JP2005071215A - Image normalization device, image normalization method, computer readable recording medium in which program is recorded and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To normalize a satellite image and to use a threshold in common for extraction in an image normalization device which performs preprocessing of processing for extracting an area of a reflector being an object from the satellite image. <P>SOLUTION: A histogram calculation part 4 calculates average and standard deviation before normalization for an original image and a normalization parameter calculation part 8 calculates average and standard deviation after the normalization so that distribution of areas including the object is extended. A normalization LUT calculation part 11 calculates luminance after the normalization corresponding to luminance before the normalization and a normalization image generation part 13 generates a normalization image based on its result. The normalization image is devoted to extraction processing of an object area using a common threshold. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to preprocessing of an image processing apparatus that extracts a region of a reflector, which is a target, from a satellite image, and relates to a technique for normalizing a satellite image and sharing a threshold value used for extraction.

  When extracting a region of a reflector such as a ship from a satellite image, a threshold for luminance is set, and the region of the reflector as a target is extracted using the threshold. If an appropriate threshold is not set, the reflector region cannot be extracted accurately. Therefore, it is important to use an appropriate threshold.

  However, the satellite image has a difference in brightness depending on the shooting conditions (weather, shooting time, etc.). For this reason, the luminance distribution differs for each satellite photograph. FIG. 23 is a diagram illustrating a relationship between a histogram and a threshold value.

  The histogram about the image A and the image B is shown. The histogram uses luminance as a class and number of pixels as a frequency. Since the image A is a dark image as a whole, the distribution is concentrated on the lower luminance side. On the other hand, since the image B is a relatively bright image, the luminance distribution spreads from the center.

  The appropriate threshold is different for each image. An appropriate threshold value for the image A is significantly smaller than an appropriate threshold value for the image B. If the threshold value used for the image B is used as it is in the image A, no target region can be extracted.

Therefore, it is necessary to extract the target area while adjusting the threshold value for each image, which is very troublesome.
JP-A 63-220375 JP 09-114977 A

  The present invention has been made in order to eliminate the above-described drawbacks of the prior art, and an object of the present invention is to provide an image processing apparatus that extracts a reflector region as a target from satellite image information. It is an object of the present invention to normalize each satellite image as processing so that the threshold value used for extraction is made common, and the accuracy of extraction is improved and stabilized.

An image normalization apparatus according to the present invention includes:
(1) An image input unit that inputs an image composed of a multi-tone pixel group (2) An image storage unit that stores an input image (3) At least one of the stored images An image normalization processing unit that normalizes a region of the image and generates a normalized image obtained by normalizing the region.

The image normalization processing unit
For the stored image, the histogram before calculating the average before the normalization with respect to the histogram with the luminance as the class and the number of pixels as the frequency, and the standard deviation before normalization,
A normalization parameter calculation unit that calculates an average after normalization and a standard deviation after normalization based on the average before normalization and the standard deviation before normalization;
An image conversion unit that converts an image before normalization into an image after normalization based on an average before normalization, a standard deviation before normalization, an average after normalization, and a standard deviation after normalization; It is characterized by having.

  The normalization parameter calculation unit calculates the average after normalization and the normalization so as to convert the luminance before normalization corresponding to a predetermined deviation from the average before normalization into the predetermined luminance after normalization. The standard deviation is calculated.

  The normalization parameter calculation unit calculates, as the histogram characteristic of the image to be normalized, a value assumed to be smaller than the luminance before normalization corresponding to a predetermined deviation from the average before normalization. It is stored as a predetermined luminance after conversion.

  The normalization parameter calculation unit stores a minimum value in a range indicating luminance as the predetermined luminance after normalization.

  The normalization parameter calculation unit is configured to convert the actual maximum luminance, which is the maximum value among the luminances of all the pixels included in the region before normalization, into a predetermined luminance after normalization. An average and a standard deviation after normalization are calculated.

  The normalization parameter calculation unit stores, as the histogram characteristics of the image to be normalized, a value assumed to be larger than the actual maximum luminance as the predetermined luminance after the normalization. To do.

  The normalization parameter calculation unit stores a maximum value in a range indicating luminance as the predetermined luminance after normalization.

  The normalization parameter calculation unit converts the luminance before normalization corresponding to a predetermined deviation from the average before normalization into a predetermined small side luminance after normalization, and in the region before normalization. Calculates the average after normalization and the standard deviation after normalization so that the actual maximum luminance, which is the maximum value among the luminances of all the pixels included in, is converted to the specified large-side luminance after normalization It is characterized by doing.

  The normalization parameter calculation unit calculates, as the histogram characteristic of the image to be normalized, a value assumed to be smaller than the luminance before normalization corresponding to a predetermined deviation from the average before normalization. It is stored as a predetermined small-side luminance after normalization, and further, a value assumed to be larger than the actual maximum luminance is stored as the predetermined large-side luminance after normalization.

  The normalization parameter calculation unit stores the minimum value in the range indicating luminance as the predetermined small-side luminance after normalization, and further calculates the maximum value in the range indicating luminance as the predetermined value after normalization. It is memorized as large luminance.

  The normalization parameter calculation unit converts the luminance before normalization corresponding to a predetermined small side deviation from the average before normalization into a predetermined small side luminance after normalization, and an average before normalization. To calculate the average after normalization and the standard deviation after normalization so that the luminance before normalization corresponding to the predetermined large-side deviation from is converted to the predetermined large-side luminance after normalization It is characterized by.

  The normalization parameter calculation unit, as a histogram characteristic of the image to be normalized, a value assumed to be smaller than the luminance before normalization corresponding to a predetermined small side deviation from the average before the normalization, Stored as the predetermined small-side luminance after normalization, and further, a value assumed to be larger than the luminance before normalization corresponding to a predetermined large-side deviation from the average before the normalization, It is memorized as a predetermined large-side luminance later.

  The normalization parameter calculation unit stores the minimum value in the range indicating luminance as the predetermined small-side luminance after normalization, and further calculates the maximum value in the range indicating luminance as the predetermined value after normalization. It is memorized as large luminance.

