JP7031584B2 - Area identification device and area identification method and area identification program - Google Patents

Description

The present invention relates to a technique for identifying an observation target area of an image taken on the ground, in the sky, in space, or the like.

A device that identifies a cloud region based on edge information of a cloud region of an image acquired from an artificial satellite (Patent Document 1), and an image processing device that identifies an observation target region such as a cloud from the brightness and saturation of an image (Patent Document 1). 2) is disclosed. In these devices, the observation target area is set while the feature amount for the user to identify the observation target area is viewed on the graph. In addition, an image identification device that identifies the category to which an image belongs by combining a region division method represented by the k-means method and machine learning represented by SVM (Support Vector Machine), and extracts an observation target region such as a cloud. (Patent Document 3) is disclosed.

Japanese Unexamined Patent Publication No. 5-333160 JP-A-2015-64753 International Publication No. 2012/11236 International Publication No. 2012/095938

However, in the apparatus disclosed in Patent Document 1, an object having edge information similar to the cloud region on the ground surface is erroneously determined as a cloud region. In addition, since clouds have various edge characteristics depending on the type, the cloud area is missed. The apparatus disclosed in Patent Document 2 is premised on a color image, and cannot be identified by a monochrome image such as an infrared image. Further, in these devices, the observation target area is set while the user sees the feature amount for identifying the observation target area on the graph. Therefore, the feature amount is limited to about two, and it is difficult to set detailed boundary conditions.

As a result of the above, there is a problem that these devices cannot improve the identification accuracy of the observation target area. Further, Patent Document 1 and Patent Document 2 refer to the case of processing the input still image, but do not refer to the real-time processing of moving images and the like.

On the other hand, in the apparatus disclosed in Patent Document 3, the identification accuracy of the observation target area can be improved. However, since this device combines a region division method represented by the k-means method and machine learning represented by SVM, the amount of calculation becomes enormous, and there is a problem that it is not suitable for real-time processing such as moving images. be.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an area identification device capable of real-time area identification with high accuracy regardless of whether the image is color or monochrome. To do.

The region identification device of the present invention divides an input image into local regions, and derives one or more of the feature quantities that can be derived for each local region based on the pixel information of the image. An observation target area determination means for determining whether or not each local region is an observation target by comparing the derived means with the derived feature amount and a criterion for determining the observation target area set in a predetermined dictionary. Have.

In the region identification method of the present invention, the input image is divided into local regions, and one or more of the feature quantities that can be derived based on the pixel information of the image is derived and derived for each local region. It is determined whether or not the feature amount is an observation target for each local region by comparing the feature amount with the reference for determining the observation target region set in a predetermined dictionary.

The region identification program of the present invention derives one or more of a process of dividing an input image into local regions and a feature amount that can be derived for each local region based on the pixel information of the image. Have the computer execute the process and the process of comparing the derived feature amount with the criterion for determining the observation target area set in the predetermined dictionary and determining whether or not the feature is the observation target for each local area. ..

According to the present invention, it is possible to provide a region identification device that enables real-time region identification with high accuracy regardless of whether the image is color or monochrome.

