JP5219901B2 - Image processing device - Google Patents

Description

  This invention is an index for capturing minute changes occurring on the ground surface with high accuracy from radar images obtained by observing the same region multiple times at different times using radar mounted on an aircraft or an artificial satellite. The present invention relates to an image processing apparatus that estimates a complex correlation (coherence) distribution to be used with high accuracy using a plurality of coherence maps having different resolutions.

  CCD (Coherent Change Detection) is one of the techniques for extracting changes occurring on the ground surface from a plurality of radar images having different acquisition times. Since the CCD performs change extraction using phase information in addition to the luminance information of the radar image, the CCD has a feature that the sensitivity to the change is higher than the change extraction technique based only on the luminance change.

  A complex correlation (coherence) between two radar images is used as an index for extracting a change area in the CCD. This coherence has a property of taking a value close to 1 in a region where there is no change in two radar images and showing a value close to 0 in a region where there is a change.

  Using this property, there is a method of extracting a difference (change) between two radar images by performing threshold processing based on the strength of a coherence distribution (coherence map) (for example, Non-Patent Document 1). reference).

  In addition, in order to accurately estimate the coherence between two radar images, a local region for estimating the coherence (hereinafter referred to as a window) is divided into preset change shapes, and the same attribute as the target pixel due to its similarity There is a technique of selecting a pixel in the area, reconstructing a window with the selected pixel, and estimating coherence using the reconstructed window (see, for example, Patent Document 1).

  Furthermore, in order to accurately estimate the coherence between two radar images, there is a technique for estimating the coherence of a target pixel by performing maximum likelihood estimation processing on a coherence value estimated using a window having a preset size. Yes (for example, see Non-Patent Document 2).

JP 2008-310391 A

Preiss, M, Gray, D. and Stacy, N. "A change detection technique for repeat pass interferometric SAR." IEEE Geoscience and Remote Sensing Sympoium Proceesingd, 2003. IGARASS'03, vol.2, pp.21-25, 2003 . Touzi. R, Lopes. A, Bruniquel. J, and Vachon. P.W, "Coherence estimation for SAR imagery," IEEE Transactions on Geoscience and Remote Sensing, vol. 37, pp.135-149, 1999.

  When the change extraction processing by the CCD is performed with high accuracy, it is necessary to obtain the coherence between the two radar images with high accuracy. However, since the coherence between the two radar images is given by the ensemble average, the actual processing is performed. In this case, it is necessary to estimate this. As a general coherence estimation method, there is a method of estimating from a data included in a window composed of a pixel of interest and its neighboring pixels by a spatial averaging process as shown in the following equation (1).

Here, Z _n (y, x) represents the data value of the position (y, x) of the n-th radar image, and Z _{n + 1} (y, x) represents the position (y, x) of the n + 1-th radar image. x) indicates the data value possessed. Further, l is the number of pixels used for calculating the coherence of the target pixel (y, x), and corresponds to a range for calculating the coherence. * Indicates a complex conjugate. Coherence takes a value close to “1” when there is no change in the ground surface under the assumption that the two radar images used for coherence calculation are acquired in the same trajectory. It is known to take a small value.

  In the estimation of coherence by spatial averaging, there is a conflict between the spatial resolution of the coherence map and the estimation accuracy of the coherence, and it is difficult to obtain a high-precision and high-resolution coherence map. Therefore, in order to obtain high coherence estimation accuracy and high spatial resolution of the coherence map, the conventional image processing apparatus described in Patent Document 1 assumes a shape of a change that has occurred on the ground surface in advance, and the shape The coherence was estimated while adaptively changing the window shape to estimate the coherence.

  However, in the conventional image processing apparatus described in Patent Document 1, since the shape of the change that has occurred on the ground surface is assumed in advance, if the shape of the change is a complicated shape, it is The shape of the constructed window did not correspond to the shape of the change that occurred on the ground surface, and as a result, there was a problem that the estimation accuracy of coherence was lowered.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to obtain an image processing apparatus capable of capturing changes occurring on the ground surface with high accuracy.

