JPH10243292A - Method and device for correcting dispersion in ir fpa - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve precision of calibrated data and to prevent a scene component from being intruded in the calibrated data by sampling a local area of a size required to generate the calibrated data, evaluating a degree of an amount of the scene component included in a distribution of output levels of elements in the area and generating the calibrated data, while reducing the size of the local area depending on the magnitude of the evaluation result. SOLUTION: Results of calculation by each of size mean value calculation sections 20-1-20-n in a calibration data generating section 2, that is a part for conducting real time correction processing in a sensitivity correction circuit, are given to a size selection section 21. On the other hand, each calculation result by area homogeneous degree calculation sections 30-1-30-n in an area homogeneous degree evaluation section 3 is compared with each threshold value of a size-depending homogeneous degree threshold value setting memory 31 at each of comparator sections 32, a proper size decision section 33 decides a proper size, and the result is outputted to a size selection section 21 as size information. Then one of proper sizes is selected, based on this information and a corresponding local area mean value is fed to an adder 22.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for correcting variations in the level of an output signal of an infrared detector, in particular, a two-dimensional IRFPA (Infrared Focal Plane Array).

[0002] An infrared detector for detecting a temperature distribution is used in a guidance device or the like, and a two-dimensional IRFPA suitable for performing target detection by a two-dimensional infrared image has been developed. When acquiring infrared image data of a scene (scene) using a two-dimensional IRFPA infrared detector, each point of the scene is electrically scanned by two-dimensionally arranged infrared detecting elements constituting the detector. An output signal corresponding to the infrared intensity is generated, and an infrared image is formed. However, in the two-dimensional IRFPA infrared detector, the input / output characteristics of each detection element still vary, and as a result, the output signal from each element varies even if an infrared input with a uniform intensity enters each element. Such variations in the output signal of each element are caused by fixed pattern noise (FPN: Fixed Patt
ern Noise), which together with the temporary noise that indicates the temporal fluctuation of the output signal, is a factor of noise that degrades the quality of the infrared image.

In particular, in a two-dimensional IRFPA infrared detector excluding some low-sensitivity types, the FPN is a dominant factor in image quality, and it is desired to effectively remove the FPN.

[0004]

2. Description of the Related Art In order to reduce FPN (fixed pattern noise), a signal processing method for making the input / output characteristics of elements uniform is generally called sensitivity correction or variation correction.

There are two factors in the variation of the input / output characteristics for each element: a variation in static characteristics that does not basically change with time, and a variation that indicates a change with time. For elements that do not change over time, calibration data for sensitivity correction can be precisely created at the stage of manufacturing test, and the data is stored in ROM.
(Read Only Memory) and read it out to calculate sensitivity correction. On the other hand, as for the element indicating the change with time, the variation factor increases or decreases due to the change in the operating condition of the IRFPA or the change in the scene input level. Therefore, it is necessary to perform the sensitivity correction in real time during use. In the real-time sensitivity correction, how to create calibration data, that is, in the presence of an input to each element from a scene that changes every moment, the characteristic variation of each element is corrected without changing the scene signal as much as possible. Therefore, how to create calibration data is an important issue.

FIG. 8 is a block diagram of an infrared sensor to which the present invention is applied. 8, reference numeral 80 denotes an infrared sensor; 81, an optical system including a lens; 82, an IRFPA in which infrared detecting elements are arranged in a two-dimensional plane;
A gimbal control unit that drives a portion including the RFPA to perform operations such as up / down, left / right, and rotation, and 84 is an IRFPA
An A / D converter 82 converts an output signal (analog signal) of each element into a digital signal. Reference numeral 85 denotes a sensitivity correction circuit.
50 is a multi-point correction unit, 851 is a calibration ROM, 852 is a real-time correction processing unit, 853 is an offset correction unit, 8
Reference numeral 54 denotes a calibration data calculation unit, and 86 denotes a target extraction unit.

The portion including the optical system 81 and the IRFPA 82 controls the gimbal control unit 8 to move vertically, horizontally, and rotationally toward a scene in a specific direction for detecting or tracking a target.
3 is performed. In the example of the figure, time 1 to time 4
Shows how the scene is rotated and scanned in response to the lapse of time. In addition, as a scene from which a target is extracted by the infrared sensor, there are various scenes such as a sky including clouds and flying objects, a sea surface including waves, and a ground including artificial objects.