It is a block diagram which shows the structure of the area identification apparatus of 1st Embodiment of this invention. It is a block diagram which shows the structure of the area identification apparatus of 2nd Embodiment of this invention. It is a flowchart which shows the operation which sets the determination criterion of the observation target area of the area identification apparatus of 2nd Embodiment of this invention. It is a figure which shows the example of the image for learning which sets the determination criterion of the observation target area of the area identification apparatus of the 2nd Embodiment of this invention. It is a figure which shows the example which divided the image for learning which sets the determination criterion of the observation target area of the area identification apparatus of 2nd Embodiment of this invention into a block. It is a figure which shows the example of the observation target correct area information of the image for learning which sets the determination criterion of the observation target area of the area identification apparatus of the 2nd Embodiment of this invention. It is a figure which shows the example which the area identification apparatus of the 2nd Embodiment of this invention associated a feature quantity with an observation target correct answer area by machine learning. It is a figure which shows the example which the area identification apparatus of the 2nd Embodiment of this invention represented the relationship between the feature amount and the observation object correct answer area in the coordinate system of the feature amount, and set the boundary of the observation object determination criteria. It is a figure which shows the example of the boundary of the determination criteria of the observation target set by the area identification apparatus of the 2nd Embodiment of this invention. It is a flowchart which shows the operation which identifies the observation target area of the area identification apparatus of 2nd Embodiment of this invention. It is a figure which shows the example which divided the image which identifies the observation target area of the area identification apparatus of 2nd Embodiment of this invention into a block. It is a flowchart which shows the operation of discriminating three or more kinds of regions of the region identification apparatus of 2nd Embodiment of this invention. It is a figure which shows the example of the boundary of the determination criteria which discriminates three or more kinds of regions set by the region identification apparatus of the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the area identification apparatus of 3rd Embodiment of this invention. It is a block diagram which shows the structure of the area identification apparatus of 4th Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. However, although the embodiments described below have technically preferable limitations for carrying out the present invention, the scope of the invention is not limited to the following.
(First Embodiment)
FIG. 1 is a block diagram showing a configuration of a region identification device according to a first embodiment of the present invention. The area identification device 1 of the present embodiment divides the input image into local regions, and derives one or more of the feature quantities that can be derived based on the pixel information of the image for each local region. It has a feature amount deriving means 11. Further, it has an observation target area determination means 12 for determining whether or not each local region is an observation target by comparing the derived feature amount with a standard for determining an observation target region set in a predetermined dictionary.

According to the area identification device 1 of the present embodiment, it is not necessary for the user to set the target area while looking at the feature amount on the graph, and it is possible to set detailed boundary conditions with a wide variety of feature amounts. As a result, the identification accuracy of the target area can be improved regardless of whether it is a color image or a monochrome image. Further, since the target area can be identified only by comparing the feature amount of the image with the predetermined dictionary data, the calculation amount at the time of identification can be small, so that the identification in real time such as a moving image is possible.

As described above, according to the present embodiment, it is possible to provide an area identification device that enables real-time area identification with high accuracy regardless of whether the image is color or monochrome.
(Second embodiment)
FIG. 2 is a block diagram showing a configuration of the area identification device 2 according to the second embodiment of the present invention. The area identification device 2 includes an image input unit 21, a feature quantity derivation unit 22, a dictionary storage unit 23, an observation target area determination unit 24, an observation target area information output unit 25, and a learning unit 26.

The image input unit 21 inputs image data of an image for identifying an observation target area such as a cloud.

The feature amount derivation unit 22 divides the image input by the image input unit 21 into local regions, and calculates the feature amount based on the pixel information of the image for each local region.

The dictionary storage unit 23 stores the dictionary data in which the determination criteria of the observation target area are set, and inputs the dictionary data to the observation target area determination unit 24.

The observation target area determination unit 24 compares the feature amount calculated by the feature amount derivation unit 22 with the standard for determining the observation target area set by the dictionary data input from the dictionary storage unit 23, and observes the observation target. Judge whether or not.

The observation target area information output unit 25 outputs the determination result of the observation target area determination unit 24 by expressing it in an image that identifies the observation target area such as clouds.

The learning unit 26 generates dictionary data in which the determination criteria of the observation target area are set.

The area identification device 2 can be an information processing device (computer) such as a PC (Personal Computer) or a server. Information processing equipment includes a CPU (Central Processing Unit) as a computing resource, a memory and an HDD (Hard Disk Drive) as a storage resource, a communication board and an input interface as a communication resource, and a keyboard and a mouse as an input resource. It is equipped with a touch panel, displays and printers that are display resources. By operating the program on the CPU and using these components, it is possible to realize each part constituting the area identification device 1.

That is, the image input unit 21 is realized by a communication unit that receives image data transmitted from a satellite or an aircraft, an input interface that inputs image data from the outside, and the like. The feature amount derivation unit 22, the observation target area determination unit 24, and the learning unit 26 are realized by operating a program on the CPU. The dictionary storage unit 23 is realized by a memory, an HDD, or the like. The observation target area information output unit 25 is realized by a display, a printer, or the like.

