JP2014232477A - Subject specifying device, imaging device, subject specifying method, and subject specifying program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the accuracy of estimation of the position of a subject included in an image and extraction of the subject. A position estimation unit that estimates a position of a point of interest in a first image using a saliency map generated based on a feature amount of the input first image; luminance information and color of the first image; A binarization processing unit that binarizes a plurality of second images generated based on information using luminance information and color information, and generates a plurality of binarized images; an estimated position of a target point; And a subject estimation unit that estimates a subject included in the first image from a plurality of binarized images. [Selection] Figure 2

Description

  The present invention relates to a subject specifying device, an imaging device, a subject specifying method, and a subject specifying program.

  Some commonly provided imaging devices specify a subject position based on an AF area selected by a user and perform focus adjustment processing on the specified subject. Such an imaging apparatus performs focus adjustment processing based on an area selected by the user, and does not specify the size or shape of the subject. For this reason, it has been proposed to binarize the image divided into the luminance component and the color difference component and specify the position, shape, and size of the subject from the binarized image (see Patent Document 1). ).

Japanese Patent Laid-Open No. 2004-20585

  In this patent document 1, an estimated subject position consisting of a position designated by the user and a face position detected from the image is used as a reference for each white pixel area included in the binarized image. An evaluation value is obtained, and a white pixel region having a high evaluation value is specified as a subject. However, if the user does not specify the subject estimation position or the face position, the accuracy of the evaluation value for the white pixel area is poor, and the subject cannot be specified accurately.

  An object of the present invention is to provide a subject specifying device, an imaging device, a subject specifying method, and a subject specifying program capable of improving the accuracy of estimating the position of a subject included in an image and extracting the subject.

  In order to solve the above-described problem, the subject specifying device of the present invention estimates the position of a point of interest in the first image using a saliency map generated based on the feature amount of the input first image. A plurality of second images generated on the basis of the luminance information and the color information of the first image, and binarizing the plurality of second images using the luminance information and the color information. A binarization processing unit to be generated, and a subject estimation unit that estimates a subject included in the first image from the estimated position of the target point and the plurality of binary images. And

  In the imaging apparatus of the present invention, an imaging optical system that captures subject light, an image sensor that outputs an image signal based on the captured subject light, and an image based on the image signal output from the image sensor are a first image. And a control unit that executes focus adjustment control of the imaging optical system using a region including the subject estimated by the subject specifying device. Features.

  Further, the subject specifying method of the present invention includes a position estimation procedure for estimating a position of a point of interest in the first image using a saliency map generated based on a feature amount of the input first image, A binarization procedure for binarizing a plurality of second images generated based on luminance information and color information of the first image using the luminance information and color information, and generating a plurality of binarized images; A subject estimation procedure for estimating a subject included in the first image from the estimated position of the target point and the plurality of binarized images.

  Further, the subject specifying program of the present invention uses a saliency map generated based on the input first image feature amount to estimate a position of a point of interest in the first image, A binarization procedure for binarizing a plurality of second images generated based on luminance information and color information of the first image using the luminance information and color information, and generating a plurality of binarized images; And a subject estimation procedure for estimating a subject included in the first image from the estimated position of the target point and the plurality of binarized images.

  ADVANTAGE OF THE INVENTION According to this invention, the precision of estimation of the position of the to-be-photographed object included in an image and extraction of a to-be-photographed object's outline can be improved.

It is a functional block diagram showing one embodiment of an imaging device. It is a functional block diagram which shows the structure of a to-be-identified part. 3A shows an image input to the subject specifying unit, FIG. 3B shows a Gabor map obtained from the luminance component of the image shown in FIG. 3A, and FIG. It is a figure which shows the gravity center position calculated | required from a Gabor map. It is a figure which shows an example of the sample image at the time of binarizing the Y image of the input image, Cb image, and Cr image with a respectively different threshold value. It is a flowchart which shows the flow of the process which specifies a to-be-photographed object. Of the samples shown in FIG. 4, an example of each sample when narrowing islands that are subject candidates is shown. It is a figure which shows AF area | region based on the result of a subject specific process.