The image converter
Luminance after normalization for each value in the range indicating luminance before normalization based on average before normalization, standard deviation before normalization, average after normalization, and standard deviation after normalization A normalized LUT calculator for calculating the value of
An LUT array storage unit that stores the normalized luminance value for each value in the range indicating the luminance before normalization calculated by the normalized LUT calculation unit;
A normal value that reads the value of each pixel included in the image before normalization, reads the luminance value after normalization for that value from the LUT array storage unit, and writes it to the value of the pixel at the same position included in the image after normalization And a converted image generation unit.

  The image normalization apparatus further includes a target area extraction unit that extracts a target area that is a reflector from an area corresponding to the area of the normalized image using a predetermined threshold value for luminance. It is characterized by that.

An image normalization method according to the present invention includes:
(1) a step of inputting an image composed of a multi-tone pixel group, (2) a step of storing the input image, and (3) at least a partial region of the stored image. Normalizing and generating a normalized image obtained by normalizing the area.

A computer-readable recording medium on which a program according to the present invention is recorded,
A computer-readable recording medium that records a program for causing a computer as an image normalization apparatus to execute the following processing. (2) Processing for storing the input image (3) Processing for normalizing at least a part of the stored image and generating a normalized image obtained by normalizing the region.

The program according to the present invention is:
(1) A procedure for inputting an image composed of a multi-tone pixel group (2) An input image is stored. Procedure (3) A procedure of normalizing at least a part of the stored image and generating a normalized image obtained by normalizing the region.

  By performing the above-described normalization for image groups constituting different histograms, the background and the target area can be separated by the same threshold value. Further, by expanding the distribution of the target area, it is possible to increase the luminance difference from the background remaining in the target area. Thereby, the accuracy of the threshold value to be set is improved, and the quality of automatic target extraction from the image groups constituting different histograms is improved and stabilized.

Embodiment 1 FIG.
Hereinafter, the present invention will be described based on embodiments shown in the drawings. FIG. 1 is a diagram illustrating a configuration of an image normalization apparatus according to the first embodiment.

  The image normalization apparatus includes an original image input unit 1, an image storage unit 2, a land area removal unit 3, a histogram calculation unit 4, a histogram array storage unit 5, an average value storage unit 6, a standard deviation storage unit 7, and a normalization parameter calculation. Unit 8, normalized average value storage unit 9, normalized standard deviation storage unit 10, normalized LUT (lookup table) calculation unit 11, LUT array storage unit 12, normalized image generation unit 13, and normalized image Each element of the storage unit 14 is included.

  Among these elements, the normalized LUT calculation unit 11, the LUT array storage unit 12, and the normalized image generation unit 13 constitute an image conversion unit that converts an image before normalization into an image after normalization. .

  The histogram calculation unit 4, the normalization parameter calculation unit 8, and the image conversion unit described above constitute an image normalization processing unit.

  In this example, the target region extraction unit 15 that extracts a target region (for example, a region of a reflecting object having a high luminance such as a ship) using the normalized image and the target region extraction unit 15 further extract the target region. The target area information storage unit 16 is provided for storing the target area information.

  FIG. 2 is a diagram illustrating an overall processing flow of image normalization according to the first embodiment. The outline of the processing will be described in order.

First, an original image input process (S201) by the original image input unit 1 is performed to input an original image. In this example, a satellite image (a photographic image taken by a satellite) including a land area (referred to as a land area) and a marine area (referred to as a sea area) is used as an original image. The total number of pixels can be obtained by multiplying the number of pixels in the east-west direction (number of pixels) by the number of pixels in the north-south direction (number of lines). In this example, an image having a resolution corresponding to 1 square m per pixel and a gradation of 2 ¹⁶ (brightness K: 0 to 2 ¹⁶ -1 = 65535) is used. Specifically, the original image is read from a storage medium such as a CD-ROM or a hard disk. The read original image is stored in the image storage unit 2.

  Next, a land area removal process (S202) by the land area removal unit 3 is performed, and a process of distinguishing the land area as not subject to processing is performed. In this embodiment, the brightness of the pixels in the land area is changed to 0 to distinguish it from the sea area, and the brightness of the pixels having 0 brightness in the sea area is changed to 1 to distinguish it from the land area. That is, the image is converted so that a pixel having a luminance of 0 is a land area and a pixel having a luminance of 1 or more is a sea area. The converted image (referred to as a sea area image) is stored in the image storage unit 2.

  The land area removal process (S202) will be described later with reference to FIG.

  Next, a histogram calculation process (S203) by the histogram calculation unit 4 is performed to calculate a histogram array (frequency distribution) having luminance as a class for all pixels included in the sea area image, and further, pixels included in the sea area (luminance) Average value m and standard deviation σ are calculated for pixels other than 0). The histogram array is stored in the histogram array storage unit 5, the average value m is stored in the average value storage unit 6, and the standard deviation σ is stored in the standard deviation storage unit 7.

  The histogram calculation process (S203) will be described later with reference to FIGS.

Next, normalization parameter calculation processing (S204) by the normalization parameter calculation unit 8 is performed to calculate parameters for normalizing the sea area image. Specifically, an average value m ′ after normalization and a standard deviation σ ′ after normalization are calculated. Normalization is performed by converting the maximum luminance (real maximum luminance K _max ) among the luminances included in the sea area image into the maximum theoretically possible luminance (theoretical maximum luminance: 2 ¹⁶ −1 in this example), In addition, the luminance m + Aσ included in the sea area image is converted into luminance 0. The normalized average value m ′ is stored in the normalized average value storage unit 9, and the normalized standard deviation σ ′ is stored in the normalized standard deviation storage unit 10.

  The normalization parameter calculation process (S204) will be described later with reference to FIG.

  Next, normalized LUT calculation processing (S205) by the normalized LUT calculation unit 11 is performed to calculate an LUT (luminance conversion table) array. The LUT array is information of an array type (described as LUT (K)) with the pre-normalized brightness K as an index for storing the post-normalized brightness K ′ corresponding to each pre-normalized brightness K. The post-normalization brightness K ′ is calculated using the pre-normalization brightness K, the pre-normalization average value m, the pre-normalization standard deviation σ, the post-normalization average value m ′, and the post-normalization standard deviation σ ′. . The LUT array is stored in the LUT array storage unit 12.

  The normalized LUT calculation process (S205) will be described later with reference to FIG.

  Next, normalized image generation processing (S206) is performed by the normalized image generation unit 13, and the luminance of all pixels included in the sea area image is converted using the LUT array. As a result, a normalized image having the converted luminance is obtained.