FIG. 3 is a flowchart showing an operation of setting a determination criterion of an observation target area, which is executed by the area identification device of the present embodiment. This flowchart is started by starting a program that executes an operation of setting a criterion for an observation target area.

First, in step S01, a learning image showing the observation target is input to the learning unit 26 from the outside. FIG. 4 is a diagram showing an example of an image for learning when the observation target is a cloud. As the image for learning, for example, if the observation target is a cloud, a plurality of images of various clouds, for example, several to several tens of images are input.

The image for learning is preferably, but is not limited to, the image that actually identifies the observation target area and the camera to be photographed, the image-taking conditions, the weather conditions, the time, the season, and the like. When it is difficult to obtain an image satisfying these conditions, an image or a simulated image different from some of these conditions can be used.

Further, the image for learning is preferably an unprocessed image (also referred to as a RAW image) that has not been image-processed in order to increase the resolution of the feature amount calculated later. This is because even if the raw image has a resolution of, for example, ¹⁴ bits (214), the resolution of the feature amount is lowered by reducing the resolution to, for example, ⁸ bits (28) by image processing. This is because it will be.

Next, in step S02, the learning unit 26 calculates the feature amount for each local region of the input learning image from the pixel information of the image data. The type of the feature amount may be the same as the feature amount used in the image that actually identifies the observation target area, and may be one or more feature amounts. For example, when the image is a visible image, the feature amount-1 can be the average luminance and the feature amount-2 can be the dispersion of the luminance, but the feature amount is not limited to these. The type of feature amount can be predetermined.

FIG. 5 shows an example in which the learning unit 26 divides the image into blocks in order to calculate the feature amount of the local region of the image for learning. The learning unit 26 calculates the feature amount of each block as the feature amount of the local region.

Next, in step S03, the observation target correct answer area information of the observation target in the input image is input to the learning unit 26. FIG. 6 shows a binarized image obtained by binarizing the cloud portion and the other portion, which is the observation target correct area information. The cloud area of this binarized image, that is, the white area in FIG. 6 is the observation target correct answer area. This binarized image can be created in advance by the user by designating a cloud portion from the learning image or by performing image processing on the binarized image.

Next, in step S04, the learning unit 26 performs machine learning processing from the feature amount of the local region of the learning image and the observation target correct answer region information. As machine learning, for example, a method with a small amount of calculation at the time of identification, such as GLVQ (Generalized Learning Vector Quantization) disclosed in Patent Document 4, is desirable. According to GLVQ, the feature amount of each block calculated in step S02 is associated with the observation target correct answer area of the observation target correct answer area information input in step S03, and the feature amount for becoming the observation target correct answer area is set. be able to.

The machine learning method is a method that can set the feature amount to be the observation target correct answer area by associating the feature amount of each block with the observation target correct answer area of the observation target correct answer area information. , Not limited to GLVQ.

FIG. 7 shows an example in which the learning unit 26 associates the feature amount with the observation target correct answer region (for example, a cloud region and a region other than the cloud) by machine learning. FIG. 7 shows a case where two learning images are input to make 1000 local regions per image, and two types of feature quantities are calculated for each local region, but the present invention is not limited to this. ..

The set of the feature amount -1 and the feature amount -2 of each local region of the image-1 is (a ₁ , b ₁ ), (a ₂ , b ₂ ), ..., (A ₁₀₀₀ , b ₁₀₀₀ ). , It is determined whether each set belongs to 1 (cloud region) or 0 (region other than cloud) of the observation target correct answer class. Further, the set of the feature amount -1 and the feature amount -2 of each local region of the image-2 is (a ₁₀₀₁ , b ₁₀₀₁ ), ..., (A ₂₀₀₀ , b ₂₀₀₀ ), and each set is. It is determined whether it belongs to 1 or 0 of the observation target correct answer class. It should be noted that the numerical values of the feature quantities derived by calculation or the like are input to a ₁ to a ₂₀₀₀ and b ₁ to b ₂₀₀₀ , respectively.