  As shown in FIG. 1, the imaging apparatus 10 includes an imaging optical system 15, an imaging element 16, an A / D converter 17, a driver 18, a timing generator (TG) 19, a buffer memory 20, an image processing circuit 21, an encoding / coding circuit. A decoding circuit 22, a subject specifying circuit 23, a display control circuit 24, a monitor 25, a media controller 26, a CPU 30, a built-in memory 31, a release button 32, and a setting operation unit 33 are provided. Here, the A / D converter 17, buffer memory 20, image processing circuit 21, encoding / decoding circuit 22, display control circuit 23, media controller 26, CPU 30, and built-in memory 31 are connected via a bus 35. The

  The imaging optical system 15 includes a lens group including a zoom lens, a focus lens, and the like that are not shown. These zoom lens and focus lens are moved in the direction of the optical axis L by a lens driving mechanism (not shown). Note that subject light captured via the imaging optical system 15 is imaged on the light receiving surface of the imaging element 16.

  The image pickup device 16 is configured by, for example, a charge coupled device (CCD), a complementary metal-oxide semiconductor (CMOS), or the like. The image sensor 16 is driven and controlled by a driver 18. The drive control of the image sensor 16 includes controlling the accumulation of signal charges in each pixel of the image sensor 16 and the output of the accumulated signal charges of each pixel. The signal charge of each pixel is output to the A / D converter 17 as an image signal. The A / D converter 17 converts an input image signal from an analog signal to a digital signal. The digital image signal output from the A / D converter 17 is stored in the buffer memory 20.

  The timing generator 19 outputs a clock pulse to the A / D converter 17 and the driver 18. By outputting clock pulses from the timing generator 19, the drive timing of the image sensor 16 and the A / D converter 17 is synchronized.

  The buffer memory 20 is composed of a non-volatile memory such as EPPROM, for example. The buffer memory 20 temporarily receives the image signal output from the storage medium 40 in addition to the image signal output from the A / D converter 17 and the image signal subjected to image processing by the image processing circuit 21. Remembered.

  The image processing circuit 21 performs image processing such as white balance processing, color interpolation processing, contour compensation processing, and gamma processing on the image signal written in the buffer memory 20. Hereinafter, an image signal subjected to image processing is referred to as image data.

  The image processing circuit 21 performs resolution conversion processing on the image data. By this resolution conversion process, image data having a lower resolution than the original image data is generated. As an example of the low-resolution image data, there is image data (thumbnail image data) used when displaying a list of images. Further, the image processing circuit 21 executes resolution conversion processing on the image data so that the resolution is matched with the resolution of the monitor 25 when displaying an image based on the image data stored in the buffer memory 20 on the monitor 25. .

  Further, the image processing circuit 21 performs a process of specifying a subject for the image data acquired when the imaging standby state is set. Note that processing for specifying a subject (hereinafter, subject specifying processing) will be described later.

  The encoding / decoding circuit 22 reads the image data stored in the buffer memory 20 and executes an encoding process on the read image data. By this encoding process, for example, JPEG compressed image data is generated. The encoded image data is stored in the buffer memory 20. The encoding / decoding circuit 22 performs a decoding process on the image data that has been subjected to the encoding process.

  The subject specifying circuit 23 uses the image data stored in the buffer memory 20 to specify the position, shape, and size of the subject. Here, as the image data used in the subject specifying circuit, through image data obtained in the shooting standby state can be cited. Information on the position, shape and size of the subject specified by the subject specifying circuit 23 is output to the CPU 30.

  The display control circuit 24 reads the image data that has been subjected to resolution conversion processing by the image processing circuit 21. Then, the display control circuit 24 displays an image based on the read image data on the monitor 25. Here, examples of the monitor 25 include an LCD and an organic EL display.

  The media controller 26 is electrically connected to a storage medium 40 such as an optical disk or a magnetic disk, in addition to a memory card made of, for example, a flash memory. When the image data is stored in the storage medium 40, the CPU 30 generates shooting information, the model of the imaging device 10, shooting date and time, thumbnail image data, and the like as additional information, and generates the generated additional information. Then, the CPU 30 writes the encoded image data and supplementary information stored in the buffer memory 20 into the storage medium 40 as, for example, an EXIF format image file.