  The normalized image generation process (S206) will be described later with reference to FIG.

  Finally, the normalized image is stored in the normalized image storage unit 14 by the normalized image writing process (S207). The normalized image storage unit 14 is secured in a temporary area on a memory managed by a storage medium such as a hard disk or a system, for example.

  The normalized image is used for the process of extracting the target area by the target area extracting unit 15. In the target area extraction process, for example, the process of comparing the brightness of each pixel with a predetermined threshold to determine a high brightness pixel and the process of selecting a pixel in a predetermined positional relationship with the determined high brightness pixel are combined. To extract the target area.

  Details of each process will be described below.

  First, the land area removal process (S202) will be described. FIG. 3 is a diagram showing a flow of land removal processing (S202).

  In order to remove the land area in the original image, the position and brightness are sequentially acquired for all the pixels, and the following processing is performed on each pixel (S301). The pixel number I is used as a loop counter that sequentially increments (+1). The pixel position can be specified by the pixel number I. For example, the pixel number can be divided by the number of lines, the quotient can be the line number, and the remainder can be the pixel number.

  Next, land area determination processing (S302) is performed. In this processing, information that determines whether each pixel is a land area or a water area by a conventional method (for example, a method disclosed in Japanese Patent Application Laid-Open No. 2000-353234, “Sea Surface Extraction Processing Method from Satellite Image”) is used. In this example, land area information including a pixel number group of pixels included in the land area is used. When the pixel number I of the pixel is included in the land area information, it is determined that the pixel is a land area. On the other hand, when the pixel number I of the pixel is not included in the land area information, it is determined that the pixel is not a land area (a sea area). Alternatively, it is also effective to use a bitmap as land area information, manage each pixel in association with each bit, and distinguish the land area from the sea area by turning the bit on and off.

  If it is determined that the pixel is a land area, land area brightness setting processing (S303) is performed. Thereby, the luminance of the pixel is changed to zero. Thereafter, the luminance 0 is used as a code indicating the land area.

  If it is determined that the pixel is not a land area, a sea area luminance determination process (S304) is performed. In this process, it is determined whether or not the brightness of the pixel is 0. If the brightness is 0, the sea area brightness changing process (S305) is performed. This process changes the luminance of the pixel to 1 which is a value approximating the original luminance 0 in order to avoid erroneous determination as a land area in spite of being a sea area. When all the pixels have been processed, the process ends (S306).

By the above-described processing, a sea area image processed so that the land area can be identified by a specific code (luminance 0) is generated on the image storage unit 2. In this sea area image, the brightness of the pixel that showed 0 brightness in the sea area is changed to 1 as described above. Therefore, the range of luminance that can be taken by the pixels in the sea area is 1 to 2 ¹⁶ −1 (theoretical maximum luminance), and the changed luminance 1 is the minimum value of the range.

  Next, the histogram calculation process (S203) will be described. 4 and 5 are diagrams showing the flow of the histogram calculation process (S203), and FIG. 6 is a diagram showing an example of the histogram arrangement.

  First, the configuration of the histogram array will be described with reference to FIG. As shown in the figure, the histogram array is information for storing a frequency distribution in which the luminance is class and the number of pixels is frequency. Specifically, it is array-type information that stores the number of pixels using luminance as an index.

  Next, processing will be described. First, the histogram array is initialized by histogram initialization processing (S401). Specifically, the value of each array is set to 0 in order to prepare for counting up the number of pixels.

Then, the brightness of all the pixels is sequentially acquired, and the histogram calculation process (S403) is repeated for each pixel (S402). The pixel number I is used as a loop counter that sequentially increments (+1). In the histogram calculation process (S403), the luminance of the pixel is acquired from the image storage unit 2, and the value of the histogram array using the luminance as an index is incremented (+1). Thereby, the number of pixels having the luminance is counted up. The program is described as follows.

Histogram array (I-th pixel brightness)
= Histogram array (I-th pixel brightness) +1;

Then, when all the pixels have been processed, the process proceeds to the next process (S404).

  Thereafter, a process for obtaining the average value m and the standard deviation σ is performed.

  First, internal variable initialization processing (S405) is performed. Two variable storage areas, the luminance sum S and the deviation square sum T, are secured in the internal variable area, and both variables are initialized to zero.

  Then, in order to obtain the total luminance of all the pixels in the sea area, the value of the histogram array excluding the luminance 0 (frequency at each luminance = number of pixels) is acquired, and the luminance total calculation processing (S407) is performed for each luminance. Repeat (S406). Luminance K is used as a loop counter that sequentially increments (+1).

In the brightness summation calculation process (S407), for each brightness, the value (number of pixels) of the histogram array is obtained using the brightness K as an index, the value of the histogram array is multiplied by the brightness K, and the resulting product is sequentially added to the brightness sum S. Perform the addition process. The program is described as follows.

Luminance sum S = luminance sum S + luminance K × histogram array (luminance K);

When all the luminances (1 to 2 ¹⁶ -1) that can be theoretically taken by the pixels in the sea area have been processed, the loop processing is ended (S408). Thereby, the sum of luminance in the sea area is obtained.

Next, an average value calculation process (S409) is performed. By this processing, the average value of luminance in the sea area is calculated. The number of pixels in the sea area is obtained by subtracting the histogram array (luminance 0) from the total number of pixels, and the average value is obtained by dividing the above-mentioned total luminance S by the number of pixels in the sea area. The program is described as follows.

Average value m = total luminance S / (total number of pixels−histogram arrangement (luminance 0));

Next, in order to obtain the sum of the squares of the deviations for all the pixels in the sea area (referred to as the sum of squared deviations or the sum of squared differences), the value of the histogram array excluding the luminance of 0 (the number of pixels of each luminance) is acquired. The deviation square sum calculation process (S411) is repeated for each luminance (S410). Luminance K is used as a loop counter that sequentially increments (+1).

In the deviation square sum calculation process (S411), the average value is subtracted from the luminance K for each luminance, the deviation is obtained as a variable, the deviation is squared, and the deviation square is obtained as a variable. Further, the histogram array value (number of pixels) is acquired using the luminance K as an index, the deviation square is multiplied by the value of the histogram array (number of pixels), and the obtained product is sequentially added to the deviation square sum T. The program is described as follows.

