JP6104115B2 - Image processing device - Google Patents

Description

  The present invention relates to an image processing apparatus that extracts a white and high-brightness observation target (for example, a cloud) from an RGB image acquired using, for example, a satellite-mounted sensor.

For example, if the cloud amount can be grasped from a satellite image observed by an earth observation satellite, weather information can be obtained, and whether or not the ground surface and the atmosphere can be observed can be determined.
In particular, the cloud coverage can be used as an auxiliary function for the spectroradiometer mounted on the satellite, which enables control of the observation visual field direction and an index for analyzing desired data from a vast amount of data. .

For example, in Patent Document 1 below, by performing differentiation on the pixel value in the horizontal direction in the pixel data, the amount of change of the edge of the object existing in the pixel is calculated as the feature amount, An image processing apparatus is disclosed that identifies whether an object is a cloud or a cloud other than a feature amount.
However, the edge varies greatly depending on the type of cloud. For example, a cloud whose thickness changes depending on the position, such as cumulus clouds and cumulonimbus clouds, has a large edge change amount, and a cloud having a substantially uniform thickness, such as a stratocloud or stratocumulus cloud, has a small edge change amount.
On the other hand, since the outline of a building, which is a feature on the ground surface, is clear, the amount of change in the edge is large, and the amount of change in the edge is small in the sea surface or forest.

JP-A-5-333160 (FIG. 2)

  Since the conventional image processing apparatus is configured as described above, the amount of change in the edge of the cloud varies greatly depending on the type of cloud, and some of the features on the surface have the same amount of change in the edge of the cloud. Some have a degree of edge change. For this reason, in the cloud detection method using the change amount of the edge of the object existing in the pixel as an index, there is a problem that a feature on the ground surface may be erroneously detected as a cloud.

  The present invention has been made to solve the above-described problems, and an object thereof is to obtain an image processing apparatus capable of detecting an observation target with high accuracy.

  The image processing apparatus according to the present invention includes image data conversion means for converting image data having RGB values into image data in the HSV color space, and the color of each pixel in the image data in the HSV color space converted by the image data conversion means. Based on the quantity and brightness, the quantity estimation means for estimating the quantity of the observation target in each pixel, and whether the observation target exists in each pixel based on the quantity of the observation target in each pixel estimated by the quantity estimation means An observation target determination means for determining whether or not the observation target quantity calculation means has an observation target by the observation target determination means among the observation target quantities in each pixel estimated by the quantity estimation means. Only the quantity of the observation target in the determined pixel is integrated to calculate the quantity of the observation target of the entire image.

  According to this invention, the observation target determining means for determining whether or not the observation target exists in each pixel based on the quantity of the observation target in each pixel estimated by the quantity estimation means is provided, and the observation target quantity calculation is performed. The means integrates only the quantity of the observation target in the pixel determined to be present by the observation target determination means out of the quantity of the observation target in each pixel estimated by the quantity estimation means. Since the entire observation target quantity is calculated, there is an effect that the observation target can be detected with high accuracy.

1 is a configuration diagram illustrating an image processing apparatus according to Embodiment 1 of the present invention; It is explanatory drawing which shows SV distribution of a white and high-intensity cloud. It is explanatory drawing which shows SV distribution of a white and high-intensity cloud. It is a block diagram which shows the image processing apparatus by Embodiment 2 of this invention. It is a block diagram which shows the image processing apparatus by Embodiment 3 of this invention. It is a block diagram which shows the image processing apparatus by Embodiment 4 of this invention.

Embodiment 1 FIG.
1 is a block diagram showing an image processing apparatus according to Embodiment 1 of the present invention.
In FIG. 1, an image input unit 1 is composed of, for example, a receiving device that receives image data transmitted from an earth observation satellite or an input interface device that inputs image data from the outside, and has image data having RGB values. Execute the process of inputting.
The image data having RGB values may be image data of RGB images, but may be a hyperspectral image acquired by a multispectral sensor or the like, for example.