The IRFPA 82 is arranged on a plane (for example, 1
Each of the 25 infrared detecting elements (25 elements × 125 elements) outputs a signal corresponding to the intensity of each infrared ray input from the optical system. The output signal of each element has a DC component corresponding to the absolute value of the infrared radiation level from the scene, while a signal difference of 1/1000 or less of this DC component is used to identify the detailed information of the scene. Need to be disassembled. However, the responsivity is, for example, 0.
If there is a 1% variation, the "signal difference" due to this variation is equivalent to 1/1000 of the scene signal difference to be resolved. IRF
It is not easy to keep the variation of PA from 0.1% or less to each element, and therefore, the sensitivity correction described below is performed in an infrared sensor using IRFPA.

The output signal of the IRFPA 82 (corresponding to the absolute value of the infrared radiation level) is converted to a digital signal by an A / D converter 84 and supplied to a sensitivity correction circuit 85. The sensitivity correction circuit 85 first corrects the static sensitivity variation, which is the variation that does not change with time described above, in order to correct the static sensitivity variation.
Using the OM 851, output data corresponding to each element is corrected by the multi-point correction unit 850, and the data subjected to the static correction is then corrected by the real-time correction processing unit 852 for an element of variation showing a temporal change. Real-time correction is performed. The contents of this real-time correction processing are shown in FIG. 9 described later, and the multi-point correction unit 85 is calculated based on the calibration data obtained by performing the calculation in the calibration data calculation unit 854.
The data from 0 is corrected by the offset correction unit 853, infrared image data is output, and a target is extracted from the data by the target extraction unit 86.

FIG. 9 shows the configuration of a conventional real-time correction process, and the method shown here is known to be the most effective. In FIG. 9, reference numeral 90 denotes a smoothing unit, and reference numeral 96 denotes a calibration data creation unit. The configuration combining these 90 and 96 corresponds to the calibration data calculation unit 854 in FIG.
Reference numeral 9 corresponds to the offset correction unit 853 in FIG.

In the method shown in FIG. 9, when attention is paid to an arbitrary element, and when the output level of the element is different from the output level of a nearby (peripheral) element, the difference is considered to be due to the variation in element characteristics. The difference between the output level of the element and the average value of the output of the element in the vicinity local area is defined as calibration data.

More specifically, the multi-point correction unit 85 shown in FIG.
0 at multiple points of the pixel number i of frame number f is a static result of the correction is performed correction data d _i (f) and is input to the smoothing unit 90 adder 91, a divider 92, prior to the adder 93 frame subtraction and (frame number f-1) of the smoothed data C _i of the same pixel number i (f-1), division, are added, the pixel number i of the smoothed frame number f data C _i ( f) is obtained and stored in the corresponding position of the frame memory 94 for the frame number f. The contents of the frame memory 94 for the frame number f indicate that the frame number f
-1 is transferred to the frame memory 95.

The smoothing section 90 smoothes the fluctuation of the output signal corresponding to the elapse of a predetermined time (smoothing time constant T) of each element to an average level during the predetermined time. This is because the difference between the smoothed data up to the previous frame and the output signal of the same element in the current frame is obtained by the adder 91, the difference is divided by the time T in the divider 92, and the result of the division is calculated by the previous frame. Adder 9 to the smoothed data up to
3, the smoothed data of the current frame is obtained, and the smoothed data is stored in the memory 9 for the f frame.
4 is stored. As the smoothing time constant T at this time, a time of a plurality of frames, for example, a time of 32 frames is used.

In this way, the variation (FPN) of each pixel caused by the change of the scene (change over time)
Absorb. On the other hand, in the calibration data creation unit 96, the pixel number i of the frame number (f-1) is stored in the frame memory 95.
When the smoothed data C _i (f-1) is input, performs averaging by the smoothing data of the element surrounding the pixel number i in the local averaging unit 97 (corresponding to a pixel). This averaging will be described with reference to FIG.

FIG. Of the target pixel number i (corresponding to the element i) among a number of elements (125 × 125 in this example) constituting the IRFPA shown in FIG.
B. As shown in FIG. 3, 3 × 3 (=
9) pixels are extracted. Each of the pixels (elements) in the local area is assigned a number of i-4,..., I,.

FIG. , The smoothed data of the eight pixels except for the pixel i of the attention area of the local area is extracted and added, and the result of addition is divided by 8 to obtain the average value of the local area. Local average data,
That is, the output r _i (f−1) of the local averaging unit 97 in FIG. 9 is obtained. By subtracting the smoothed data d _i (f-1) of the pixel of interest from the frame memory 95 from the local average data r _i (f-1) of the frame f-1 by the operation unit 98 in FIG. Is obtained as calibration data R _i (f). The calibration data by adding a multi-point correction data d _i (f) in the arithmetic unit 99, the correction of both static correction and real-time correction has been performed corrected output data D _i (f) is obtained.