An image processing apparatus according to the present invention includes a data storage unit that stores a plurality of radar images obtained by observing the same region having different acquisition times, and selects two radar images from the plurality of radar images. Coherence calculation processing means for calculating a plurality of coherence maps, which are complex correlation distributions of radar images, with different window sizes, and the likelihood of the joint probability density function of the coherence of the pixel of interest in the obtained coherence maps with different resolutions. Coherence maximum likelihood estimation processing means for performing maximum likelihood estimation processing for estimating true coherence of the target pixel using a degree function, and output storage means for storing the estimated coherence result.

  According to the present invention, the maximum likelihood estimation process is performed from the coherence values estimated at the respective resolutions of the target pixel based on the complex correlation distribution between the plurality of radar images having different resolutions. Can be obtained with high accuracy.

It is a block block diagram which shows the structure of the image processing apparatus by Embodiment 1 of this invention. It is a conceptual diagram of each coherence map estimated with a different window size. It is a conceptual diagram of a joint probability density function when D is a parameter with true coherence. It is explanatory drawing of a likelihood function. It is a block block diagram which shows the structure of the image processing apparatus in Embodiment 2 of this invention. It is explanatory drawing of a likelihood function in case an attention pixel is located in the vicinity of the boundary which has two different coherence. It is a block block diagram which shows the structure of the image processing apparatus in Embodiment 3 of this invention. It is explanatory drawing of the likelihood function at the time of changing the combination of windows when an attention pixel is located in a uniform area | region. It is explanatory drawing of the likelihood function at the time of changing the combination of windows when the attention pixel is located in the vicinity of the boundary which has two different coherences. It is a block block diagram which shows the structure of the image processing apparatus in Embodiment 4 of this invention. It is explanatory drawing explaining the data acquisition method of the area | region comprised with an attention pixel and its surrounding pixel.

  Hereinafter, preferred embodiments of an image processing apparatus of the present invention will be described with reference to the drawings.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration of an image processing apparatus according to Embodiment 1 of the present invention. In the first embodiment, a ground surface SAR image obtained by a synthetic aperture radar (SAR) will be described as an example of a radar image. However, the present invention is limited to a SAR image. However, the type is not limited as long as it is similar to this. In addition, although the SAR image to be used is described as two sheets, a plurality of pieces of data may be used.

  Hereinafter, the configuration will be described with reference to FIG. The image processing apparatus according to the first embodiment shown in FIG. 1 stores a data storage unit 1 that stores two SAR images obtained by observing the same region twice at different acquisition times, and stores them in the data storage unit 1. The coherence calculation unit 10 that calculates the coherence of the pixel of interest based on the above-described equation (1) using L windows having different sizes from the two SAR images that have been generated, and the L calculated by the coherence calculation unit 10 The likelihood function calculation unit 20 that calculates the likelihood function of the joint probability density function of the pixel of interest using the coherence D as a parameter from the number of coherences, and the coherence of the pixel of interest calculated in the likelihood function calculation unit 20 are different. Estimate the true coherence of the pixel of interest by calculating the coherence that maximizes the likelihood function of the combined probability density function obtained when estimating with the size window. To include the maximum likelihood estimation processing section 30, and an output storage section 2 for storing the coherence of the pixel of interest calculated by the maximum likelihood estimation processing unit 40. Note that the likelihood function calculation unit 20 and the maximum likelihood estimation processing unit 30 constitute coherence maximum likelihood estimation processing means for performing maximum likelihood estimation processing from the value of the coherence of the target pixel in each coherence map having different resolutions.

  Next, the operation will be described. The data storage unit 1 stores complex data of the ground surface in each observation region obtained by performing two observations with different acquisition times on the same observation region.