The HSV conversion unit 2 includes, for example, a semiconductor integrated circuit on which a CPU is mounted or a one-chip microcomputer, and the image data input by the image input unit 1 is converted into image data in an HSV (Hue Saturation Value) color space. Perform the process of converting to. The HSV conversion unit 2 constitutes image data conversion means.
The SV composite image generation unit 3 is composed of, for example, a semiconductor integrated circuit mounted with a CPU or a one-chip microcomputer, and the saturation of each pixel in the image data of the HSV color space converted by the HSV conversion unit 2 Using S (i, j) and lightness V (i, j), a process of estimating the number of observation objects (for example, cloud amount) Z (i, j) in each pixel is performed. Note that the SV composite image generation unit 3 constitutes a quantity estimation unit.

The observation target determination unit 4 is composed of, for example, a semiconductor integrated circuit on which a CPU is mounted, or a one-chip microcomputer. The observation target quantity Z (i, If j) is equal to or greater than a preset threshold value Z _th (for example, Z _th = 0), it is determined that an observation target exists at the pixel position (i, j), and the quantity Z (i, j) Is less than the threshold value _Zth , a process of determining that an observation target does not exist at the pixel position (i, j) is performed.
Further, the observation target determination unit 4 sets the mask M (i, j) for the pixel position (i, j) determined to have an observation target to “1”, and determines that the observation target does not exist. A process of setting the mask M (i, j) for the pixel position (i, j) set to “0” is performed. The observation target determination unit 4 constitutes an observation target determination unit.

The quantity calculation unit 5 includes, for example, a semiconductor integrated circuit on which a CPU is mounted, a one-chip microcomputer, or the like, and the quantity Z (i, j to be observed in each pixel estimated by the SV composite image generation unit 3. ), The number C of observation objects included in the entire image (for example, for example) is obtained by accumulating only the quantity Z (i, j) in the pixels determined to be present by the observation object determination unit 4. The processing for calculating the cloud amount of the observation target of the entire image is performed.
That is, after the quantity calculation unit 5 multiplies the quantity Z (i, j) estimated by the SV composite image generation unit 3 by the mask M (i, j) set by the observation target determination unit 4 for each pixel. Then, the multiplication result for each pixel is integrated, and the integration result is divided by the total number of pixels, thereby executing the process of calculating the quantity C to be observed included in the entire image. The quantity calculator 5 constitutes an observation target quantity calculator.
The quantity output unit 6 transmits the observation target quantity C included in the entire image calculated by the quantity calculation unit 5 to the outside, or the display apparatus displays the observation target quantity C included in the entire image on the display. Alternatively, it is composed of an output interface device that outputs the observation target quantity C included in the entire image to a storage device such as a hard disk.

In the example of FIG. 1, each of the image input unit 1, the HSV conversion unit 2, the SV composite image generation unit 3, the observation target determination unit 4, the quantity calculation unit 5, and the quantity output unit 6 that are components of the image processing apparatus is dedicated. However, the image processing apparatus may be configured with a computer.
When the image processing apparatus is configured by a computer, the processing contents of the image input unit 1, HSV conversion unit 2, SV composite image generation unit 3, observation target determination unit 4, quantity calculation unit 5, and quantity output unit 6 are described. May be stored in the memory of the computer, and the CPU of the computer may execute the program stored in the memory.

Next, the operation will be described.
In the first embodiment, an example in which a white and high-intensity cloud is an observation target will be described.
First, the image input unit 1 inputs image data having RGB values from the outside.
The image data only needs to have at least RGB values. For example, the image data may be RGB image data or hyperspectral image data.

When the image input unit 1 inputs image data having RGB values, the HSV conversion unit 2 converts the image data into image data in the HSV color space.
Hereinafter, the image data conversion processing by the HSV conversion unit 2 will be described in detail.
First, the HSV conversion unit 2 extracts R (i, j), which is an R value (a value corresponding to the Red band) at a pixel position (i, j), and a pixel position (i, j) from image data having RGB values. G (i, j) that is the G value (value corresponding to the Green band) and B (i, j) that is the B value (value corresponding to the Blue band) of the pixel position (i, j) is extracted.
Here, the values of R (i, j), G (i, j), and B (i, j) are normalized to a range of 0.0 to 1.0 using the bit depth of the image. Shall.