The corrected output data D _i (f) and the calibration data R _i
(f), local average data r _i (f-1) and averaged data C _i
Each operation expression of (f-1) can be expressed by the following expressions.
Here, n is the local area size (number of elements), and n = 9 in the example of FIG.

[0018]

(Equation 1)

In the method shown in FIG. 9, the calibration data is created by using the output signal smoothed on the time axis without using the output signal itself from the detection element. In the direct output signal, the output level distribution of each sensing element is dominated by the component corresponding to the infrared intensity distribution of the input scene (referred to as scene component), but in the output signal after smoothing, the scene moves not stationary but moves. If it is, or if the scene is moved by scanning the visual axis of the IRFPA, the output signal from the scene is attenuated by the smoothing operation, and the level distribution of the output signal for each sensing element is mainly FP.
This is due to N (fixed pattern noise) components.

[0020]

In the above-mentioned conventional method, since the local local area when generating the calibration data is one type and is fixed, the accuracy of the calibration data and the reproducibility of the scene components are second. However, there was a problem as described in.

In order to improve the accuracy of the calibration data, it may be necessary to take a wide area as a local area in the vicinity and to create the calibration data based on the average value of the output levels of more elements. The reason is that when the average value is obtained for a wide local region, the average value variation for each local region is small, but when the local region is narrow and the number of elements to be averaged is small, the average value variation for each local region is large. It is because it becomes. That is, when the output levels of the elements are randomly distributed, if the average value of the output levels of n elements is taken, the variation in the average value decreases in proportion to the square root of n.

On the other hand, when the range of the local region for which the calibration data is created is widened, the distribution of the output level of the elements in the region becomes larger in the distribution of the scene component than in the FPN component. As a result, the influence of the scene component remains in the calibration data created based on the average value of the output levels of the elements in the area. Since the calibration data is created using the smoothed signal instead of the element output signal itself, the scene components remaining in the calibration data are different from the actual scene information, so the scene information is mixed. When the correction operation is performed using the corrected calibration data, false information different from the actual scene information is mixed in the correction output.

If the time constant of the smoothing is increased, the scene components are sufficiently attenuated and it is possible to prevent the scene components from being mixed in the calibration data. The time difference (delay) from the output signal increases. As a result, there is a problem that the created calibration data does not match as calibration data for correcting the output signal of the sensing element at that time.

The present invention solves the above-mentioned problems, and realizes both a method of correcting variation in IRFPA and a method of correcting variation in IRFPA, which can realize both high precision (high precision correction) of calibration data and prevention of mixing of scene components into calibration data. An object of the present invention is to provide an apparatus for performing variation correction.

[0025]

SUMMARY OF THE INVENTION The present invention first takes a local region of a size necessary to generate sufficiently high-precision calibration data and, for the local region of that size, distributes the output level of elements in the region. Is evaluated, and if the scene components are small enough to be ignored, calibration data is created using a local region of that size. If the scene components are large enough to be ignored, the local region Is reduced to such a degree that the scene components can be ignored, and calibration data is created using the reduced local region.

FIG. 1 is a diagram showing the principle of the present invention. The principle configuration of FIG. 1 shows only the configuration for performing the real-time correction processing according to the present invention in the sensitivity correction circuit (85 in FIG. 8), and the input thereof is the same as the conventional configuration for correcting the variation in static characteristics that does not change with time. A multipoint correction unit (85 in FIG. 8) not shown
0) is the multipoint correction data corrected.

In FIG. 1, reference numeral 1 denotes a smoothing unit (90 in FIG. 9) having a configuration similar to the conventional one, 2 denotes a calibration data creation unit,
2a is an average value calculating means for each size corresponding area, 2b is a size selecting means, 2c is an adding means, 3 is an area homogeneity evaluation unit,
3a is each size corresponding homogeneity calculating means, 3b is each size corresponding homogeneity threshold storage means, 3c is evaluation means for comparing the homogeneity corresponding to each size with the corresponding threshold to determine an appropriate size, 3d is storage means, Reference numeral 4 denotes a correction unit.