Based on the two SAR images stored in the data storage unit 1, the coherence calculation unit 10 generates two images from the data included in the L windows having different sizes composed of the pixel of interest and its neighboring pixels. L coherences d ₁ , d ₂ ,..., D _L related to the target pixel on the SAR image are calculated based on the equation (1).

The conceptual diagram is shown in FIG. A coherence map 1 in FIG. 2 shows a coherence map estimated with the smallest window size. The coherence map 2 is estimated with a window larger than the coherence map 1 and the window with the largest coherence map L is estimated. This is a coherence map. The coherence of the pixel of interest in these 1 to L coherence maps is d ₁ , d ₂ ,..., D _L.

  Note that L coherences calculated using windows of different sizes may be calculated in advance and stored in the data storage unit 1, and in this case, the coherence calculation unit 10 may be omitted. .

In the likelihood function calculation unit 20, the joint probability density function p (d ₁ , d ₂ ,. ..., A likelihood function based on d _L | D) is calculated. Here, the description is performed using the likelihood function. However, as long as the probability density function is a parameter in which the value of coherence estimated using windows of different sizes with the true coherence D as a parameter is used as a parameter, such as a posteriori probability distribution. Anything can be used.

FIG. 3 is a conceptual diagram showing, by contour lines, the joint probability density distribution when coherence is estimated using two windows of different sizes for a pixel of interest whose true coherence is D. If the coherence values estimated in two different size windows are d ₂ and d ₃ , respectively, the joint probability when data with true coherence D is estimated in two different window sizes Density is the intersection of two coherences in the figure.

Thus, the likelihood function shown in FIG. 4 can be obtained by calculating the joint probability density located at the intersection of two coherences for each D. Here, the likelihood function lik (D) = p (d ₁ , d ₂ ,..., D _L | D). Note that it is possible to omit calculating the likelihood function by calculating the joint probability density function or likelihood function calculated in advance for each D and the size of each window and storing the function in a database or the like. This makes it possible to increase the processing speed (not shown).

  The likelihood function of the joint probability density function at the target pixel calculated by the likelihood function calculation unit 20 has the maximum likelihood when the coherence D is true as shown in FIG. Therefore, the maximum likelihood estimation processing unit 30 can estimate the true coherence D at the target pixel with high accuracy by calculating the coherence D having the maximum likelihood based on the equation (2).

  The output storage unit 2 stores the result calculated in the first embodiment and displays it on a display unit (not shown), for example.

  Therefore, according to the first embodiment, by performing the maximum likelihood estimation process from the coherence value estimated at each resolution of the pixel of interest based on the complex correlation distribution between a plurality of radar images having different resolutions, The true coherence of the pixel of interest can be obtained with high accuracy, and changes occurring on the ground surface can be captured with high accuracy.

Embodiment 2. FIG.
In the first embodiment, the maximum likelihood estimation process is performed from the coherence values estimated by the L different size windows set in advance for the target pixel, and the coherence value of the target pixel is estimated. When the size of the window that estimates coherence increases, the estimation accuracy of the coherence estimation result decreases near the boundary between regions with different coherence. It is thought that the accuracy of the results also decreases.

  When the vicinity of the boundary is estimated with the size of a large window, it is considered that the likelihood of the joint probability density is low for data in which two regions having different coherence are mixed. In the second embodiment, using this property, a threshold process set in advance for the likelihood is performed, and when the likelihood is equal to or smaller than the threshold, the coherence value estimated with a certain window size is obtained. Let it be the coherence of the pixel of interest.

  FIG. 5 is a block diagram showing the structure of the image processing apparatus according to Embodiment 1 of the present invention. Note that portions that are the same as or equivalent to those in FIG. 1 are denoted by the same or corresponding reference numerals, and detailed description thereof is omitted.