Deviation = Luminance K-Average value m;
Squared deviation = squared function (deviation);
Deviation sum of squares T =
Deviation square sum T + deviation square x histogram array (luminance K);

When the waters of pixels processed for all of the luminance may take theoretically (1-2 ¹⁶ -1), and ends the loop processing (S412). As a result, the sum of the squares of the deviations in the sea area is obtained.

Next, standard deviation calculation processing (S413) is performed. By this processing, the standard deviation σ of luminance in the sea area is calculated. The number of pixels in the sea area is obtained by subtracting the histogram array (luminance 0) from the total number of pixels, and the above-mentioned deviation square sum T is divided by the number of pixels in the sea area to obtain the variance and temporarily store the square root of the variance. The standard deviation σ is obtained by calculation. The program is described as follows.

Variance = total deviation square sum T / (total number of pixels−histogram arrangement (luminance 0));
Standard deviation σ = square root function (variance);

FIG. 7 is a diagram showing a histogram before normalization and a histogram after normalization. The upper part of the figure is a histogram before normalization, and the average value m and standard deviation σ of this histogram can be obtained by the above-described calculation process. The histogram distinguishes between a luminance range of pixels corresponding to the sea surface as a background and a luminance range including pixels corresponding to the target area. The reference luminance 701 for distinguishing these is obtained by adding a product of a constant A × standard deviation σ to the average value m. The maximum luminance (real maximum luminance K _max ) 702 among the luminances included in the sea area image is obtained by a process of sequentially comparing the luminance of each pixel and selecting a large luminance.

The lower part shows a histogram for an image obtained by normalization. In the normalization parameter calculation process (S204), which will be described later, the actual maximum brightness K _max 702 is converted into the maximum possible brightness (theoretical maximum brightness: 2 ¹⁶ −1 in this example) 712, and the reference brightness A parameter for normalization in which (average value m + constant A × standard deviation σ) 701 is converted to luminance 0 shown in 711 is obtained. Specifically, the average value m ′ after normalization and the standard deviation σ ′ after normalization are obtained.

Here, an expression for calculating the average value m ′ after normalization and the standard deviation σ ′ after normalization will be described. In general, when converting an original histogram represented by a normal distribution of mean value m ₀ and standard deviation σ ₀ to a new histogram represented by a normal distribution of mean value m ₁ and standard deviation σ ₁ , Accordingly, the luminance K ₀ of the original histogram is converted into the luminance K _{1 of the} new histogram.

Luminance K ₁ =
(Standard deviation σ ₁ / standard deviation σ ₀ ) × (luminance K ₀ −average value m ₀ ) + average value m ₁

Therefore, the following simultaneous equations can be derived from the above normalization conditions.

Maximum theoretical brightness (2 ¹⁶ -1) =
(Standard deviation after normalization σ ′ / standard deviation before normalization σ) ×
(Real maximum brightness K _max -average value m before normalization) + average value m ′ after normalization

Luminance 0 =
(Standard deviation after normalization σ ′ / standard deviation before normalization σ) ×
((Average value before normalization m + constant A × standard deviation before normalization σ) −average value before normalization m)
+ Normalized value m ′ after normalization

Accordingly, the normalized average value m ′ and the normalized standard deviation σ ′ are obtained by the following equations.

Standard deviation after normalization σ ′ = maximum theoretical brightness (2 ¹⁶ −1) × standard deviation before normalization σ /
(Real maximum brightness K _max -average value before normalization m-constant A × standard deviation σ before normalization)

Average value after normalization m ′ = − constant A × standard deviation after normalization σ ′

Next, the normalization parameter calculation process (S204) will be described. FIG. 8 is a diagram showing a flow of normalization parameter calculation processing (S204).

First, a normalized standard deviation σ ′ is calculated by a normalized standard deviation calculation process (S801). In order to calculate the denominator as a variable, the standard deviation σ before normalization is multiplied by a constant A, and the product is temporarily stored as a variable. Then, the average value m before normalization is subtracted from the actual maximum luminance K _max , the product temporarily stored is subtracted from the difference, and the obtained difference is temporarily stored as the denominator value. In order to calculate a numerator as a variable, the theoretical maximum luminance (2 ¹⁶ −1) is multiplied by a standard deviation σ before normalization, and the product is temporarily stored as a numerator value. Finally, the values of the numerator and denominator are read out, the value of the numerator is divided by the value of the denominator, and the obtained quotient is stored in the normalized standard deviation storage unit 10 as the normalized standard deviation σ ′. The program is described as follows.

Product of variables = constant A × standard deviation σ before normalization;
Variable denominator = actual maximum brightness K _max -average value before normalization m -variable product;
Variable numerator = theoretical maximum luminance (2 ¹⁶ -1) x standard deviation σ before normalization;
Standard deviation after normalization σ ′ = variable numerator / variable denominator;

Next, a normalized average value calculation process (S802) is performed. The normalized standard deviation σ ′ obtained in S801 is multiplied by a constant A stored in advance, the sign of the obtained product is inverted, and the obtained value is used as a normalized average value m ′. Store in the value storage unit 9. The program is described as follows.

Normalized average value m ′ = − constant A × normalized standard deviation σ ′;

Next, the normalized LUT calculation process (S205) will be described. FIG. 9 is a diagram showing a flow of normalized LUT calculation processing (S205). FIG. 10 is a diagram illustrating an example of the LUT array. The LUT array is information for storing the luminance after normalization with respect to the luminance before normalization. Specifically, it is array type information that stores the luminance after normalization using the luminance before normalization as an index.

  First, as a pre-processing, the LUT array is initialized (S901), and the land area identification code 0 is similarly set to the normalized luminance with respect to the land area identification code (luminance 0) (S902). That is, 0 is stored in the LUT array (0). Thereby, also in the image after normalization, the land area and the sea area can be distinguished as before the normalization.

  In order to store the normalized luminance in the LUT array, the following processing (S904 to S907) is repeated in order (S903). The luminance K before normalization is used as a loop counter that sequentially increments (+1).