  CPU30 controls each part of the imaging device 10 centrally by executing the control program which abbreviate | omitted illustration. A release button 32 and a setting operation unit 33 are connected to the CPU 30, and the CPU 30 executes processing according to these operations. Examples of processing in the CPU 30 include AF processing and AE processing.

  Next, the configuration of the subject specifying circuit 23 will be described with reference to FIG. As illustrated in FIG. 2, the subject specifying circuit 23 includes a data generation unit 50, a target point estimation unit 51, and a subject estimation unit 52. Hereinafter, image data (YCbCr image data) shown in the YCbCr color space will be described as an example of image data input to the subject specifying circuit 23.

  The data generation unit 50 generates luminance component image data and color difference component image data from the input YCbCr image data. Hereinafter, the luminance component image data is referred to as Y image data, and the color difference component image data is referred to as Cb image data and Cr image data. The data generation unit 50 outputs Y image data among the generated image data to the attention point estimation unit 51. Further, the data generation unit 50 outputs Y image data, Cb image data, and Cr image data to the subject estimation unit 52. When image data represented in a color space other than the YCbCr color space is input, the data generation unit 50 performs color space conversion processing on the input image data, and then performs Y image data, Cb Image data and Cr image data may be generated.

  The attention point estimation unit 51 estimates the position of the attention point in the image using the input Y image data. This attention point is a point to be noted when the user observes the image. The attention point estimation unit 51 includes a filter processing unit 55, a binarization processing unit 56, and a position calculation unit 57.

  The filter processing unit 55 performs filter processing using, for example, a Gabor function on the input Y image data. As is well known, the Gabor function is a filter based on an object recognition method in human vision. The Gabor function is a product of a trigonometric function (sine function) and a Gaussian function. In the present embodiment, a two-dimensional Gabor function that is a product of a two-dimensional trigonometric function and a two-dimensional Gaussian function is used. Hereinafter, the expression (1) is an expression indicating a Gabor function.

  Here, x and y are coordinate values of an image, λ is a wavelength corresponding to a spatial frequency, θ is an azimuth (direction), ψ is a phase, σ is a standard deviation of a Gaussian distribution, and γ is an aspect ratio. The spatial frequency here is defined as a change in contrast between white and black (contrast) that falls within a viewing angle of 1 ° (degree) of the human eye, and its unit is cpd (cycler degree).

  The filter processing unit 55 convolves the Gabor function striped pattern represented by the above-described equation (1) with the image from a plurality of directions with respect to the Y image data, and adds the calculation results. Here, examples of the plurality of directions include directions of θ = 45 ° and θ = 135 °. 3A shows an example of an image input to the subject specifying circuit 23, and FIG. 3B is a convolution with the luminance (Y) component of the image shown in FIG. 3A. It is a map (hereinafter referred to as a Gabor map) generated when an operation is performed. This Gabor map is an example of a saliency map.

  The binarization processing unit 56 performs binarization processing on the Gabor map obtained by the filter processing unit 55. Note that a threshold value used in the binarization process is a threshold value A. The threshold A is made up of a value obtained by multiplying a statistical value such as an average value or standard deviation of each pixel in Y image data by a coefficient (for example, 3). Here, the value of the coefficient is a value set in advance by experiments, statistics, or the like.

  The position calculation unit 57 calculates the coordinates of the center of gravity of the white pixel included in the binarized Gabor map. The position calculation unit 57 uses the coordinates of the white pixels included in the binarized Gabor map to determine the coordinates of the position that is the center of gravity of the white pixels. In FIG. 3C, the centroid of the white pixel included in the binarized Gabor map is indicated by a mark “x”. In general, a position (hereinafter, a point of interest) in which an image is noticed by a user is not a center of the image but a subject included in the image or a region including the vicinity thereof. In addition, the region of the subject included in the image includes a higher frequency component than other regions. That is, when a Gabor map is generated from Y image data and the generated Gabor map is binarized, the subject area included in the image tends to appear as a white pixel area. Accordingly, the attention point estimation unit 51 estimates the centroid of the white pixel included in the binarized Gabor map calculated by the position calculation unit 57 as the attention point. The position calculation unit 57 outputs the coordinate data of the center of gravity in the labeling region to the subject estimation unit 52 as the coordinate data of the estimated point of interest.