Next, in step S05, the learning unit 26 generates dictionary data in which the criteria for identifying the observation target area by the feature amount is set from the correspondence between the observation target area and the feature amount obtained by the above machine learning process. ..

In FIG. 8, the learning unit 26 represents the relationship between the feature amount of each local region of FIG. 7 and the observation target correct answer region in the coordinate system of the feature amount-1 and the feature amount-2, and sets the boundary of the observation target determination criterion. An example is shown. Each of the feature quantities of image-1 and image-2 is collectively represented in the coordinate system of feature quantity-1 and feature quantity-2. The boundary is set so that the discrepancy between the classification of class 1 and class 0 according to the provided boundary and the classification of FIG. 7 is minimized by machine learning using, for example, GLVQ. Therefore, there may be a discrepancy between the classification of class 1 and class 0 according to the boundary provided by GLVQ and the classification of FIG. 7.

According to GLVQ, a misclassification scale is used as an index to express a predetermined classiness. The misclassification scale takes a value between -1 and 1, and in the case of the misclassification scale for class 1, the closer the misclassification scale is to -1, the higher the probability that the misclassification scale is class 1. On the contrary, the closer the misclassification scale is to 1, the higher the probability that it is not class 1, that is, it is class 0. The boundary between class 1 and class 0 is usually set to the misclassification scale 0 and is a region that tends to contain ambiguity. Also, for example, if you want to reliably determine the cloud part as a cloud even if the sky part is a little wrong as a cloud, you can make adjustments such as identifying a positive value with a misclassification scale for clouds greater than 0 as a threshold value. Is possible.

FIG. 9 shows an example of the boundary of the observation target determination standard set by the learning unit 26. The learning unit 26 sets a boundary between the observation target and the other in the coordinate system of the feature amount -1 and the feature amount -2 as shown in FIG. Generate dictionary data that can determine whether or not there is. The generated dictionary data is stored in the dictionary storage unit 23 and ends.

In FIG. 9, in order to bring the boundary for determining the observation target closer to the actual observation target, the feature having high separability between the above classes, that is, the feature amount near the boundary in FIG. 8 is small. It is useful to choose the amount. For this purpose, it is possible to execute the above operation of determining the boundary with several kinds of features in advance, and select a feature amount that can determine the boundary with as little ambiguity as possible.

FIG. 10 is a flowchart showing an operation for identifying an observation target area executed by the area identification device of the present embodiment. This flowchart is started by starting a program that executes an operation for identifying an observation target area.

First, in step S11, an image that identifies the observation target area is input to the image input unit 21 from the outside. The input of the image can be directly input to the image captured by the camera, but the input is not limited to this. The image may be input as an image stored in the storage medium.

The image that identifies the observation target area is preferably an unprocessed image that has not been image-processed in order to improve the resolution of the feature amount calculated later. This is because even if the raw image has a resolution of, for example, 14 bits, the resolution of the feature amount is lowered by reducing the resolution to, for example, 8 bits by image processing.

Next, in step S12, the feature amount derivation unit 22 divides the image into blocks, sets each block as a local area, and calculates the feature amount for each local area. FIG. 11 is a diagram showing an example in which an image identifying an observation target area is divided into blocks. As shown in FIG. 11, for example, 8 vertical pixels and 8 horizontal pixels can be one block, but the block is not limited thereto. The feature amount derivation unit 22 can use, for example, the average value of the feature amounts of each pixel as the feature amount of the block. Further, the dispersion of the feature amount of each pixel can be used as the feature amount of the block, but the feature amount is not limited thereto.

Further, the method of calculating the feature amount for each local region is not limited to the method of dividing into blocks as shown in FIG. For example, a method may be used in which 8 vertical pixels and 8 horizontal pixels are set as one local area, and the feature amount is continuously calculated while scanning by shifting this area by 1 pixel. Further, the feature amount can be one or more feature amounts. For example, when the image is a visible image, the feature quantities can be two of saturation and lightness, two of luminance and luminance dispersion, or all four of these, but are not limited thereto. The type of feature amount can be predetermined.