Next, as shown in the following formula (1), the HSV conversion unit 2 is the largest among R (i, j), G (i, j), and B (i, j) for each pixel. The value Max (i, j) and the smallest value Min (i, j) are specified.

ＨＳＶ変換部２は、最大値Ｍａｘ（ｉ，ｊ）がＧ（ｉ，ｊ）であれば、下記の式（３）に示すように、画素位置（ｉ，ｊ）のＢ（ｉ，ｊ）、Ｒ（ｉ，ｊ）、Ｍａｘ（ｉ，ｊ）及びＭｉｎ（ｉ，ｊ）を用いて、画素位置（ｉ，ｊ）の色相Ｈ（ｉ，ｊ）を算出する。

ＨＳＶ変換部２は、最大値Ｍａｘ（ｉ，ｊ）がＢ（ｉ，ｊ）であれば、下記の式（４）に示すように、画素位置（ｉ，ｊ）のＲ（ｉ，ｊ）、Ｇ（ｉ，ｊ）、Ｍａｘ（ｉ，ｊ）及びＭｉｎ（ｉ，ｊ）を用いて、画素位置（ｉ，ｊ）の色相Ｈ（ｉ，ｊ）を算出する。

式（２）〜（４）は、色相Ｈ（ｉ，ｊ）を３６０度の角度で表現した例であり、例えば、２πラジアンや、最大値１．０で規格化して表現してもよい。 When the HSV conversion unit 2 specifies the maximum value Max (i, j) and the minimum value Min (i, j) for each pixel, if the maximum value Max (i, j) is R (i, j), As shown in the following equation (2), using G (i, j), B (i, j), Max (i, j) and Min (i, j) at the pixel position (i, j), H (i, j) which is the hue (Hue) of the pixel position (i, j) is calculated.

If the maximum value Max (i, j) is G (i, j), the HSV conversion unit 2 uses B (i, j) at the pixel position (i, j) as shown in the following equation (3). , R (i, j), Max (i, j), and Min (i, j) are used to calculate the hue H (i, j) of the pixel position (i, j).

If the maximum value Max (i, j) is B (i, j), the HSV conversion unit 2 calculates R (i, j) of the pixel position (i, j) as shown in the following equation (4). , G (i, j), Max (i, j), and Min (i, j) are used to calculate the hue H (i, j) of the pixel position (i, j).

Expressions (2) to (4) are examples in which the hue H (i, j) is expressed by an angle of 360 degrees, and may be expressed by being normalized by 2π radians or a maximum value of 1.0, for example.

さらに、ＨＳＶ変換部２は、下記の式（６）に示すように、画素位置（ｉ，ｊ）の明度（Ｖａｌｕｅ）であるＶ（ｉ，ｊ）として、最大値Ｍａｘ（ｉ，ｊ）を設定する。

これにより、ＲＧＢ値を有する画像データがＨＳＶ色空間の画像データに変換されたことになるが、Ｒ（ｉ，ｊ）、Ｇ（ｉ，ｊ）、Ｂ（ｉ，ｊ）の値が０．０〜１．０の範囲に規格化されているので、彩度Ｓ（ｉ，ｊ）及び明度Ｖ（ｉ，ｊ）についても０．０〜１．０の範囲に規格化された値として取り扱うことができる。 Further, the HSV conversion unit 2 uses the maximum value Max (i, j) and the minimum value Min (i, j) of the pixel position (i, j) as shown in the following formula (5). S (i, j) which is the saturation of (i, j) is calculated.

Furthermore, the HSV conversion unit 2 sets the maximum value Max (i, j) as V (i, j) which is the brightness (Value) of the pixel position (i, j) as shown in the following equation (6). Set.