  The subject estimation unit 52 estimates the subject using the input Y image data, Cb image data, and Cr image data, and the attention point coordinate data obtained by the attention point estimation unit 51. The subject estimation unit 52 includes a binarization processing unit 61 and a shape extraction unit 62.

  The binarization processing unit 61 performs binarization processing on each of the Y image data, Cb image data, and Cr image data. First, the binarization processing unit 61 obtains an average luminance value M and a standard deviation B of density values of each pixel from each data of Y image data, Cb image data, and Cr image data. Then, the binarization processing unit 61 calculates a threshold D used when the binarization process is performed using the obtained average luminance value M and the standard deviation B of the density values. The threshold value D may be calculated from the following equation (2).

[Equation 2]
D = M + α × B (2)
Here, the coefficient α is a value obtained in advance from experiments, statistics, and the like. Hereinafter, the case where α ₁ = 1.6, α ₂ = 1, α ₃ = −1.3 is used as the coefficient α will be described.

The binarization processing unit 61 executes binarization processing on each image data using the above-described equation (2). FIG. 4 shows a sample when the binarization process is performed on each of the Y image, the Cb image, and the Cr image using each of the above-described three types of coefficients α ₁ , α ₂ , and α ₃ . Here, the binarized images shown in Sample 1, Sample 2, and Sample 3 are binarized with a coefficient α = α ₁ (= 1.6) for the Y image, Cb image, and Cr image. It is a binarized image obtained when executed. The binarized images shown in Sample 4, Sample 5 and Sample 6 are obtained when binarization processing with coefficient α = α ₂ (= 1) is executed on the Y image, Cb image and Cr image. It is the binarized image obtained. Furthermore, the binarized images shown in Sample 7, Sample 8 and Sample 9 are binarized with a coefficient α = α ₃ (= −1.3) for the Y image, Cb image and Cr image. It is a binarized image obtained when executed.

  The shape extraction unit 62 performs a labeling process on the acquired binarized image, and extracts a group of white pixels as a labeling region. Hereinafter, the labeling region extracted by the shape extraction unit 62 is referred to as an island. The shape extraction unit 62 excludes, for example, islands in which the area ratio of the total area to the total area of the binarized image is 60% or more and islands in which the area ratio is 1% or less among the extracted islands. In addition, the shape extraction unit 62 excludes the island at the left end or the right end of the image from the extracted islands. After these processes, the shape extraction unit 62 sets the remaining islands that are not excluded as islands that are subject candidates.

  The shape extraction unit 62 obtains a rectangular area (hereinafter referred to as a rectangular area) including islands that are subject candidates for each island. Then, the shape extraction unit 62 narrows down the islands that are subject candidates by the area ratio (ratio) occupied by white pixels with respect to the rectangular area and the aspect ratio in the rectangular area for each obtained rectangular area. Obtained as an evaluation value of 1 and a second evaluation value. The shape extraction unit 62 uses the calculated evaluation value to narrow down islands that are subject candidates. As the narrowing-down process, a process for determining whether or not the first evaluation value is equal to or less than a preset threshold value (for example, 0.2) and a second evaluation value for a predetermined range (for example, 0.2). And less than 5).

  For example, by performing a process of determining whether or not the first evaluation value is equal to or less than a preset threshold value, an island that has irregularities or cavities that are not normally possible as a subject is excluded from islands that are subject candidates. Can do. In addition, by performing a process of determining whether or not the second evaluation value obtained for each target island is included in a predetermined range, an elongated island that cannot be a subject is excluded from the subject candidate islands. be able to.

  In addition, after narrowing down the islands that are subject candidates, the shape extracting unit 62 calculates the area of the island that is the subject candidate and the moment of inertia with the attention point estimated by the attention point estimation unit 51 as the center. Here, the method of calculating the moment of inertia is well known and will not be described in detail. For example, it can be calculated by the sum of the square of the pixel distance from the target point × (0 or 1).