Next, in step S13, the dictionary storage unit 23 inputs the dictionary data generated by the learning unit 26 to the observation target area determination unit 24. The observation target area determination unit 24 observes each divided region by comparing the feature amount for each local region obtained by the feature quantity derivation unit 22 with the determination standard for the observation target region set in the dictionary data. Determine if it is a target.

Next, in step S14, the observation target area information output unit 25 outputs the determination result in step S13 as, for example, a binary image in which the observation target is 1 and the non-observation target is 0, and the process ends.

FIG. 12 is a flowchart showing an operation of determining three areas executed by the area identification device 2 of the present embodiment. Here, a case where clouds, the sea, and other regions are identified as observation target regions will be described.

The learning unit 26 sets the determination criteria of the observation target area shown in FIG. 3 in advance, and obtains the observation target correct area information for each of the sea and the cloud in the learning image in FIG. 13 from the learning image. Generate dictionary data with the criteria for the observation target area as shown.

FIG. 13 is a diagram showing an example of a boundary of a criterion for identifying three regions. The learning unit 26 sets the boundary between the observation target-1 (cloud), the observation target-2 (sea), and the non-observation target in the coordinate system of the feature amount -1 and the feature amount -2 as shown in FIG. , Generate dictionary data that can identify the observation target by the feature quantity using this boundary as a criterion. The learning unit 26 stores the generated dictionary data in the dictionary storage unit 23.

The flowchart of FIG. 12 is started by starting a program that executes an operation of identifying an observation target area. First, in step S21, an image that identifies the observation target area is input to the image input unit 21. Next, in step S22, the feature amount derivation unit 22 divides the image into blocks, sets each block as a local area, and calculates the feature amount for each local area.

Next, in step S23, the dictionary storage unit 23 inputs the dictionary data generated by the learning unit 26 to the observation target area determination unit 24. The observation target area determination unit 24 compares the feature amount for each local region obtained by the feature amount derivation unit 22 with the determination standard for the observation target area set in the dictionary data, and thereby clouds for each divided region. Determine whether it is the sea or something else.

Next, in step S24, the observation target area information output unit 25 sets the determination result in step S23 as, for example, a ternary image in which the cloud area is 1, the sea area is 2, and the other areas are 0, and the result is externalized. Output and exit.

By the above method, for example, when a moving body is present in an image identifying an observation target area, or when the place where the moving body is located is the sea, the moving body can be estimated to be a ship. Further, when the place where the moving body exists is a cloud, the moving body can be estimated as a flying body.

The observation target area is not limited to one area or two areas as described above. The number of observation target areas can be any positive number, and dictionary data that can identify the observation target corresponding to the observation target area may be generated in advance.

The observation target area is an image acquired from the air, ground, or underwater, such as clouds, sea, sky, land, ice, snow, mountains, grasslands, forests, deserts, coral reefs, and structures such as buildings and roads. It can be an area inside. Further, the observation target area is not limited to these, and may be a region in an image acquired from a room or a living body, or a region in an image magnified by a microscope.

The feature amount can be saturation, brightness, luminance, luminance dispersion for a visible image, and luminance, luminance dispersion, etc. for an infrared image. In addition, it can be a feature amount described by a histogram of the intensity of each pixel in a local region such as a block, a state of distribution (pattern), SIFT (Scale-Invariant Features Transition Form), and HOG (Histograms of Oriented Gradients). .. The feature amount is derived by extracting a value from the information such as saturation, brightness, and brightness of the pixel of the image, and calculating using the extracted value.

According to the area identification device 2 of the present embodiment, the user does not need to set the target area while looking at the feature amount on the graph, and detailed boundary conditions can be set by a wide variety of feature amounts. As a result, the identification accuracy of the target area can be improved regardless of whether it is a color image or a monochrome image. Further, since the target area can be identified only by comparing the feature amount of the image with the predetermined dictionary data, the calculation amount at the time of identification can be small, so that the identification in real time such as a moving image is possible.