As a result, image data having RGB values is converted to image data in the HSV color space, but the values of R (i, j), G (i, j), and B (i, j) are 0. Since it is standardized in the range of 0 to 1.0, the saturation S (i, j) and lightness V (i, j) are also handled as values standardized in the range of 0.0 to 1.0. be able to.

When the HSV conversion unit 2 converts the image data having RGB values into image data in the HSV color space, the SV composite image generation unit 3 converts the saturation S (i, j) and brightness of each pixel in the image data in the HSV color space. Using V (i, j), the observation target quantity Z (i, j) in each pixel is estimated.
Hereinafter, the estimation process using the cloud amount Z (i, j) as an example of the quantity Z (i, j) by the SV composite image generation unit 3 will be specifically described.

FIG. 2 is an explanatory diagram showing the SV distribution of white and high-intensity clouds.
In FIG. 2, the circles represent the saturation S (i, j) and lightness V (i, j) of the cloud, and white and high-intensity clouds have a small saturation S (i, j). It can be seen that the brightness V (i, j) is large and there is a bias in the SV distribution.
Therefore, in FIG. 2, it is possible to estimate that the object existing on the upper left of the straight line Z ₁ is a cloud.
Therefore, the SV composite image generation unit 3 estimates the cloud amount Z (i, j) of each pixel using the quantity model equation of each pixel. For example, the SV composite image generation unit 3 uses the following equation (7) to estimate the cloud amount of each pixel. The cloud amount Z (i, j) is estimated.

In Expression (7), α ₁ , β ₁ , and Z ₁ are constants set in advance, and are set, for example, by conducting a simulation analysis of the SV characteristics of the cloud in advance.
As shown in the following equation (8), if Z ₁ capable of estimating that an object existing on the upper left side of the straight line Z ₁ is a cloud is set, the cloud amount Z (i, j) of each pixel is set. Is positive, it is considered appropriate to estimate that it is a cloud. On the other hand, if the cloud amount Z (i, j) of each pixel is negative, it is inappropriate to estimate that it is a cloud. it is conceivable that.

When the SV composite image generation unit 3 estimates the cloud amount Z (i, j) of each pixel, the observation target determination unit 4 compares the cloud amount Z (i, j) of each pixel with a preset threshold value _Zth .
In the first embodiment, it is assumed that Z _th = 0. However, the threshold Z _th is not limited to 0 as long as it is appropriately set according to the SV characteristic of the observation target. Therefore, the threshold value Z _th may be a value larger than 0 or a value smaller than 0.
The observation target determination unit 4 is an observation target at the pixel position (i, j) if the cloud amount Z (i, j) of each pixel is equal to or greater than the threshold Z _th (Z (i, j) ≧ Z _th ). It is determined that a cloud exists.
On the other hand, if the cloud amount Z (i, j) of each pixel is less than the threshold value _Zth (Z (i, j) < _Zth ), a cloud to be observed exists at the pixel position (i, j). Judge that there is no.

When the above-described determination process is performed, the observation target determination unit 4 performs a mask M (i, j) for the pixel position (i, j) determined to have a cloud as shown in the following equation (9). Is set to “1”, and the mask M (i, j) for the pixel position (i, j) determined to have no cloud is set to “0”.

式（１０）において、ｐｉｘｅｌは画像の全画素数であり、雲量Ｃは０．０〜１．０の値を取り得る。 When the observation target determination unit 4 performs the determination process, the quantity calculation unit 5 determines the observation target by the observation target determination unit 4 out of the cloud amount Z (i, j) of each pixel estimated by the SV composite image generation unit 3. The cloud amount C of the entire image is calculated by accumulating only the cloud amount Z (i, j) at the pixel determined to be present.
That is, the quantity calculation unit 5 estimates the mask M (i, j) set by the observation target determination unit 4 for each pixel by the SV composite image generation unit 3 as shown in the following equation (10). After multiplying the cloud amount Z (i, j), the multiplication results for each pixel are integrated, and the integration result is divided by the total number of pixels of the image, thereby calculating the cloud amount C of the entire image.