  The shape extraction unit 62 calculates an evaluation value for shape extraction by dividing the area of the island as a subject candidate by the moment of inertia. Then, of the islands that are subject candidates, the island with the largest evaluation value for shape extraction is extracted as the island corresponding to the subject.

  Next, an embodiment of a process for extracting a subject will be described based on the flowchart of FIG. Hereinafter, the process from step S102 to step S105 is a process of estimating the attention point. The processing from step S106 to step S111 is processing for estimating the shape of the subject.

  Step S101 is processing for generating Y image data, Cb image data, and Cr image data. The data generation unit 50 generates Y image data, Cb image data, and Cr image data using the input YCbCr image data. Then, the data generation unit 50 outputs Y image data among the generated image data to the attention point estimation unit 51. At the same time, the data generation unit 50 outputs Y image data, Cb image data, and Cr image data to the subject estimation unit 52.

  Step S102 is processing for generating a Gabor map. The filter processing unit 55 performs filter processing using a Gabor function on the input Y image data. More specifically, the Gabor function stripe pattern represented by the above-described equation (1) is convolved with the image from a plurality of directions with respect to the Y image data, and the calculation results are added (calculation of the convolution sum). Here, the two directions refer to directions of θ = 45 ° and 135 ° in the equation (1). A Gabor map based on the input Y image data is generated by performing the process of calculating the convolution sum for all the pixels (see FIG. 3B).

  Step S103 is a process for binarizing the Gabor map. The binarization processing unit 56 executes binarization processing using the threshold A on the generated Gabor map.

  Step S104 is a process of calculating the coordinates of the center of gravity included in the binarized Gabor map. The position calculation unit 57 uses the coordinates of the white pixels included in the binarized Gabor map to determine the coordinates of the position that is the center of gravity of the white pixels. After this processing, the position calculation unit 57 outputs the obtained coordinate data of the center of gravity of the white pixel to the subject estimation unit 52 as the coordinate data of the estimated point of interest.

Step S105 is a process of performing binarization processing on the Y image data, Cb image data, and Cr image data. As described above, the subject image estimation unit 52 receives Y image data, Cb image data, and Cr image data. First, the binarization processing unit 61 obtains an average luminance value M and a standard deviation B of density values of input Y image data, Cb image data, and Cr image data. Then, the binarization processing unit 61 obtains the threshold value D by using the obtained average luminance value M and standard deviation B of the density value and the expression 2). Here, as the coefficient α, for example, values of α ₁ = + 1.6, α ₂ = 1, α ₃ = −1.3 are used. Then, the binarization processing unit 61 binarizes each image data using the obtained threshold value D. Thereby, a plurality of binarized image data is generated.

  Step S106 is a labeling process. The shape extraction unit 62 performs a labeling process on each of the binarized image data. A plurality of islands are extracted by this labeling process. The shape extraction unit 62 excludes, for example, an island having an area ratio of 60% or more with respect to the total area of the binarized image or an island having an area ratio of 1% or less from among the extracted islands. Further, the shape extraction unit 62 excludes the island at the left end or the right end of the binarized image. Then, the shape extraction unit 62 sets the remaining islands that are not excluded as islands that are subject candidates.

  Step S107 is processing to calculate the first and second evaluation values. The shape extraction unit 62 obtains a rectangular area including the island for each of the islands that are subject candidates. Next, the shape extraction unit 62 determines, for each obtained rectangular area, the ratio of the white pixels to the rectangular area (first evaluation value) and the aspect ratio of the rectangular area (second evaluation value). And ask.

  Step S108 is processing to narrow down islands that are subject candidates using the first and second evaluation values. First, the shape extraction unit 62 determines whether or not the first evaluation value obtained for each island that is a subject candidate is equal to or less than a preset threshold value (for example, 0.2). Then, the shape extraction unit 62 excludes islands whose first evaluation value is equal to or less than the threshold from the islands that are subject candidates.