As described above, according to the present embodiment, it is possible to provide an area identification device that enables real-time area identification with high accuracy regardless of whether the image is color or monochrome.
(Third embodiment)
FIG. 14 is a block diagram showing the configuration of the area identification device 3 according to the third embodiment of the present invention. The area identification device 3 includes a first image input unit 31-1, a second image input unit 31-2, a feature quantity derivation unit 32, a dictionary storage unit 33, an observation target area determination unit 34, and an observation target area information output unit. It has 35 and a learning unit 36. The first aspect of the area identification device 3 of the present embodiment is that it is provided with a plurality of image input units in order to input two or more types of images of scenes including the same observation target at the same time to identify the area. It is different from the second embodiment.

FIG. 14 shows a case where two types of images are input to identify an area. First, two types of images obtained by capturing the same scene at the same time, for example, a visible image is input to the first image input unit 31-1 and an infrared image is input to the second image input unit 31-2. The feature amount derivation unit 32 divides each image into blocks and calculates the feature amount for each local region.

The two types of image features can be, for example, four types: saturation and brightness of a visible image, average luminance and luminance dispersion of an infrared image. The learning unit 36 machine-learns the four types of features in advance, and generates dictionary data for setting boundaries in the four-dimensional feature space. The dictionary storage unit 33 stores the dictionary data generated by the learning unit 36 and inputs it to the observation target area determination unit 34.

The observation target area determination unit 34 observes each divided region by comparing the feature amount of the local region obtained by the feature amount derivation unit 32 with the determination standard of the observation target area set in the dictionary data. Determine if it is a target.

The observation target area information output unit 35 outputs the determination result of the observation target area determination unit 34 as, for example, a binary image in which the observation target is 1 and the non-observation target is 0.

As described above, by identifying the observation area from a plurality of different images such as a visible image and an infrared image, for example, a region such as clouds and ice or snow that is difficult to identify only by the visible image can be combined with the infrared image. This enables accurate identification.

The image input unit is not limited to two. The number of image input units can be any positive number. Further, a plurality of images may be input from one image input unit.

According to the area identification device 3 of the present embodiment, the user does not need to set the target area while looking at the feature amount on the graph, and detailed boundary conditions can be set by a wide variety of feature amounts. As a result, the identification accuracy of the target area can be improved regardless of whether it is a color image or a monochrome image. Further, since the target area can be identified only by comparing the feature amount of the image with the predetermined dictionary data, the calculation amount at the time of identification can be small, so that the real-time identification of a moving image or the like is possible.

As described above, according to the present embodiment, it is possible to provide an area identification device that enables real-time area identification with high accuracy regardless of whether the image is color or monochrome.
(Fourth Embodiment)
FIG. 15 is a block diagram showing the configuration of the area identification device 4 according to the fourth embodiment of the present invention. The area identification device 4 includes a first image input unit 41-1, a second image input unit 41-2, a feature quantity derivation unit 42, a dictionary storage unit 43, an observation target area determination unit 44, and an observation target area information output unit. It includes 45, a learning unit 46, and a dictionary selection unit 47. The area identification device 4 of the present embodiment is different from the third embodiment in that it includes a dictionary selection unit 47 in order to properly use a plurality of dictionaries depending on the image shooting conditions.

Here, a case will be described in which a cloud region in an image is identified in real time in a scene in which a two-wavelength infrared camera is used to capture the horizon direction from an aircraft.

First, the a-wavelength infrared image 1 is input to the first image input unit 41-1 and the b-wavelength infrared image 2 is input to the second image input unit 41-2. Here, the two types of images are taken at almost the same time, and may be taken by switching filters with one camera and taking two wavelengths alternately, or may use two cameras. For each image, an unprocessed image is used in order to obtain brightness corresponding to the physical properties and temperature of the cloud region to be observed.

The feature amount derivation unit 42 divides each image into blocks and calculates the feature amount for each local region. As the feature amount of the two types of images, for example, a four-dimensional feature amount of the average luminance and the dispersion of the luminance of each of the two wavelength infrared images can be used.