A normalized luminance calculation process (S904) is performed, and a normalized luminance K ′ is calculated as a variable. First, the standard deviation after normalization σ ′ and the standard deviation before normalization σ are read out, the standard deviation after normalization σ ′ is divided by the standard deviation before normalization σ, and the obtained quotient is temporarily stored as a variable. Next, the average value m before normalization is read, and the average value m before normalization is subtracted from the luminance K before normalization, which is a loop counter, to obtain a difference. The stored quotient is multiplied by the obtained difference, and the obtained product is temporarily stored as a variable. Furthermore, the average value m ′ after normalization is read out, and in addition to the stored product, the obtained sum is temporarily stored as the normalized luminance K ′. The program is described as follows.

Variable quotient = standard deviation after normalization σ ′ / standard deviation before normalization σ;
Variable product = variable quotient × (luminance K before normalization−average value m before normalization);
Normalized luminance K ′ = variable product + normalized average value m ′;

Next, in the luminance determination process (S905), it is determined whether the calculated normalized luminance K ′ is included in the normalized luminance range (1 to theoretical maximum luminance 2 ¹⁶ −1). In this example, when the normalized luminance K ′ is 0 or less, it is determined that it is out of the normalized luminance range, and when it is 1 or more, it is determined that it is within the normalized luminance range. .

  If it is determined that the luminance is not within the normalized luminance range, the luminance setting process (S906) sets the normalized luminance K ′ to 1 which is the minimum luminance in the normalized luminance range. When it is determined that the luminance is within the normalized luminance range, the calculated normalized luminance K ′ is used as it is.

The value of the normalized luminance K ′ thus determined is stored in the LUT array that is the normalized luminance corresponding to the pre-normalized luminance K in the LUT array storage unit 12 by the LUT setting process (S907). The program is described as follows.

LUT arrangement (luminance K before normalization) = luminance K ′ after normalization;

When it is determined that the luminance K before normalization is the actual maximum luminance K _max , the loop processing is terminated (S908).

  Next, the normalized image generation process (S206) will be described. FIG. 11 is a diagram showing a flow of normalized image generation processing (S206).

In order to convert the luminance of the sea area image into normalized pixels, the luminance of all the pixels of the sea area image is sequentially acquired, and the luminance conversion process (S1102) is repeated for each pixel (S1101). The pixel number I is used as a loop counter that sequentially increments (+1). In the luminance conversion process (S1102), the luminance is converted using the LUT array. Specifically, the luminance of the pixel specified by the pixel number I is acquired from the image storage unit 2, and an LUT array using the luminance as an index is read from the LUT array storage unit 12. Then, the value is stored in the pixel on the normalized image storage unit 14 specified by the pixel number as the normalized luminance. The program is described as follows.

I-th normalized luminance K ′ = LUT array (I-th pre-normalized luminance K);

Through the above-described processing, a normalized image forming the normalized histogram shown in the lower part of FIG. 7 is generated and stored in the normalized image storage unit 14. However, to be precise, as shown in FIG. As shown in FIG. 21, image groups forming histograms having different distributions are converted into image groups forming histograms having similar distributions as a result of normalization.

In the normalized image, most of the pixels corresponding to the sea surface as the background have a luminance of 1, and the range of luminance from 2 to 2 ¹⁶ -1 is a small number of pixels corresponding to the target and pixels corresponding to the sea surface. Part.

  The normalized image is used for the process of extracting the target area by the target area extracting unit 15. In the target area extraction process, a predetermined area is used as a common threshold, and a target area is extracted by determining a pixel having a luminance equal to or higher than (or higher than) the threshold as a pixel of the target area. As a result, even when image groups having different contrasts are used, it is not necessary to individually adjust the threshold value.

Embodiment 2. FIG.
In the first embodiment, image data including a code for identifying a land area is first generated and processed based on the image data. In the present embodiment, a pixel in the sea area is selected each time processing is performed. Will be described.

  FIG. 12 is a diagram illustrating a configuration of an image normalization apparatus according to the second embodiment. The land area removal unit 3 in the first embodiment is omitted, and the histogram calculation unit 4, the normalization parameter calculation unit 8, the normalization LUT calculation unit 11, and the normalized image generation unit 13 input land area information.

  FIG. 13 is a diagram showing an overall processing flow of image normalization according to the second embodiment. In the present embodiment, the land area removal process (S202) is omitted.

  14 and 15 are diagrams illustrating a histogram calculation processing flow according to the second embodiment. In the loop processing of S402, it is determined whether the pixel is a sea area based on the sea area information (S1401). Then, the histogram calculation process (S403) is performed only when the area is a sea area, and the process is not performed when the area is a land area. Therefore, the generated histogram targets only pixels in the sea area. FIG. 16 is a diagram illustrating an example of a histogram array according to the second embodiment. In the present embodiment, the frequency with 0 brightness indicates the number of pixels with 0 brightness in the sea area.

In the luminance sum calculation processing (S407), the luminance sum S from 0 to 2 ¹⁶ -1 including the luminance 0 is calculated (S1402). In the average value calculation process (S1403), the sum total of the histogram array is calculated, and the number of pixels in the sea area is obtained as a variable. Then, the average value is obtained by dividing the luminance sum S described above by the number of pixels in the sea area. The program is described as follows.

Number of sea area pixels = summation function (histogram array);
Average value m = total luminance S / number of sea area pixels;

The deviation square sum calculation process (S411) calculates the luminance deviation square sum T from 0 to 2 ¹⁶ -1 including the luminance 0 (S1404). In the standard deviation calculation process (S1405), the total of the histogram array is calculated, and the number of pixels in the sea area is obtained as a variable. The program is described as follows.

Number of sea area pixels = summation function (histogram array);
Variance = Deviation square sum T / Number of sea area pixels;
Standard deviation σ = square root function (variance);

In the normalized standard deviation calculation process (S801) in the normalization parameter calculation process (S204) shown in FIG. 8, the maximum luminance among the pixels in the sea area of the original image is used as the actual maximum luminance. This value is obtained by sequentially comparing the brightness of pixels determined to be sea areas based on the land area information and selecting a larger brightness.

FIG. 17 is a normalized LUT calculation process flow according to the second embodiment. In the initialization of S901, 0 is set in all LUT arrays. Then, the normalized luminance calculation process (S904) calculates the normalized luminance K ′ from 0 to 2 ¹⁶ −1 including the luminance 0 (S1701).