  Next, the shape extraction unit 62 determines whether or not the second evaluation value for an island that is a subject candidate excluding the excluded island is included in a predetermined range (for example, 0.2 or more and less than 5). Then, the shape extraction unit 62 excludes islands whose second evaluation values are included in a predetermined range from islands that are subject candidates. By performing the processing in step S108, islands that are subject candidates are narrowed down.

  Step S109 is processing for calculating the area and moment of inertia of the island that is the subject candidate. The shape extraction unit 62 obtains the area of islands that are subject candidates narrowed down by the processing in step S108. Further, the shape extraction unit 62 obtains the moment of inertia using the binarized image data in which the island that is the subject candidate is detected. Here, the coordinate data of the center of gravity obtained in step S104 is input to the subject estimation unit 52 as the coordinate data of the point of interest. The shape extraction unit 62 obtains the inertia moment of the island using the coordinate data of the target point.

  Step S110 is processing for obtaining an evaluation value for shape extraction. The shape extraction unit 62 calculates an evaluation value for shape extraction using the island area and the inertia moment of the island obtained in step S109. Specifically, the shape extraction unit 62 calculates an evaluation value for shape estimation by dividing the area of the island by the inertia moment of the island.

  Step S111 is processing for extracting a shape. The shape extraction unit 62 estimates the island having the maximum shape extraction evaluation value obtained in step S110 among the islands that are subject candidates as the subject region, and uses the island shape as the subject shape. Extract.

  As illustrated in FIG. 4, for example, one or more labeling regions are detected in each of the binarized images obtained by binarizing the Y image, the Cb image, and the Cr image using different threshold values. Here, by performing the above-described processing of step S106 to step S109, the binarized image in which the island as the subject candidate exists becomes the binarized image of the sample 1 and the sample 3 (see FIG. 6). In FIG. 6, reference numeral 70 shown in the binarized image of sample 1 and reference numeral 71 shown in the binarized image of sample 3 are islands that are subject candidates. In each binarized image, the mark “×” is the attention point obtained by the attention point estimation unit 51. Here, the value of the moment of inertia increases as the distance from the attention point increases, for example. That is, the evaluation value for shape extraction decreases as the moment of inertia increases. Therefore, in the case of FIG. 6, the evaluation value for the island 71 included in the binarized image of the sample 3 is larger than the evaluation value of the island 70 included in the binarized image of the sample 1. Therefore, the shape of the island 71 included in the binary image of the sample 3 is extracted as the shape of the subject. As a result, it is possible to improve the accuracy with respect to estimation of the position of the subject and extraction of the shape of the subject.

  Information relating to the subject (subject information) extracted by the shape extraction unit 62 is output to the CPU 30. The CPU 30 sets an AF area based on the subject information and executes AF processing. That is, an AF process is performed in which the area of the frame 72 shown in FIG.

  In this way, the subject estimation unit 52 estimates the position of the subject using the Gabor map, and then the size of the white pixel clusters (islands) obtained from the image and the inertia of the island centered on the estimated position. The shape of the subject is evaluated by an evaluation value using the moment. By using such a method, for example, even when the position of the subject is unknown, the approximate position of the subject can be estimated, and as a result, the position and shape of the subject included in the image can be estimated. It becomes.

  In the present embodiment, the processing for estimating the subject is performed after the processing for estimating the attention point is performed. However, the present invention is not limited to this. For example, the processing for estimating the attention point and the shape of the subject are performed. It is also possible to perform the estimation process in parallel.

  In addition to this, it is also possible to perform the process of estimating the point of interest after performing the process of estimating the shape of the subject. In this case, the subject estimation unit 52 evaluates an island that is a subject candidate using, for example, the center of the image data or the center of the face area of the subject obtained by performing face detection processing. Using this evaluation and the coordinate data of the attention point obtained by the attention point estimation unit 51, the obtained evaluation value is re-evaluated, and the island having the highest re-evaluation result is identified as the subject.

  In the present embodiment, the details of the image size of the image data input to the subject specifying circuit 23 are not described, but by performing resolution conversion processing after image processing executed by the image processing unit 21. It is preferable that the image data has a reduced image size.