Here, for example, the a-wavelength infrared image 1 is an infrared image having a wavelength of 3 to 4 μm, and the b-wavelength infrared image 2 is an infrared image having a wavelength of 4 to 5 μm. Infrared images are greatly affected by sunlight when the wavelength is 3 to 4 μm. Therefore, the influence of sunlight can be extracted by taking the difference between the infrared image of 3 to 4 μm and the infrared image of 4 to 5 μm. This makes it possible to eliminate the influence of sunlight and identify the observation target area. The infrared wavelength of the infrared image is not limited to the above combination.

The learning unit 46 learns four types of features in advance and generates dictionary data for setting boundaries in a four-dimensional feature space. Here, as the learning image, a plurality of sets of images for each shooting condition are prepared, and dictionary data is created for each shooting condition. Examples of the shooting conditions include weather conditions, time of day and night, and seasons. The dictionary storage unit 43 stores the dictionary data generated by the learning unit 46. The dictionary selection unit 47 selects a dictionary having the closest shooting conditions to the conditions at the time of actual shooting, and inputs the dictionary to the observation target area determination unit 44.

The observation target area determination unit 44 compares the feature amount of the local region obtained by the feature amount derivation unit 42 with the determination standard of the observation target area set by the dictionary data, and thereby observes each divided region. To identify.

The observation target area information output unit 45 outputs the determination result of the observation target area determination unit 44 as, for example, a binary image in which the observation target is 1 and the non-observation target is 0.

As described above, by selecting the dictionary that has the closest shooting conditions to the shooting conditions of the image for which the observation target area is to be identified and identifying the observation target, it is possible to suppress erroneous identification caused by changes in the shooting conditions. Can be done.

According to the area identification device 4 of the present embodiment, the user does not need to set the target area while looking at the feature amount on the graph, and detailed boundary conditions can be set by a wide variety of feature amounts. As a result, it is possible to improve the identification accuracy of the target area regardless of whether it is a color image or a monochrome image. Further, since the target area can be identified only by comparing the feature amount of the image with the predetermined dictionary data, the calculation amount at the time of identification can be small, so that the identification in real time such as a moving image is possible.

As described above, according to the present embodiment, it is possible to provide an area identification device that enables real-time area identification with high accuracy regardless of whether the image is color or monochrome.

Although the invention of the present application has been described above with reference to the embodiments, the invention of the present application is not limited to the above-described embodiment. Various changes that can be understood by those skilled in the art can be made within the scope of the present invention in terms of the configuration and details of the present invention.

In addition, some or all of the above embodiments may be described as in the following appendix, but are not limited to the following.
(Appendix 1)
A feature quantity deriving means that divides the input image into local regions and derives one or more of the feature quantities that can be derived based on the pixel information of the image for each local region.
Area identification having an observation target area determination means for determining whether or not each local area is an observation target by comparing the derived feature amount with a criterion for determining an observation target area set in a predetermined dictionary. Device.
(Appendix 2)
It has a learning means for creating the dictionary for setting the observation target area of the feature amount based on the feature amount of the learning image having the observation target and the correct answer region information of the observation target of the learning image. The area identification device according to Appendix 1.
(Appendix 3)
The area identification device according to Appendix 2, wherein the learning means performs machine learning to set the observation target area.
(Appendix 4)
The area identification device according to item 1 of Addendums 1 to 3, which has a dictionary selection means for selecting the dictionary based on the shooting conditions of the image.
(Appendix 5)
The area identification device according to item 1 of Supplementary note 1 to 4, which has an image input means for inputting the image.
(Appendix 6)
The area identification device according to Appendix 5, wherein the image input means is plural.
(Appendix 7)
Divide the input image into local areas and
One or more of the feature quantities that can be derived based on the pixel information of the image for each local region are derived.
A region identification method for determining whether or not each local region is an observation target by comparing the derived feature amount with a criterion for determining an observation target region set in a predetermined dictionary.
(Appendix 8)
Addendum 7 to create the dictionary for setting the observation target area of the feature amount based on the feature amount of the learning image having the observation target and the correct answer region information of the observation target of the learning image. Area identification method.
(Appendix 9)
The area identification method according to Appendix 7 or 8, wherein the observation target area is set by machine learning.
(Appendix 10)
The area identification method according to item 1 of Appendix 7 to 9, wherein the dictionary is selected based on the shooting conditions of the image.
(Appendix 11)
The process of dividing the input image into local areas and
A process of deriving one or more of the feature quantities that can be derived based on the pixel information of the image for each local region.
An area identification program that causes a computer to execute a process of comparing the derived feature amount with a standard for determining an observation target area set in a predetermined dictionary and determining whether or not each local area is an observation target. ..
(Appendix 12)
The computer executes a process of creating the dictionary for setting the observation target area of the feature amount based on the feature amount of the learning image having the observation target and the correct answer area information of the observation target of the learning image. The area identification program according to Appendix 11.
(Appendix 13)
The area identification program according to Appendix 11 or 12, which causes a computer to execute a process of setting an observation target area by machine learning.
(Appendix 14)
The area identification program according to item 1 of Addendums 11 to 13, which causes a computer to execute a process of selecting the dictionary based on the shooting conditions of the image.