  The image processing apparatus according to the second embodiment shown in FIG. 5 is different from the configuration of the first embodiment shown in FIG. 1 in that the coherence estimated by different L windows in the target pixel calculated by the likelihood function calculation unit 20. In the likelihood function of the joint probability density, the maximum likelihood calculating unit 31 that calculates the maximum likelihood is compared with the maximum likelihood of the target pixel calculated in the maximum likelihood calculating unit 31 and a preset threshold value. A likelihood determination processing unit 40 for determining whether the pixel of interest is a pixel located in the vicinity of a region having different coherence, and a coherence output for outputting a coherence value at the pixel of interest estimated with a preset window size And a unit 50.

  Next, the operation will be described. The maximum likelihood calculating unit 31 calculates the maximum likelihood in the likelihood function related to the joint probability density function of L coherences regarding the target pixel calculated with different window sizes, based on Expression (3).

  Note that max [f (x)] is an operator that calculates the maximum value of the function f (x).

  The likelihood determination processing unit 40 compares and determines the maximum likelihood Lik calculated by the maximum likelihood calculation unit 31 and a preset threshold value based on the equation (4). For example, when the pixel of interest is a pixel located near the boundary having two different coherences, if the estimated window is small, it is unlikely that data having the coherence of these two boundaries is mixed, but the size of the window When the length is increased, the coherence is estimated from data in which data having different coherences are mixed. In this case, as shown in FIG. 6, the obtained likelihood function is considered to show a value having a lower likelihood compared to the case where the present process is performed on a uniform region. Using this property, it is possible to determine whether the pixel of interest is a pixel located near the boundary having two different coherences.

  The coherence output unit 50 outputs the coherence of the pixel of interest estimated with a preset window size. For example, when it is determined by the likelihood determination processing unit 40 that the pixel of interest is a pixel located near the boundary of a region having different coherence, the value of coherence estimated by a small window is used. It is possible to reduce the influence of coherence estimation due to data in which data of two areas are mixed. In this example, the coherence value estimated using a preset small window is output. However, the user may select the optimum value, or the coherence estimated using several set window sizes. You may use the average value of.

  By performing threshold processing on likelihood, it becomes possible to avoid performing coherence estimation using data having two different coherences, and the accuracy of estimation of coherence near the boundary having two different coherences can be avoided. It is possible to prevent the decrease.

Embodiment 3 FIG.
In the second embodiment, when a pixel of interest is determined using likelihood as an index indicating that it is a pixel located near the boundary of a region having different coherence, and is determined to be a pixel located near the boundary Outputs a coherence estimated with a preset window size, thereby preventing a decrease in coherence estimation accuracy near the boundary.

  However, if the threshold value for determining whether the target pixel is a pixel located near the boundary, the size of the coherence window to be output, etc. are set in advance, the performance will be obtained if the set value does not satisfy the performance. May not be enough. In Embodiment 3 of the present invention, a plurality of joint probability densities of coherence estimated using windows of different sizes are prepared in different combinations, and the optimum one is selected from the combinations, and the coherence at the target pixel is determined by maximum likelihood estimation processing. presume.

  FIG. 7 is a block diagram showing the structure of the image processing apparatus according to Embodiment 3 of the present invention. 1 and 5 are denoted by the same or corresponding reference numerals, and detailed description thereof is omitted. Hereinafter, the function of each block diagram shown in FIG. 7 will be described.

  The image processing apparatus according to the third embodiment shown in FIG. 7 is different from the configuration of the first embodiment shown in FIG. 1 in terms of L different window sizes used for coherence estimation in the coherence calculation unit 10. A combination selection unit 60 for setting a combination probability density function of coherence estimated by different window sizes calculated by the likelihood function calculation unit 20 and a combination of the likelihood functions, and an attention calculated by the likelihood function calculation unit 20 In the likelihood function of the joint probability density of the coherence estimated by different L windows in the pixel, the maximum likelihood calculation unit 31 for calculating the maximum likelihood and the window obtained with a combination of different size windows, respectively. A maximum likelihood calculation processing unit 32 that calculates the maximum likelihood from among the maximum likelihoods of the maximum likelihood calculation unit, and the likelihood function calculation unit 20 and the maximum likelihood calculation unit 31 are combined with a window. Only the number of combinations is provided.