In the luminance determination process (S905), it is determined whether the calculated normalized luminance K ′ is included in the normalized luminance range (0 to the theoretical maximum luminance 2 ¹⁶ −1). In this example, when the normalized luminance K ′ is less than 0, it is determined that it is outside the normalized luminance range, and when it is 0 or more, it is determined that it is within the luminance range.

  In the luminance setting process (S1704), when it is determined that the luminance is out of the normalized luminance range, the normalized luminance K ′ is set to 0 which is the minimum luminance in the normalized luminance range. FIG. 18 is a diagram illustrating an example of the LUT array according to the second embodiment.

  FIG. 19 is a diagram illustrating a normalized image generation processing flow according to the second embodiment. In the loop processing of S1101, it is determined whether the pixel is a sea area based on the land area information (S1901). Then, the luminance conversion process (S1102) is performed only when the area is a sea area, and the process is not performed when the area is a land area. Therefore, only the sea area pixels are normalized, and the land area pixels remain unchanged.

  The target area extraction unit 15 uses the land area information to extract a target area within a range determined as a sea area.

Embodiment 3 FIG.
In this embodiment, a mode in which an image not including a land area, that is, an image including only a sea area is used will be described. FIG. 20 is a diagram illustrating a configuration of an image normalization apparatus according to the third embodiment. In this embodiment, land area information is not used.

  The entire process is performed in the same manner as in the second embodiment in principle according to the flow shown in FIG. However, it is not necessary to judge the sea area.

  Of the histogram calculation processing according to the second embodiment shown in FIGS. 14 and 15, the determination of the sea area in S1401 is not necessary.

In the average value calculation process (S1403), the average value is obtained by dividing the luminance sum S by the total number of pixels. The program is described as follows.

Average value m = total luminance S / total number of pixels;

In the standard deviation calculation process (S1405), the total number of pixels is used instead of the number of pixels in the sea area. The program is described as follows.

Variance = deviation square sum T / total number of pixels;
Standard deviation σ = square root function (variance);

In the normalized standard deviation calculation process (S801) in the normalization parameter calculation process (S204) shown in FIG. 8, the maximum luminance among the pixels of the original image is used as the actual maximum luminance. This value is obtained by sequentially comparing the luminance of all the pixels and selecting a larger luminance.

  In the normalized image generation processing according to Embodiment 2 shown in FIG. 19, the determination in S1901 is not necessary.

Embodiment 4 FIG.
In the above-described embodiment, as shown in FIG. 7, the luminance 701 before normalization corresponding to a predetermined deviation from the average before normalization is converted into the minimum value 711 in the range indicating the luminance, and normalized. Although the example in which the previous actual maximum brightness 702 is normalized so as to be converted into the theoretical maximum brightness value 712 and the target area is expanded most has been shown, in this embodiment, the distribution of the target area is generally extended. An example will be described.

  FIG. 21 is a diagram showing a histogram before normalization and a histogram after normalization according to the fourth embodiment. In this form, the luminance 2101 corresponding to a predetermined deviation from the average before normalization is converted into a predetermined small-side luminance 2111 after normalization, and the actual maximum luminance 2102 before normalization is converted to a predetermined value after normalization. Is normalized so as to be converted into a large-side luminance 2112.

  In this example, the normalization parameter calculation unit 8 stores and uses a value assumed to be smaller than the luminance 2101 corresponding to a predetermined deviation as the small luminance 2111 as the histogram characteristic of the image to be normalized. ing. Similarly, a value assumed to be larger than the actual maximum luminance 2102 is stored and used as the large-side luminance 2112 as the histogram characteristics of the image to be normalized. The small-side luminance 2111 and the large-side luminance 2112 are commonly used in normalizing each image.

  As described above, when the difference between the large-side luminance 2112 and the small-side luminance 2111 is larger than the difference between the actual maximum luminance 2102 and the luminance 2101 corresponding to a predetermined deviation, the target area is enlarged.

An expression for calculating the average value m ′ after normalization and the standard deviation σ ′ after normalization will be described. From the normalization conditions described above, the following simultaneous equations can be derived.

Large side brightness K ⁺ =
(Standard deviation after normalization σ ′ / standard deviation before normalization σ) ×
(Real maximum brightness K _max -average value m before normalization) + average value m ′ after normalization

Small side luminance K ^- =
(Standard deviation after normalization σ ′ / standard deviation before normalization σ) ×
((Average value before normalization m + constant A × standard deviation before normalization σ) −average value before normalization m)
+ Normalized value m ′ after normalization

Accordingly, the normalized average value m ′ and the normalized standard deviation σ ′ are obtained by the following equations.

Standard deviation after normalization σ ′ =
(Large luminance K ⁺ −small luminance K ⁻ ) × standard deviation before normalization σ /
(Real maximum brightness K _max -average value before normalization m-constant A × standard deviation σ before normalization)

Average value after normalization m ′ = small luminance K ⁻ −constant A × standard deviation after normalization σ ′

In the normalized standard deviation calculating process (S801) of the normalized parameter calculating process (S204), the normalized standard deviation σ ′ is calculated as follows.

In order to calculate the denominator as a variable, the standard deviation σ before normalization is multiplied by a constant A, and the product is temporarily stored as a variable. Then, the average value m before normalization is subtracted from the actual maximum luminance K _max , the product temporarily stored is subtracted from the difference, and the obtained difference is temporarily stored as the denominator value. In order to calculate the numerator as a variable, the small luminance K ⁻ is subtracted from the large luminance K ⁺ and the luminance difference is temporarily stored as a variable. The luminance difference is multiplied by the standard deviation σ before normalization, and the product is temporarily stored as a numerator value. Finally, the values of the numerator and denominator are read out, the value of the numerator is divided by the value of the denominator, and the obtained quotient is stored in the normalized standard deviation storage unit 10 as the normalized standard deviation σ ′. The program is described as follows.

Product of variables = constant A × standard deviation σ before normalization;
Variable denominator = actual maximum brightness K _max -average value before normalization m -variable product;
Variable luminance difference = larger luminance K ⁺ −smaller luminance K ⁻ ;
Variable numerator = Variable brightness difference x Standard deviation σ before normalization;
Standard deviation after normalization σ ′ = variable numerator / variable denominator;

The brightness difference may be stored in advance, and the stored brightness difference may be used instead of the calculation of the variable brightness difference.