In Expression (10), pixel is the total number of pixels of the image, and the cloud amount C can take a value of 0.0 to 1.0.

For example, when the cloud is thick, the scattering cross section of the cloud increases, so the brightness V (i, j) increases and the cloud amount Z (i, j) increases. On the other hand, when ground reflected light is included in the reflected light from the cloud like a thin cloud, the saturation S (i, j) decreases due to the color of the ground being mixed, and the cloud amount Z (i, j) is reduced. descend. Thus, the cloud amount C of the entire image can be estimated using the magnitude of the cloud amount Z (i, j).
When the quantity calculation unit 5 calculates the cloud amount C of the entire image, the quantity output unit 6 outputs the cloud amount C of the entire image to the outside.

As apparent from the above, according to the first embodiment, if the cloud amount Z (i, j) of each pixel estimated by the SV composite image generation unit 3 is equal to or greater than a preset threshold value _Zth , the pixel If it is determined that a cloud to be observed exists at the position (i, j), and the cloud amount Z (i, j) is less than the threshold value _Zth , the pixel position (i, j) is the object to be observed. An observation target determination unit 4 that determines that a certain cloud does not exist is provided, and the quantity calculation unit 5 determines an observation target among the cloud amount Z (i, j) of each pixel estimated by the SV composite image generation unit 3. Since it is configured to calculate the cloud amount C of the entire image by accumulating only the cloud amount Z (i, j) in the pixels determined to be present by the unit 4, the observation target can be detected with high accuracy. There is an effect that a certain cloud can be detected.

In the first embodiment, it has been shown that the object existing at the upper left of the straight line Z ₁ is a cloud. However, by preparing a plurality of straight lines (in the example of FIG. 3, two straight lines are provided). Z ₁ and Z ₂ are prepared), and when the cloud amount Z (i, j) of each pixel satisfies all the conditions corresponding to each straight line, the observation target determining unit 4 determines that the cloud to be observed You may make it determine with existing in a pixel.
In the example of FIG. 3, it is determined that a cloud exists when it exists on the upper left side of the straight line Z ₁ and exists on the upper left side of the straight line Z ₂ . For this reason, in FIG. 2, an object having a low saturation S (i, j) and lightness V (i, j) (Δ in the upper left of the line Z _{1 in the} figure) is erroneously detected as a cloud. However, in FIG. 3, since such an object can be determined not to be observed, the identification accuracy of the observed object can be improved.

式（１１）において、ｋは１〜Ｎの自然数である。
あるいは、多数の直線に代わり、Ｖ（ｉ，ｊ）又はＳ（ｉ，ｊ）、あるいは、その両方に対する高次の関数として、各画素の雲量Ｚ_k（ｉ，ｊ）を推定してもよい。

また、観測対象判定部４が、各直線に対応する閾値毎に、観測対象である雲が各画素位置（ｉ，ｊ）に存在しているか否かを判定して、画素位置（ｉ，ｊ）に対するマスクＭ_k（ｉ，ｊ）を設定し、数量算出部５が、下記の式（１２）に示すように、画像全体の雲量Ｃを算出する。

As described above, when N straight lines are prepared, the SV composite image generation unit 3 estimates the cloud amount Z _k (i, j) of each pixel using the threshold value Z _k corresponding to each straight line.

In Expression (11), k is a natural number of 1 to N.
Alternatively, the cloud amount Z _k (i, j) of each pixel may be estimated as a high-order function for V (i, j) and / or S (i, j), or both, instead of multiple straight lines. .

In addition, the observation target determination unit 4 determines whether or not a cloud to be observed exists at each pixel position (i, j) for each threshold corresponding to each straight line, so that the pixel position (i, j ) For the mask M _k (i, j), and the quantity calculation unit 5 calculates the cloud amount C of the entire image as shown in the following equation (12).