Next, the operation will be described. The combination selection unit 60 calculates a joint probability density function and its likelihood function in the likelihood function calculation unit 20 for the coherence estimated by the L different windows calculated in the coherence calculation unit 10. Set a combination of S windows. For example, when the number of combinations is two, there are combinations such as L ₁ = {1,2}, L ₂ = {3, 4}, L _S = {L−1, L}. Here, data having two adjacent L's is used, but this combination method is not limited to this, and any combination may be used. Further, the number of combinations is not limited to two.

  The maximum likelihood calculation processing unit 32 uses the S maximum likelihoods in each combination calculated by the likelihood function calculation unit 20 and the maximum likelihood calculation unit 31 based on the S combinations set by the combination selection unit 60. The one with the maximum likelihood is calculated, and the optimal window combination likelihood function is output.

  FIG. 8 and FIG. 9 are conceptual diagrams showing the relationship between window combinations and likelihoods obtained when the sizes of the windows to be combined are sequentially increased. FIG. 8 assumes a region where the coherence is uniform, and FIG. 9 assumes a case where a pixel of interest exists near the boundary between regions having different coherence. 8 and 9, the size of the window used for the combination increases as the size of the combination of the windows on the horizontal axis increases.

  In the case of the uniform region shown in FIG. 8, the likelihood calculated from a combination with a small window size tends to be larger than the likelihood estimated with a small combination. On the other hand, in the pixel located near the boundary as shown in FIG. 9, when the window combination is increased, the likelihood is the maximum value when the combination of the window sizes is the maximum of the combinations in which the data of the two regions are not mixed. The likelihood tends to decrease when it is larger than that. Therefore, the window combination that maximizes the likelihood is the optimal window combination, and true coherence at the pixel of interest can be obtained by performing maximum likelihood estimation processing.

  In this case, the maximum likelihood estimation processing unit 30 can be omitted by calculating the true coherence D having the maximum likelihood. Further, in the second embodiment, when the maximum likelihood calculated by the maximum likelihood calculation processing unit 32 is equal to or less than a preset threshold value, a coherence value calculated under a preset condition is output. It is also possible to combine with these methods.

Embodiment 4 FIG.
In Embodiments 1 to 3 described above, the joint probability density function is calculated using L different coherences for the target pixel, and the true coherence is estimated from the likelihood function. Next, in Embodiment 4 of the present invention, a joint probability density function is calculated using data of pixels included in a certain region formed by the target pixel and surrounding pixels.

  FIG. 10 is a block diagram showing the structure of an image processing apparatus according to Embodiment 4 of the present invention. 1, FIG. 5, and FIG. 7 are denoted by the same or corresponding reference numerals, and detailed description thereof is omitted. Hereinafter, the function of each block diagram shown in FIG. 10 will be described.

The image processing apparatus according to the fourth embodiment shown in FIG. 10 is an area selection that selects data included in the size of the pixel of interest and the surrounding area set in advance with respect to the configuration of the first embodiment shown in FIG. The maximum likelihood estimation processing by the likelihood function calculation unit 20 and the maximum likelihood estimation processing unit 30 is performed using data of pixels included in the region selected by the region selection unit 70.
It is.

  Next, the operation will be described. As shown in the conceptual diagram shown in FIG. 11, the region selection unit 70 selects a pixel of interest set in advance and pixels included in the surrounding region. Thereby, in the uniform region, it is possible to improve the estimation accuracy of the maximum likelihood estimation process.