In the present invention, image data obtained by temporally smoothing the multi-point correction data in the smoothing unit 1 is created, and a plurality of sizes of local regions (for example, 3 ×) are obtained for the smoothed data of each pixel. The average value (average value excluding the pixel at the center of interest) within 3, 5, 5 x 5, 7 x 7, etc.) is calculated by each size corresponding region average value calculation means 2a.

On the other hand, in the area homogeneity evaluation section 3, the area homogeneity in the local area corresponding to a plurality of sizes for each element is obtained by the size homogeneity calculating means 3a. This calculation is to calculate the variance of the signal level in the local area, and to calculate the variance, standard deviation or equivalent statistic of the signal level in the local area.
(Variation unique to the element) ² + (variation in scene) ² The scene variation is large when the local region includes a scene having a complicated structure such as an edge portion of the scene, and when the scene is uniform, there is no change in the scene. Small variation.

Further, a set value of each threshold value in consideration of a variation (FPN) unique to an element corresponding to a local region of each size is set in advance in each size corresponding threshold value storage means 3b. The evaluation means 3c is a homogeneity calculating means 3a corresponding to each size.
The homogeneity in the local area corresponding to each size obtained in the above is sequentially compared with the set value of the threshold corresponding to each size taken out from the size corresponding threshold storage unit 3b. Determines that the variation in the scene components is too large, and compares the value of the homogeneity corresponding to the smaller local region size with the set value of the corresponding threshold. Thus, if the value of homogeneity is smaller than the set value,
The size at that time is determined as the optimal local area size for obtaining the calibration data of the element, and the size is stored in the storage unit 3d. The same processing is performed for each element, and the optimum size of the local area for each pixel of the frame is stored in the storage unit 3d.

In the calibration data creation unit 2, the local average value corresponding to various sizes is obtained for the smoothed data of each element of the frame, and the size selection unit 2b stores the optimum local area size corresponding to each element. The local average value r _i (f−1) corresponding to the size extracted from the means 3d is selected and supplied to the adding means 2c. The difference between the peripheral is obtained as calibration data R _i (f) by subtracting the value of the adding means 2c in smoothed data d _i (f-1) from the local average value r _i (f-1). The calibration data by adding a multi-point correction data d _i (f) in the correction unit 4 a, static correction and real-time correction has been performed corrected output data D _i (f) is obtained.

Although the optimum size of the local area can be found by the above-described principle, if the process for finding the local area takes a long time, the delay until the generation of the calibration data may be too large. In order to cope with this, the size of the local region should be set so that at least two types of local regions, large and small, can be applied, and the degree of the scene components included in the distribution of the output levels of the elements in the region for the large-sized local region first. (Dispersion, standard deviation, etc.) is evaluated. If the scene component is small enough to be ignored, calibration data is created using the large-sized local region. If the scene component is large enough to be ignored, small. Calibration data is created using a local region of size.

As the evaluation of the scene component, the variance or standard deviation of the output level of each element or a statistic substantially equivalent thereto is calculated, and the calculated value is larger than a preset threshold. When the scene component is large, and when the scene component is small, the following method is available in addition to the method of making the local region appropriate.

The local area is divided into a plurality of sub-areas including at least four sides of the area, the average value or the sum of the output signal levels of the elements in each sub-area is calculated, and the maximum value of the calculated value in the area is calculated. When the difference between the threshold value and the minimum value is larger than a preset threshold value, the scene component is determined to be large.

[0035]

FIG. 2 is a hardware configuration diagram of the first embodiment, which is a configuration for realizing the above-described principle of FIG. 1 by hardware. In FIG. 2, reference numerals 1 to 4 correspond to the same reference numerals in FIG. 1, respectively. Reference numeral 1 denotes a smoothing unit, which has the same configuration as the conventional smoothing unit (90 in FIG. 9). The parts denoted by reference numerals 14 to 14 are 91 to 9 in FIG.
Same as 5, 10 is an adder, 11 is a divider, 12 is an adder, 13 is a frame memory composed of a RAM for storing an image for f frame, and 14 is a RAM for f-1 frame. It is a configured frame memory.

Reference numeral 2 denotes a calibration data creation unit, and 20-1 and 20-1
20-2,..., 20-n are average value calculation units of each size for calculating a local average value in a local area corresponding to each size of size 1, size 2,. ), 21 is a size selector provided with a switch mechanism (corresponding to 2b in FIG. 1), 22 is an adder (corresponding to 2c in FIG. 1)
It is. However, the size 1 of the local area is 3 elements × 3 elements,
Size 2 is 5 elements × 5 elements,..., Size n is (2n +
1) Element × (2n × 1) element.

