (17) to find the interpolation coefficient W.

=W @P (19)(
ii) 1<x≦2 Similarly, Q(x)=WAPt(x-2)L1+APZ(X-1
) I z+1ilAPi(x-]) I s+WAP
a(x-1)I 4
From (20), (in)2<x≦3 Similarly, Q (x) =WAPt(x-2)Iz+1jAPz(
x-2) I s+APa(x-2) I 4+WA
Pa (x-2) r From 11 (22), the term indicated by -1 here is the coefficient of pe, but it is clear that there is no sudden change between adjacent pixels in the satellite image, and that P3
．． Considering that the weighting of Pa is about 10-2 of the weighting of Pl, it is approximated as Pa:Pa.

(fv)3<x≦2−D Similarly, Q (x)=WAPt(x−3)Ia+1IAP2(X
-3)I 4+WAPa(x-3)I R+WAPa(
x-3) I e (24) Next, the amount of shape distortion occurring on the uncorrected image 6 is calculated in advance by the shape distortion amount calculation process 13 using satellite system parameters (e.g. orbit, attitude data), and the shape superposition information file is created. Store it in 12.

Using this amount of shape distortion, the uncorrected image 6 is interpolated in an interpolation process 9 to determine the pixel intensity at the correct position and obtain a corrected image 11.

Here, the portion where the uneven amount occurs is calculated by the uneven amount calculation process using the amount of shape distortion in the shape distortion amount information file 12, and the portion where the uneven amount is generated is calculated by using the amount of shape distortion in the shape distortion amount information file 12, and the portion where the amount of uneven space is generated is Equally spaced interpolation coefficients are obtained in interpolation coefficient selection process 8, and pixel positions with uneven intervals are interpolated in interpolation process 9, and corrected image 1 is obtained.
Get 1.

Interpolation processing 9 involves performing a sum-of-products calculation of the interpolation function (coefficient) as shown in equation (18) and surrounding pixel intensities, using observation sample data P260 at 6 points and nonuniformly spaced samples jtD (internal). Interpolation) By using the interpolation function W corresponding to the position X, estimated sample data 290 of the point of interest can be obtained.

When the interpolation accuracy of the method of the present invention was evaluated using the pattern of the following formula, a high accuracy value of 0.5 quantization level or less was obtained in the range of underlap 2.0 pixels to overlap 1.5 pixels. .

5(x)=127.5(cos(2πx-f)+1)
(26) f: Frequency component of pattern For example, high-quality reconstructed images without linking ring artifacts can be obtained.

〔Effect of the invention〕

According to the present invention, the following effects can be obtained.

■ It becomes possible to apply an aperture characteristic compensation filter determined based on spatial invariance to images with irregularly spaced quantities in which spatial invariance is not preserved.

(3) Three-step processing can be reduced to one step, reducing the interpolation function table capacity by 14% of all interpolation coefficients and improving calculation speed by ten times.

■ (The test observation pattern is set to a dynamic range of 255
(Evaluation using trigonometric functions that change at the quantum level)
Interpolation accuracy of 0.5 indentation level can be obtained with non-uniform spacing (±2 pixels).

[Brief explanation of drawings]

Figure 1 is an HRV image, a schematic diagram of the uneven interval distortion correction system,
Figure 2 is a CCD arrangement diagram in the HRV sensor, Figure 3 is a cable arrangement diagram, Figure 4 is a diagram of the deterioration and restoration process of the imaging system, and Figure 5 is the aperture characteristics of the ideal imaging system and the actual imaging system. An overview diagram, Figure 6 is an explanatory diagram of the interpolation method, and Figure 7 considers the aperture characteristics!
FIG. 8 is an explanatory diagram of an example of interpolation coefficient calculation, and FIG. 9 is a schematic diagram of x Pl; Ac'r system v1. Agent: Patent attorney HIyl Ogawa. Rod 1 Old No. 2 Mouth Broom 3r: 3 Houki-+ Zuo S Zutei 4 Kusashi-111E No. 7 (2) Rod B Mouth

Claims (1)

[Claims] In an image distortion correction processing method that calculates the shape correction amount using a shape distortion amount table, an interpolation coefficient table, and these table values, and calculates the point of interest by multiplying and calculating the observed image data and the interpolation coefficient value. , considering the aperture characteristics, which is characterized by comprising an interpolation coefficient calculation process that simultaneously corrects the aperture characteristics and non-uniform interval image distortion, and an interpolation process that performs optimal interpolation according to the non-uniform interval occurrence area and the non-uniform interval amount. An interpolation method for unevenly spaced image distortion.