  1 data storage unit, 2 output storage unit, 10 coherence calculation unit, 20 likelihood function calculation unit, 30 maximum likelihood estimation processing unit, 31 maximum likelihood calculation unit, 32 maximum likelihood calculation processing unit, 40 likelihood determination processing unit , 50 coherence output unit, 60 combination selection unit, 70 region selection unit.

Claims (7)

Data storage means for storing a plurality of radar images obtained by observing the same area with different acquisition times;
Coherence calculation processing means for selecting two radar images from a plurality of radar images and calculating a plurality of coherence maps, each of which is a complex correlation distribution of the two radar images, with different window sizes;
Coherence maximum likelihood estimation processing means for performing maximum likelihood estimation processing for estimating the true coherence of the pixel of interest using the likelihood function of the joint probability density function of the coherence of the pixel of interest in each of the obtained coherence maps with different resolutions;
An image processing apparatus comprising output storage means for storing the estimated coherence result. The image processing apparatus according to claim 1.
The coherence maximum likelihood estimation processing means includes:
A likelihood function calculating means for calculating a likelihood function of a joint probability density function of coherence estimated with different window sizes;
An image processing apparatus, comprising: a maximum likelihood estimation calculating unit that calculates a coherence having a maximum likelihood for the obtained likelihood function. The image processing apparatus according to claim 2,
The coherence maximum likelihood estimation processing means includes:
Maximum likelihood calculating means for calculating the maximum likelihood for the likelihood function calculated by the likelihood function calculating means;
A likelihood determination processing means for comparing the maximum likelihood calculated by the maximum likelihood calculating means with a predetermined threshold and determining whether or not the pixel of interest is a pixel located in a region having a different coherence;
An image processing apparatus further comprising coherence output means for outputting a coherence value obtained under a preset condition in accordance with a determination result by the likelihood determination processing means. The image processing apparatus according to claim 1.
A combination selection unit for selecting a combination of different window sizes for calculating the likelihood function;
The coherence maximum likelihood estimation processing means includes:
A plurality of likelihood function calculating means for calculating a likelihood function of a joint probability density function of coherence estimated by a combination of different window sizes;
Maximum likelihood calculating means for calculating a maximum likelihood for the likelihood functions respectively calculated by the plurality of likelihood function calculating means;
Maximum likelihood calculation processing means for comparing the maximum likelihood values in each combination and selecting the combination with the maximum likelihood,
An image processing apparatus comprising: a maximum likelihood estimation processing unit configured to estimate coherence of a pixel of interest by maximum likelihood estimation processing using a likelihood function calculated from a combination having maximum likelihood. In the image processing device according to any one of claims 1 to 4,
A region selection unit that selects a fixed region composed of a pixel of interest and neighboring pixels around it in each coherence map with different resolutions obtained by the coherence calculation processing unit;
The image processing apparatus, wherein the coherence maximum likelihood estimation processing means performs maximum likelihood estimation processing using data of pixels included in an area selected by the area selection means. Select two radar images from multiple radar images observed in the same area with different acquisition times, and calculate multiple coherence maps, which are complex correlation distributions of the two radar images, for each of the different window sizes. A database for storing the results obtained in advance;
Coherence maximum likelihood estimation processing means for performing maximum likelihood estimation processing for estimating the true coherence of the pixel of interest using the likelihood function of the joint probability density function of the coherence of the pixel of interest in each of the obtained coherence maps with different resolutions;
An image processing apparatus comprising: an output storage processing unit that stores an estimated coherence result. Select two radar images from multiple radar images observed in the same area with different acquisition times, and calculate multiple coherence maps, which are complex correlation distributions of the two radar images, for each of the different window sizes. A database for storing a result of calculating a likelihood function of a joint probability density function of coherence when coherence is estimated using windows of different sizes;
Maximum likelihood estimation calculating means for estimating the true coherence of the pixel of interest by the maximum likelihood estimation process using the likelihood function of the calculated joint probability density function;
An image processing apparatus comprising: an output storage processing unit that stores an estimated coherence result.

Images