This application claims priority on the basis of Japanese application Japanese Patent Application No. 2016-122510 filed on June 21, 2016 and incorporates all of its disclosures herein.

1, 2, 3, 4 Area identification device 11 Feature quantity derivation means 12 Observation target area determination means 21 Image input unit 31-1, 41-1 First image input unit 31-2, 41-2 Second image input Units 22, 32, 42 Feature quantity derivation unit 23, 33, 43 Dictionary storage unit 24, 34, 44 Observation target area determination unit 25, 35, 45 Observation target area information output unit 26, 36, 46 Learning unit 47 Dictionary selection unit

Claims (10)

A feature amount deriving means for dividing an input image into local regions and deriving a plurality of types of feature quantities that can be derived based on the pixel information of the image for each local region.
Based on the boundary set in a predetermined dictionary and set for each of a plurality of observation target areas in the feature quantity space of the dimension of the number of the type, the reference and the derived feature quantity of the plurality of the types are used. A region identification device comprising an observation target region determination means for determining which observation target is to be observed for each of the local regions by comparing the above. It has a learning means for creating the dictionary for setting the observation target area of the feature amount based on the feature amount of the learning image having the observation target and the correct answer region information of the observation target of the learning image. The area identification device according to claim 1. The area identification device according to claim 2, wherein the learning means performs machine learning to set the observation target area. The area identification device according to claim 1, further comprising a dictionary selection means for selecting the dictionary based on the imaging conditions of the image. The area identification device according to claim 1, further comprising an image input means for inputting the image. The area identification device according to claim 5, wherein the image input means is a plurality of. Divide the input image into local areas and
A plurality of types of feature quantities that can be derived based on the pixel information of the image are derived for each local region.
Based on the boundary set in a predetermined dictionary and set for each of a plurality of observation target areas in the feature quantity space of the dimension of the number of the type, the reference and the derived feature quantity of the plurality of the types are used. A region identification method for determining which observation target is to be observed for each of the local regions. The seventh aspect of claim 7, wherein the dictionary for setting the observation target area of the feature amount is created based on the feature amount of the learning image having the observation target and the correct answer region information of the observation target of the learning image. Area identification method. The process of dividing the input image into local areas and
A process for deriving a plurality of types of features that can be derived based on the pixel information of the image for each local region, and a process for deriving the features.
Based on the boundary set in a predetermined dictionary and set for each of a plurality of observation target areas in the feature quantity space of the dimension of the number of the above type, the reference and the derived feature quantity of the plurality of the types are used. A region identification program that causes a computer to execute a process of comparing and determining which observation target is to be observed for each local region. The computer executes a process of creating the dictionary for setting the observation target area of the feature amount based on the feature amount of the learning image having the observation target and the correct answer area information of the observation target of the learning image. The area identification program according to claim 9.

Images