Reference numeral 3 denotes a region homogeneity evaluation unit, and 30-1 and 30-
2, ..., 30-n are size 1, size 2, ...,
A region homogeneity calculation unit for each size (corresponding to 3a in FIG. 1) for calculating a region homogeneity in a local region corresponding to each size of size n, and 31 is a homogeneity corresponding to each of the local regions of sizes 1 to n. A size-based homogeneity threshold setting memory (corresponding to 3b in FIG. 1) in which a degree threshold is stored in advance, and is constituted by a ROM. Reference numeral 32 denotes a number of comparators (n) for comparing the calculation results from the region homogeneity calculators 30-1 to 30-n corresponding to each size with thresholds corresponding to the sizes from the size homogeneity threshold setting memory 31. (Comparative 1 to comparative n), and 33 is an appropriate size determining unit for determining an appropriate size from the comparison results of the plurality of comparators from the comparing unit 32. And the appropriate size determination unit 33 corresponds to the evaluation unit 3c in FIG.
Reference numeral 34 denotes a memory (corresponding to 3d in FIG. 1) for storing an appropriate size determined for each element for each element.
M. Reference numeral 4 denotes a correction unit constituted by an adder.

In the configuration shown in FIG.
Multipoint correction data d by the same configuration as_ifor (f)
And the signal C smoothed for each element i_i(f-1)
Output from the frame memory 14. Signal C_i(f-1) is the school
In the positive data creation unit 2, each size of each element (i)
Of each size in the average value calculation units 20-1 to 20-n
The average value in the local area is calculated in parallel. Each local area size
Is input to the size selection unit 21 and the area
One is selected based on the appropriate size information from the homogeneity evaluation unit 3.
Of the local area corresponding to the selected and output appropriate size
The average value is supplied to the adder 22. With this adder 22,
The smoothed signal C of the element from the average value_i(f-1) is subtracted
And the calibration data R_i(f) is output. Output school
Positive data R _i(f) is supplied to the adder 4 where the multipoint correction data
D_iCalibration data R from (f)_i(f) is subtracted and corrected
Output data D_i(f) is output.

In the region homogeneity evaluation unit 3, the region homogeneity calculation units 30-1 to 30-n of each size calculate the region homogeneity corresponding to the size of each local region in parallel.
In this embodiment, a standard deviation (variation) or a variance of the signal level in the local region is obtained as the region homogeneity.

When evaluating a scene component in a local region of each size, the scene component varies depending on the use environment and use conditions, and it is difficult to know the value in advance. Therefore, when determining the scene component, F
A preset value of the PN component is used. In this case, the preset value is the FP when the use condition of IRFPA and the use environment change.
The setting is made in consideration of the fluctuation of the N component.

FIG. 3 shows the relationship between the infrared input level and the variation in the element output level (FPN) when the size of the local area is 11 × 11 when the homogeneous infrared ray is input, and this relationship is obtained from experimental results. . The horizontal axis in FIG. 3 represents an input level (background temperature), the left side is low temperature, the right side is high temperature, and the vertical axis represents variation (FPN). From this data, the variation (FPN component) fluctuates with respect to the change in the scene temperature (background temperature). By using this experimental result, it is possible to set a threshold value for determining the magnitude of the distribution of the scene component.

A specific example of setting and using the homogeneity threshold value of each local region size will be described below. Uniform infrared rays are incident on each element of the IRFPA, and the output of each element at the time of homogeneous infrared input with respect to changes in the infrared input level (represented by the average value of the signal level of each element in the local area) using the local area size as a parameter The data of the level variation (FPN) is acquired.

FIG. 4 shows the variation (FPN) corresponding to each local area size when a homogeneous infrared ray is input. As shown in FIG. 4, the variation (FPN) of the element output signal varies depending on the infrared input level and also depends on the local region size to be evaluated.

4 is stored in the ROM forming the size-based homogeneity threshold setting memory 31 shown in FIG. At this time, the specified local area size and the average value of the signal level of each element in the local area at that size are set as addresses.

When evaluating the homogeneity in the local area for each size, ROM data is read using the local area size and the average signal level in the area as addresses. When the average signal level does not match the mark in FIG. 4, the nearest data is read and calculated by the proportional distribution method.