In the normalized average value calculation process (S802), the normalized standard deviation σ ′ obtained in S801 is multiplied by a constant A stored in advance, and the obtained product is temporarily stored as a variable. Then, the stored product is subtracted from the small-side luminance K ⁻ , and the difference is stored as the normalized average value m ′ in the normalized average value storage unit 9. The program is described as follows.

Product of variables = constant A × standard deviation after normalization σ ′;
Normalized average value m ′ = small luminance K ⁻ −variable product;

Other processes are the same as those in the above-described embodiment.

Embodiment 5 FIG.
In the present embodiment, another example of expanding the distribution of the target area will be described.

  FIG. 22 is a diagram showing a histogram before normalization and a histogram after normalization according to the fifth embodiment. In this form, the luminance 2201 corresponding to a predetermined small deviation from the average before normalization is converted into a minimum value 2211 in the range indicating the luminance, and corresponds to a predetermined large deviation from the average before normalization. The normalization is performed so that the luminance 2202 to be converted into the theoretical maximum luminance 2212 that is the maximum value in the range indicating the luminance.

  Note that the minimum value 2211 is an example of a predetermined small-side luminance after normalization, in particular, an example of a value assumed to be smaller than the luminance 2201 corresponding to a predetermined small-side deviation from the average before normalization. . The theoretical maximum luminance 2212 is an example of a predetermined large-side luminance after normalization, in particular, an example of a value assumed to be larger than the luminance 2202 corresponding to a predetermined large-side deviation from the average before normalization. It is.

  In this example, the normalization parameter calculation unit 8 includes a small side constant A for calculating the luminance 2201 corresponding to a predetermined small deviation from the average before normalization, and a predetermined large value from the average before normalization. A large-side constant B for calculating the luminance 2202 corresponding to the side deviation is stored in advance. These small side constant A and large side constant B are used in common in normalization of each image.

An expression for calculating the average value m ′ after normalization and the standard deviation σ ′ after normalization will be described. From the normalization conditions described above, the following simultaneous equations can be derived.

Theoretical maximum brightness K (2 ¹⁶ -1) =
(Standard deviation after normalization σ ′ / standard deviation before normalization σ) ×
((Average value before normalization m + large constant B × standard deviation before normalization σ) −average value before normalization m)
+ Normalized value m ′ after normalization

Luminance 0 =
(Standard deviation after normalization σ ′ / standard deviation before normalization σ) ×
((Average value before normalization m + small constant A × standard deviation before normalization σ) −average value before normalization m)
+ Normalized value m ′ after normalization

Accordingly, the normalized average value m ′ and the normalized standard deviation σ ′ are obtained by the following equations.

Standard deviation after normalization σ ′ =
Theoretical maximum brightness K (2 ¹⁶ -1) / (large constant B-small constant A)

Average value after normalization m ′ = − standard deviation after normalization σ ′ × small constant A

In the normalized standard deviation calculating process (S801) of the normalized parameter calculating process (S204), the normalized standard deviation σ ′ is calculated as follows.

In order to obtain the denominator as a variable, the small constant A is subtracted from the large constant B, and the difference is temporarily stored as the denominator value. Then, the theoretical maximum luminance K (2 ¹⁶ −1) is divided by the denominator value, and the obtained quotient is stored in the normalized standard deviation storage unit 10 as the normalized standard deviation σ ′. The program is described as follows.

Variable denominator = large constant B-small constant A;
Standard deviation after normalization σ ′ = theoretical maximum brightness K (2 ¹⁶ −1) / variable denominator;

In the normalized average value calculation process (S802), the normalized standard deviation σ ′ obtained in S801 is multiplied by a prestored small-side constant A, and the obtained product is temporarily stored as a variable. Then, the sign of the product is reversed, and the result is stored in the normalized average value storage unit 9 as the normalized average value m ′. The program is described as follows.

Variable product = small constant A × normalized standard deviation σ ′;
Normalized mean value m ′ = − variable product;

Other processes are the same as those in the above-described embodiment.

  The image normalization apparatus is a computer, and each element can execute processing by a program. Further, the program can be stored in a storage medium so that the computer can read the program from the storage medium.

  In the above-described embodiment, an example is shown in which a LUT array is calculated in advance and the normalized luminance is read from the array. However, each time the luminance is converted, the luminance of each pixel is set as the luminance before normalization. May be calculated.

1 is a diagram illustrating a configuration of an image normalization apparatus according to Embodiment 1. FIG. FIG. 6 is a diagram illustrating an overall processing flow of image normalization according to the first embodiment. It is a figure which shows a land area removal process flow. It is a figure which shows a histogram calculation process flow (the first half). It is a figure which shows a histogram calculation processing flow (second half). It is a figure which shows the example of a histogram arrangement | sequence. It is a figure which shows the histogram before normalization, and the histogram after normalization. It is a figure which shows the normalization parameter calculation processing flow. It is a figure which shows the normalization LUT calculation processing flow. It is a figure which shows the example of a LUT arrangement | sequence. It is a figure which shows the normalization image generation process flow. 6 is a diagram illustrating a configuration of an image normalization apparatus according to Embodiment 2. FIG. FIG. 10 is a diagram illustrating an overall processing flow of image normalization according to the second embodiment. FIG. 10 is a diagram showing a histogram calculation processing flow (first half) according to the second embodiment. FIG. 10 is a diagram showing a histogram calculation processing flow (second half) according to the second embodiment. FIG. 10 is a diagram showing an example of a histogram array according to the second embodiment. FIG. 10 is a diagram illustrating a normalized LUT calculation processing flow according to the second embodiment. FIG. 10 is a diagram showing an example of an LUT array according to the second embodiment. FIG. 10 is a diagram illustrating a normalized image generation processing flow according to the second embodiment. FIG. 10 is a diagram illustrating a configuration of an image normalization apparatus according to a third embodiment. It is a figure which shows the histogram before normalization based on Embodiment 4, and the histogram after normalization. It is a figure which shows the histogram before normalization based on Embodiment 5, and the histogram after normalization. It is a figure which shows the relationship between a histogram and a threshold value.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Original image input part, 2 Image storage part, 3 Land area removal part, 4 Histogram calculation part, 5 Histogram arrangement | sequence storage part, 6 Average value storage part, 7 Standard deviation storage part, 8 Normalization parameter calculation part, 9 Normalization Post-average value storage unit, 10 normalized standard deviation storage unit, 11 normalized LUT calculation unit, 12 LUT array storage unit, 13 normalized image generation unit, 14 normalized image storage unit, 15 target area extraction unit, 16 target Area information storage unit.