In the first embodiment, an example in which a white and high-intensity cloud is an observation target has been described. However, if an object whose SV characteristics are known in advance, an object other than a cloud is an observation target. Also good.
Therefore, for example, a building that is a feature on the ground surface can be detected as an observation target.

Embodiment 2. FIG.
FIG. 4 is a block diagram showing an image processing apparatus according to Embodiment 2 of the present invention. In the figure, the same reference numerals as those in FIG.
The coverage calculation unit 7 is composed of, for example, a semiconductor integrated circuit mounted with a CPU or a one-chip microcomputer, and the observation target determination unit 4 determines that an observation target (for example, a cloud) exists. The number of pixels obtained is integrated, and the processing for calculating the coverage C ′ of the observation target for the entire image is performed from the integration result. Note that the coverage rate calculation unit 7 constitutes a coverage rate calculation means.

In the first embodiment, the quantity calculation unit 5 has the observation target determination unit 4 out of the observation target quantity Z (i, j) estimated by the SV composite image generation unit 3. In this example, only the quantity Z (i, j) in the pixels determined to be integrated is calculated to calculate the quantity C of the entire image. The number of pixels determined to be present may be integrated and the coverage C ′ of the observation target for the entire image may be calculated from the integration result.
That is, when the observation target determination unit 4 performs the determination process, the coverage ratio calculation unit 7 masks M (i,) set by the observation target determination unit 4 for each pixel as shown in the following equation (13). j) is integrated (the mask M (i, j) is “1” if the observation target exists, “0” if the observation target does not exist), and the integration result is divided by the total number of pixels pixel. Thus, the coverage C ′ of the observation target for the entire image is calculated.

The quantity C of the observation target included in the entire image calculated by the quantity calculation unit 5 is calculated by the coverage calculation unit 7 while the density of the observation target (for example, clouds) is quantified. The coverage C ′ of the observation target with respect to the entire image displayed indicates the presence / absence of the observation target, and does not indicate the density of the observation target. Therefore, the calculation speed can be increased as compared with the first embodiment.
The coverage C ′ of the observation target with respect to the entire image is effective when recognizing the presence or absence of the observation target. For example, when detecting a region with few clouds as the observation target, the cloud amount C indicating the density of the cloud More useful information.

Embodiment 3 FIG.
5 is a block diagram showing an image processing apparatus according to Embodiment 3 of the present invention. In the figure, the same reference numerals as those in FIG.
The atmospheric correction unit 11 is composed of, for example, a semiconductor integrated circuit on which a CPU is mounted, a one-chip microcomputer, or the like, and atmospheric propagation (for example, atmospheric scattering) included in the image data input by the image input unit 1 The process of correcting the image data by removing the offset component of each band (R value, G value, B value) due to the influence of atmospheric attenuation) and outputting the corrected image data to the HSV conversion unit 2 carry out. Note that the atmospheric correction unit 11 constitutes an image data correction unit.
In the image processing apparatus of FIG. 5, the example in which the atmospheric correction unit 11 is applied to the image processing apparatus of FIG. 1 is shown, but it may be applied to the image processing apparatus of FIG. 4.

Since it is the same as that of the said Embodiment 1, 2 except the point which mounts the air | atmosphere correction | amendment part 11, only the processing content of the air | atmosphere correction | amendment part 11 is demonstrated here.
Since the color of the observation target on the ground surface is affected by atmospheric scattering and atmospheric attenuation, the clouds output as satellite images are often bluish. Therefore, the original saturation of the observation target changes, which may cause a deterioration in the accuracy of the determination process in the observation target determination unit 4.
Therefore, when the atmospheric correction unit 11 receives image data having RGB values from the image input unit 1, each band (R value) due to the influence of atmospheric propagation (for example, atmospheric scattering, atmospheric attenuation) included in the image data. , G value, B value) is removed to correct the image data.
Specifically, the image data is corrected as follows.