As described above, when the calculation results of the homogeneity of each local area are stored in the ROM forming the size-based homogeneity threshold setting memory 31 and the comparison unit 32 of FIG. The threshold corresponding to each size is read out and compared with the homogeneity of the local region of the corresponding size by each size-compatible comparator in the comparing unit 32. The comparison result is input to the appropriate size determination unit 33, and one size in which the comparison result indicating that the value of the homogeneity is smaller than the threshold value is determined (selected) as the appropriate size. If there are multiple selected sizes, select the largest size among them. The size of the local area thus selected is stored in the memory 34 as an appropriate size for the element.

In the calibration data creation unit 2, a certain element C _i (f
When calibrating -1), the size selection unit 21 takes out the appropriate size of the corresponding element from the memory 34, thereby obtaining the average value calculation units 20-1 to 20- of multiple sizes.
The calculation result corresponding to the appropriate size (local average data r _i (f) of the appropriate size) is selected from n and supplied to the adder 22. The adder 22 obtains calibration data (R _i (f)) by subtracting the smoothed data from the local average data, and the output is supplied to the adder 4. The adder 4 calculates the calibration data R _i (f) from the multipoint correction data d _i (f) of the element i.
Is output signal D _{i of} element i calibrated by subtracting
(f) is obtained.

FIG. 5 is a diagram showing a processing flow of the second embodiment, and realizes the above-mentioned principle configuration of FIG. 1 by software. In this case, the speed can be increased by using a CPU and a DSP (Digital Signal Processor) for signal processing as a processing device.

IRFPA detecting element of frame f (i)
Output the signal d _i is input (S1 in FIG. 5), the conventional example (FIG. 9) performs time smoothing for each element by preset time constant (T) by the same principle as in (Fig. 5 S2 ). Time-smoothed data of the f-th frame of each element is obtained, and the frame memory (same as 11 in FIG. 3)
(S3 in FIG. 5). The following processing is performed for the next i-th element (S4). That is, a local area A _i (n) having a size n with the element i as the center is extracted. In this case, n selects one of 3 × 3, 5 × 5, 7 × 7, 11 × 11,..., And sets the maximum to, for example, 11 × 11, and changes the area size to 9 × 11 (S8 described later). 9,7 × 7,
... and sequentially reduced. Next, the distribution (or variance) σ _{Ri (n)} of the output level of each element in the local region R _i (n) selected for the element i is calculated by the equation shown in FIG.
6). Next, it is determined whether σ _{Ri (n) as} a calculation result is equal to or smaller than σ _th which is a preset threshold (S7 in FIG. 5). This threshold value σ _th can be set in a memory (ROM) and read out in the same manner as the method of setting the homogeneity threshold value of each local area size described with reference to FIG.

In this determination, the calculated distribution σ_{Ri (n)}The value of the
Is the threshold σ_thIf it exceeds, change the area size n
(S8 in FIG. 5), the processing of S5 to S7 is repeated again. This
Is repeated to obtain the distribution σ_{Ri (n)}Is the threshold σ_thLess than
, The region R_iThe output signal level of each element in (n)
Average value (r_i) Is calculated by the formula shown in the figure.
(S9 in FIG. 5). Next, calibration data R for element i_iTo
The equation (r_i-D _i) (S1)
0), and stores the result in memory in association with element i
S11).

Next, it is determined whether or not the processing has been completed for all the elements (S12 in FIG. 5). If the processing has not been completed, i is increased (S13), and the processing of S4 and subsequent steps is executed. Is executed by the following equation (S in FIG. 5)
16). When the processing of all the elements is completed, it is determined whether the correction operation has been completed or not, based on the presence or absence of an external instruction to end the correction operation (S14 in FIG. 5). If not, f (frame number) is increased. (S15) For the next frame, the processing of S3 and subsequent steps is executed in the same manner as described above.

FIG. 6 shows measured values of the correction accuracy with respect to the size of the local area for the generation of the calibration data. This is the effect of reducing the FPN when the calibration data is created by changing the size of the local region and corrected using the calibration data, as described in the first embodiment (FIG. 2) and the second embodiment (FIG. 5). Is shown in the figure.

The horizontal axis in FIG. 6 is the size of the local area (number of elements), and the vertical axis is FPN [LSB (Least Significant) which is a unit of digital quantity representing the standard deviation (rms: root mean square) of the variation of the element output level. LSB represents millivolt in this example. In this example, 3 × 3 = 9 elements, 5 × 5 = 25 elements, 7 × 7
= 49 elements, 9 × 9 = 81 elements, and 11 × 11 = 121 elements, the measured values of FPN corresponding to each size are drawn. As can be seen from this, when the size is enlarged, the FPN can be reduced, the variation in calibration data can be reduced, and high accuracy can be achieved.