Claims (19)

An image normalization apparatus having the following elements: (1) an image input unit for inputting an image composed of a multi-tone pixel group; (2) an image storage unit for storing an input image; and (3) a stored image. An image normalization processing unit that normalizes at least a part of the region and generates a normalized image obtained by normalizing the region. The image normalization processing unit
For the stored image, the histogram before calculating the average before the normalization with respect to the histogram with the luminance as the class and the number of pixels as the frequency, and the standard deviation before normalization,
A normalization parameter calculation unit that calculates an average after normalization and a standard deviation after normalization based on the average before normalization and the standard deviation before normalization;
An image conversion unit that converts an image before normalization into an image after normalization based on an average before normalization, a standard deviation before normalization, an average after normalization, and a standard deviation after normalization; The image normalization apparatus according to claim 1, further comprising:   The normalization parameter calculation unit calculates the average after normalization and the normalization so as to convert the luminance before normalization corresponding to a predetermined deviation from the average before normalization into the predetermined luminance after normalization. The image normalization apparatus according to claim 2, wherein the standard deviation is calculated.   The normalization parameter calculation unit calculates, as the histogram characteristic of the image to be normalized, a value assumed to be smaller than the luminance before normalization corresponding to a predetermined deviation from the average before normalization. 4. The image normalization apparatus according to claim 3, wherein the image normalization apparatus stores the predetermined luminance after the conversion.   The image normalization apparatus according to claim 4, wherein the normalization parameter calculation unit stores a minimum value in a range indicating luminance as the predetermined luminance after normalization.   The normalization parameter calculation unit is configured to convert the actual maximum luminance, which is the maximum value among the luminances of all the pixels included in the region before normalization, into a predetermined luminance after normalization. The image normalization apparatus according to claim 2, wherein an average and a standard deviation after normalization are calculated.   The normalization parameter calculation unit stores, as the histogram characteristics of the image to be normalized, a value assumed to be larger than the actual maximum luminance as the predetermined luminance after the normalization. The image normalization apparatus according to claim 6.   The image normalization apparatus according to claim 7, wherein the normalization parameter calculation unit stores a maximum value in a range indicating luminance as the predetermined luminance after the normalization.   The normalization parameter calculation unit converts the luminance before normalization corresponding to a predetermined deviation from the average before normalization into a predetermined small side luminance after normalization, and in the region before normalization. Calculates the average after normalization and the standard deviation after normalization so that the actual maximum luminance, which is the maximum value among the luminances of all the pixels included in, is converted to the specified large-side luminance after normalization The image normalization apparatus according to claim 2, wherein:   The normalization parameter calculation unit calculates, as the histogram characteristic of the image to be normalized, a value assumed to be smaller than the luminance before normalization corresponding to a predetermined deviation from the average before normalization. And storing a value assumed to be larger than the actual maximum luminance as the predetermined large-side luminance after normalization. 9. The image normalization apparatus according to 9.   The normalization parameter calculation unit stores the minimum value in the range indicating luminance as the predetermined small-side luminance after normalization, and further calculates the maximum value in the range indicating luminance as the predetermined value after normalization. The image normalization apparatus according to claim 10, wherein the image normalization apparatus stores the large-side luminance.   The normalization parameter calculation unit converts the luminance before normalization corresponding to a predetermined small side deviation from the average before normalization into a predetermined small side luminance after normalization, and an average before normalization. To calculate the average after normalization and the standard deviation after normalization so that the luminance before normalization corresponding to the predetermined large-side deviation from is converted to the predetermined large-side luminance after normalization The image normalization apparatus according to claim 2.   The normalization parameter calculation unit, as a histogram characteristic of the image to be normalized, a value assumed to be smaller than the luminance before normalization corresponding to a predetermined small side deviation from the average before the normalization, Stored as the predetermined small-side luminance after normalization, and further, a value assumed to be larger than the luminance before normalization corresponding to a predetermined large-side deviation from the average before the normalization, The image normalization apparatus according to claim 12, wherein the image normalization apparatus stores the brightness as a later predetermined large-side luminance.   The normalization parameter calculation unit stores the minimum value in the range indicating luminance as the predetermined small-side luminance after normalization, and further calculates the maximum value in the range indicating luminance as the predetermined value after normalization. The image normalization apparatus according to claim 13, wherein the image normalization apparatus is stored as large-side luminance. The image converter
Luminance after normalization for each value in the range indicating luminance before normalization based on average before normalization, standard deviation before normalization, average after normalization, and standard deviation after normalization A normalized LUT calculator for calculating the value of
An LUT array storage unit that stores the normalized luminance value for each value in the range indicating the luminance before normalization calculated by the normalized LUT calculation unit;
A normal value that reads the value of each pixel included in the image before normalization, reads the luminance value after normalization for that value from the LUT array storage unit, and writes it to the value of the pixel at the same position included in the image after normalization The image normalization apparatus according to claim 2, further comprising: a normalized image generation unit.   The image normalization apparatus further includes a target area extraction unit that extracts a target area that is a reflector from an area corresponding to the area of the normalized image using a predetermined threshold value for luminance. The image normalization apparatus according to claim 1, wherein: An image normalization method having the following elements: (1) a step of inputting an image composed of a multi-tone pixel group (2) a step of storing an input image (3) at least one of the stored images Normalizing the area of the image and generating a normalized image obtained by normalizing the area. A computer-readable recording medium that records a program for causing a computer as an image normalization apparatus to execute the following processing (1) Processing for inputting an image composed of a multi-tone pixel group (2) Input image Processing for storing (3) Processing for normalizing at least a part of the stored image and generating a normalized image obtained by normalizing the region. A program for causing a computer as an image normalization apparatus to execute the following procedure (1) A procedure for inputting an image composed of a multi-tone pixel group (2) A procedure for storing an input image (3) A stored image A procedure for normalizing at least a part of the region and generating a normalized image obtained by normalizing the region.

Images