When the atmospheric correction unit 11 receives image data having RGB values from the image input unit 1, the atmospheric correction unit 11 specifies an offset component of each band (R value, G value, B value) in the image data.
The method for specifying the offset component of each band is not particularly limited, but a method of specifying the smallest value as the offset component in each band is conceivable.
However, since the smallest value may be noise, a value that is several percent higher than the smallest value may be specified as the offset component.

When the atmospheric correction unit 11 specifies the offset component of each band (R value, G value, B value) in the image data, it subtracts the offset component of each band from the value of each band of the image data. The offset component of each band is removed.
The air correction unit 11 outputs the image data from which the offset component of each band is removed to the HSV conversion unit 2 as corrected image data.
Thereby, the HSV conversion part 2 can convert the image data in which the influence of the atmospheric scattered light is reduced into image data in the HSV color space.

  As apparent from the above, according to the third embodiment, the atmospheric correction unit 11 is influenced by atmospheric propagation (for example, atmospheric scattering, atmospheric attenuation) included in the image data input by the image input unit 1. Since the image data is corrected by removing the offset component of each band (R value, G value, B value) according to, the influence of atmospheric propagation is reduced, and the cloud to be observed is detected. There is an effect that the accuracy can be increased.

Embodiment 4 FIG.
FIG. 6 is a block diagram showing an image processing apparatus according to Embodiment 4 of the present invention. In the figure, the same reference numerals as those in FIG.
The exception determination unit 21 is configured by, for example, a semiconductor integrated circuit on which a CPU is mounted, a one-chip microcomputer, or the like. Among the image data in the HSV color space converted by the HSV conversion unit 2, an observation target (for example, A process of detecting a non-observation target (for example, a white artifact such as asphalt) whose SV characteristic (saturation / lightness characteristic) is similar to that of the cloud) is performed.
The exception area removal processing unit 22 is composed of, for example, a semiconductor integrated circuit on which a CPU is mounted, or a one-chip microcomputer. The exception determination unit 21 uses image data in the HSV color space converted by the HSV conversion unit 2. A process of removing the area where the detected non-observation target exists and outputting the image data after the area removal to the SV composite image generation unit 3 is performed.
The exception determination unit 21 and the exception area removal processing unit 22 constitute non-observation target removal means.
6 shows an example in which the exception determination unit 21 and the exception area removal processing unit 22 are applied to the image processing device in FIG. 5, but the image processing device is applied to the image processing device in FIGS. 1 and 4. It may be a thing.

Since the exception determination unit 21 and the exception area removal processing unit 22 are the same as in the third embodiment except that the exception determination unit 21 and the exception area removal processing unit 22 are implemented, only the processing contents of the exception determination unit 21 and the exception area removal processing unit 22 are described here. explain.
For example, a white artifact such as asphalt has a substantially constant spectral characteristic with visible light like a cloud, and the saturation S (i, j) has a similar value.
However, since the reflectance of the artifact is lower than that of the cloud, when the cloud is thick, the lightness V (i, j) is lower than that of the cloud. For this reason, the artifact and the cloud can be easily identified by the difference in the brightness V (i, j).
On the other hand, when the reflected light from the cloud includes ground reflected light as in the case of a thin cloud, the saturation S (i, j) decreases due to the color of the ground surface being mixed, and the brightness V (i, j) and The saturation S (i, j) is lowered. For this reason, it may be difficult to identify the artifact and the cloud even if the lightness V (i, j) and the saturation S (i, j) are referred to.

Therefore, the exception determination unit 21 performs non-observation targets (for example, white asphalt) whose SV characteristics are similar to those of the observation target clouds in the HSV color space image data as follows. Artifacts) are detected.
For example, the exception determination unit 21 generates an edge image for the brightness V (i, j) of each pixel in the image data of the HSV color space converted by the HSV conversion unit 2. Since the edge image generation process itself is a known technique, a detailed description thereof will be omitted.
Clouds do not have clear edges, but artifacts have clear edges with large changes in shadows and reflectivity with the ground. In addition, since artifacts are denser than small clouds, the number of edges increases.
Therefore, the exception determination unit 21 determines that the object is an artifact (non-observation target) if the edge strength or the edge quantity (or both the intensity and quantity) is higher than a preset threshold value. If the edge strength or the edge quantity (or both the intensity and quantity) is lower than the threshold value, it is determined that the object is a cloud (observation target).