In the configurations of the first embodiment shown in FIG. 2 and the second embodiment shown in FIG. 5, the homogeneity (standard deviation, distribution or variance, etc.) of a large number (3 or more) of local regions is evaluated in real time correction. However, processing time for calculation and evaluation takes time, and target extraction may be delayed.
A third embodiment aiming at improving the processing speed will be described below.

FIG. 7 is a hardware configuration diagram of the third embodiment. In the third embodiment, in the evaluation of the region homogeneity, the variance is calculated for the size of one wide local region (for example, 9 × 9), and when the variance does not exceed the preset value,
The average of the wide local area is selected, and otherwise, the average of a predetermined narrow local area (for example, 3 × 3) is selected.

The reference numerals 1, 10 to 14, and 2, 3, and 4 in FIG. 7 have the same names as those of the corresponding reference numerals in FIG. 2 and will not be described. In the calibration data creation unit 2,
Numeral 3 denotes a narrow local average calculation unit for averaging local regions of a predetermined small size (for example, 3 × 3) for a target element, and 24 denotes a wide size (for example, 9) for a target element. × 9) is a wide local average calculating unit for averaging the local areas, 25 is a selecting unit, and 26 is an adding unit.

In the area homogeneity evaluation section 3, a variance calculation section 35 calculates a variance in a wide local area corresponding to a wide size (9 × 9) of the wide local average calculation section 24.
A comparison unit 6 compares a threshold value and a variance value in a preset wide local area.

In the third embodiment, the multi-point correction data d _i (f) input in the smoothing section 1
A similar operation is performed by the same configuration as in FIG. 2, and the smoothed data is stored in the frame memory (for f-frame) 13. The smoothed data of the frame memory 14 of the previous frame (f-1 frame) is input to the calibration data creation unit 2 and the region homogeneity evaluation unit 3. In the calibration data creating unit 2, the narrow local average calculating unit 23 and the wide local average calculating unit 24 determine the size, 3 × 3 and 9 ×, of the smoothed data C _i (f−1) of the target element i. The average of the nine local regions is calculated, and each calculation result is output to the selection unit 25.
On the other hand, in the area homogeneity evaluation unit 4, the variance of the size (9 × 9) of the wide local area is calculated by the wide local area variance calculating means 35. The calculation result is compared with a threshold value, which is a preset value, by the comparison unit 36, and an output indicating whether the value is equal to or smaller than the threshold value is generated in the calibration data creation unit 2.

When the selecting section 25 of the calibration data generating section 2 indicates that the output from the comparing section 36 is equal to or less than the threshold value, the selecting section 25 outputs the output (average value) r _i (f) of the wide local average calculating section 24.
-1) is supplied to the adder 26, and the adder 26 subtracts the value of the smoothed data C (f-1) corresponding to the element i to obtain calibration data R _i (f). The data is added to the correction unit is supplied to the fourth adder multipoint correction data d _i (f), corrects the output D _i (f) is generated.

In the case of the third embodiment, the size of the wide local area is set to 9 × 9 in the above description.
For example, 11 × 11 can be selected. The size of the narrow local region is 3 × 3, which is the minimum size.

Although the hardware configuration is shown in the third embodiment of FIG. 7, it can be realized by software processing. In that case, it is clear that the processing flow shown in FIG. 5 is simplified.

[0062]

According to the present invention, the following effects can be obtained. High-accuracy calibration data can be created for an area where scene components are homogeneous, for example, a clear sky background. As a result, the FPN component can be reduced, the quality of the infrared image data is improved, and the performance for target extraction, which is the purpose of use of the infrared sensor, is improved.

For a region where the scene components are non-homogeneous, for example, the sky where clouds are scattered, the wavy sea surface, or the ground where artificial objects are scattered, the scene structure itself is complicated.
Even if the IRFPA performance is ideal and the FPN component is 0, target extraction is not easy. Therefore, even if the size of the local region is reduced and the FPN is increased according to the present invention, the effect on the target extraction performance is the same for a homogeneous scene. Smaller than. Accordingly, the target extraction performance of the infrared sensor can be improved comprehensively.

[Brief description of the drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a hardware configuration diagram of the first embodiment.

FIG. 3 shows a local area size of 11 × 1 when a homogeneous infrared ray is input.
FIG. 3 is a diagram showing a relationship between an infrared input level and an element output level variation (FPN) in the case of 1.

FIG. 4 is a diagram illustrating a variation (FPN) in element output level corresponding to each local region size when a homogeneous infrared ray is input.