Here, an example has been shown in which the exception determination unit 21 evaluates the strength of an edge or the number of edges to determine whether the object is an observation target or a non-observation target. The variation of the hue H (i, j) of each pixel in the data is calculated, and the variation of the hue H (i, j) (for example, the standard deviation of the hue H (i, j)) is evaluated to be an observation target. Or whether it is a non-observation target.
Clouds have a substantially constant hue (white to gray), whereas artifacts have a large change in hue due to differences in color from the ground surface and colors from adjacent artifacts.
Therefore, the exception determination unit 21 calculates the standard deviation of the hue H (i, j) of each pixel, and determines that it is an artifact (non-observation target) if the standard deviation is higher than a preset threshold. If the standard deviation is lower than the threshold value, it is determined to be a cloud (observation target).

When the exception determination unit 21 detects a non-observation target, the exception area removal processing unit 22 has a non-observation target detected by the exception determination unit 21 from the image data in the HSV color space converted by the HSV conversion unit 2. The removed region is removed, and the image data after the region removal is output to the SV composite image generation unit 3.
As a result, the non-observation target data whose SV characteristics are similar to those of the observation target cloud are removed from the image data in the HSV color space, so that the detection accuracy of the observation target cloud can be improved. There is an effect.

  In the present invention, within the scope of the invention, any combination of the embodiments, or any modification of any component in each embodiment, or omission of any component in each embodiment is possible. .

  DESCRIPTION OF SYMBOLS 1 Image input part, 2 HSV conversion part (image data conversion means), 3 SV synthetic image generation part (quantity estimation means), 4 Observation object determination part (observation object determination means), 5 Quantity calculation part (observation object quantity calculation means) ), 6 Quantity output unit, 7 Coverage rate calculation unit (coverage rate calculation unit), 11 Atmospheric correction unit (image data correction unit), 21 Exception determination unit (non-observation target removal unit), 22 Exception area removal processing unit (non- Observation target removal means).

Claims (4)

Image data conversion means for converting image data having RGB values into image data in the HSV color space;
Quantity estimation means for estimating the number of observation objects in each pixel using the saturation and brightness of each pixel in the image data of the HSV color space converted by the image data conversion means;
Observation target determination means for determining whether or not the observation target is present in each pixel based on the quantity of the observation target in each pixel estimated by the quantity estimation means;
Of the quantity of the observation target in each pixel estimated by the quantity estimation unit, by adding only the quantity of the observation target in the pixel determined by the observation target determination unit to be present. And an observation target quantity calculating means for calculating the quantity of the observation target of the entire image. Image data conversion means for converting image data having RGB values into image data in the HSV color space;
Quantity estimation means for estimating the number of observation objects in each pixel using the saturation and brightness of each pixel in the image data of the HSV color space converted by the image data conversion means;
Observation target determination means for determining whether or not the observation target is present in each pixel based on the quantity of the observation target in each pixel estimated by the quantity estimation means;
An image comprising: a coverage ratio calculating means for integrating the number of pixels determined to be present by the observation object determining means and calculating the coverage of the observation object with respect to the entire image from the integration result Processing equipment.   Image data correction means for correcting the image data by removing offset components contained in image data having RGB values and outputting the corrected image data to the image data conversion means is provided. The image processing apparatus according to claim 1 or 2. Among the image data in the HSV color space converted by the image data conversion means, a non-observation object having a saturation lightness characteristic similar to that of the observation object is detected, and the non-observation object exists from the image data A non-observation object removal means for removing the region that is
The quantity estimation means uses the saturation and brightness of each pixel in the image data from which the area where the non-observation target exists is removed by the non-observation target removal means to calculate the quantity of the observation target in each pixel. The image processing apparatus according to claim 1, wherein estimation is performed.

Images