FIG. 5 is a diagram illustrating a processing flow according to a second embodiment.

FIG. 6 is a diagram showing measured values of correction accuracy with respect to the size of a local region for creating calibration data.

FIG. 7 is a hardware configuration diagram of a third embodiment.

FIG. 8 is a diagram showing a block configuration of an infrared sensor to which the present invention is applied.

FIG. 9 is a diagram showing a configuration of a conventional real-time correction process.

FIG. 10 is an explanatory diagram of averaging.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Smoothing part 2 Calibration data creation part 2a Average value calculation means in each area corresponding to each size 2b Size selection means 2c Addition means 3 Area homogeneity evaluation part 3a Uniformity calculation means for each size 3b Threshold storage means for each size 3c Evaluation means 3d storage means 4 correction unit

 ──────────────────────────────────────────────────の Continued on the front page (72) Inventor Hiroki Shimozen 1333 Uedanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Prefecture Inside Fujitsu System Integration Laboratories Co., Ltd. Fujitsu System Integration Laboratories

Claims (6)

[Claims] 1. A two-dimensional IR in which infrared detecting elements are arranged.
The output of each element of the FPA is included in a local area near the target element, centered on the target element, in a real-time correction processing section that corrects static variations by a multipoint correction section and then corrects dynamic variations. In the variation correction method of IRFPA for generating calibration data for correcting the output signal level of the target element using the time-smoothed output signal level of each element as a reference value, a plurality of different-sized local areas are used as the local area. For the local area, the distribution of the output signal level of each element included in each of the local areas is determined, and the distribution is evaluated by comparing the distribution with a preset threshold value corresponding to each size, and the distribution is set in advance. When the size of the local region smaller than the threshold value is detected, calibration data of the element is created using the reference value of the local region of that size. Variation correction method of the FPA. 2. The method according to claim 1, wherein the distribution of the output signal level of the local region of each size is obtained by calculating any of a variance or a standard deviation of each output signal or a statistical value equivalent thereto. An IRFPA characterized in that the evaluation is performed by comparing with a threshold value which is set in advance for each size and approximates a variation unique to an element.
Method for correcting variations. 3. The method according to claim 1, wherein the evaluation of the distribution in the local region of each size includes dividing the local region into a plurality of sub-regions, and averaging output signal levels of elements included in each of the sub-regions. Alternatively, a method of correcting variation in IRFPA, wherein the method is performed by using a difference between a maximum value and a minimum value of the sum. 4. The method according to claim 1, wherein two sizes, large and small, are used as the local region, and one of the sizes is selected by evaluating the distribution in the local region according to claim 2 or 3. IRFPA variation correction method. 5. A two-dimensional IR in which infrared detecting elements are arranged.
In an apparatus for performing IRFPA variation correction, comprising a multi-point correction unit for correcting static variations with respect to the output of each element of the FPA, and a real-time correction processing unit for correcting dynamic variations, the real-time correction processing unit includes a calibration unit. Calculating means for calculating a homogeneity in each of a plurality of local regions of different sizes with respect to the signal of the smoothed element, comprising a data creating unit and an area homogeneity evaluating unit; Evaluation means for comparing a plurality of outputs of the calculation means with a preset threshold value corresponding to each size to determine an appropriate size; and outputting the result determined by the evaluation means to the calibration data creation unit. To the size selection means of
The calibration data creation unit calculates a plurality of average values for a plurality of local regions of different sizes with respect to the signal of the smoothed element, and a plurality of outputs of the plurality of calculation units. A size selection unit for selecting one of the two, and an adder for subtracting an average value of the local area selected by the size selection unit from the signal of the smoothed element according to a result determined by the evaluation unit, An apparatus for performing IRFPA variation correction, wherein the multipoint correction data is corrected by an output of an adder of the calibration data creation unit. 6. The area homogeneity evaluation unit according to claim 5, wherein the area homogeneity evaluation unit is configured to calculate homogeneity of the size of a wide local area, and to calculate the calculated homogeneity in accordance with the size of the wide local area. A comparison means for comparing with a threshold value, wherein the calibration data creation unit calculates two average value calculation units for calculating an average value of the local area corresponding to the size of the narrow local area and the size of the wide and narrow local area; Selection means for selecting one of the outputs of the average value calculation unit, wherein the size selection means of the calibration data creation unit outputs the output of the average value calculation unit of one size based on the output of the comparison means of the area homogeneity evaluation unit. An apparatus for correcting variation of IRFPA, wherein