  In this embodiment, the timing for executing the process of specifying the subject in the subject specifying circuit 23 is not described in detail, but the above-described subject is applied to the through image data obtained immediately after shifting to the shooting standby state. For the through image data obtained by performing the specifying process, template matching using the position, shape, and size of the specified subject may be performed, or each of the generated through image data is processed. It is also possible to specify the subject by using. In addition to this, processing for specifying a subject may be performed on image data obtained by photographing. In this case, the result of specifying the subject can be used for processing such as image classification.

  In the present embodiment, an example of an imaging device is taken up. However, the present invention is not limited to this, and may be an image processing device that can execute the configuration of FIG. 2 or the flowchart of FIG. . In addition to this, a subject specifying program that can cause a computer to execute the function shown in FIG. 2 or the processing of the flowchart of FIG. 5 may be used. In this case, the subject specifying program is preferably stored in a storage medium that can be recognized by the computer, such as a memory card including a nonvolatile memory, an optical disk, a magnetic disk, and an HDD.

  DESCRIPTION OF SYMBOLS 10 ... Imaging device, 21 ... Image processing circuit, 23 ... Subject specific circuit, 51 ... Attention point estimation part, 52 ... Subject estimation part, 55 ... Filter processing part, 56, 61 ... Binarization processing part, 57 ... Position calculation 62, shape extraction unit

Claims (10)

A position estimation unit that estimates a position of a point of interest in the first image using a saliency map generated based on a feature amount of the input first image;
A binarization processing unit that performs binarization processing on a plurality of second images generated using either the luminance information or the color information of the first image, and generates a plurality of binarized images;
A subject estimation unit that estimates a subject included in the first image from the estimated position of the target point and the plurality of binarized images;
A subject identification device comprising: The subject specifying device according to claim 1,
The position estimation unit extracts a pixel whose feature amount in the saliency map is equal to or greater than a first threshold from among pixels included in the first image, and the extracted feature amount is equal to or greater than a first threshold. A subject identification device that estimates the position of the center of gravity as the position of the attention point. The subject specifying device according to claim 1 or 2,
A filter processing unit that generates a saliency map by applying a plurality of direction selectivity filters in different directions to the first image and adding the results obtained by applying the plurality of direction selectivity filters; A subject specifying device. The subject specifying device according to claim 3,
The image processing apparatus according to claim 1, wherein the direction selective filter is a Gabor filter. The subject specifying device according to claim 1,
The subject estimation unit has a ratio of an area of the white pixel region to a rectangular area including the white pixel region out of a white pixel region including white pixels included in the binarized image is equal to or less than a second threshold value. The white pixel region is excluded from white pixel regions that are candidates for estimating the subject. The subject specifying device according to claim 1,
The subject estimation unit includes a white pixel region including a white pixel region that includes white pixels included in the binarized image and includes a white pixel region that includes a rectangular aspect ratio that includes the white pixel region within a predetermined range. Is excluded from a white pixel region that is a candidate for estimating the subject. In the subject specifying device according to claim 5 or 6,
The subject estimation unit calculates an evaluation value using a white pixel region that is a candidate for estimating the subject and the position of the attention point, and a white pixel region having a large calculated evaluation value is used as the subject region. A subject specifying device characterized by estimating. An imaging optical system that captures subject light;
An image sensor that outputs an image signal based on the captured subject light;
The subject specifying device according to any one of claims 1 to 7, wherein an image based on an image signal output from the imaging device is input as a first image.
A control unit that performs focus adjustment control of the imaging optical system using a region including the subject estimated in the subject identifying device;
An imaging apparatus comprising: A position estimation procedure for estimating the position of the point of interest in the first image using a saliency map generated based on the input feature quantity of the first image;
A binarization procedure for generating a plurality of binarized images by binarizing a plurality of second images generated based on the luminance information and color information of the first image using the luminance information and the color information. When,
A subject estimation procedure for estimating a subject included in the first image from the estimated position of the attention point and the plurality of binarized images;
A method for identifying a subject characterized by comprising: A position estimation procedure for estimating the position of the point of interest in the first image using a saliency map generated based on the input feature quantity of the first image;
A binarization procedure for generating a plurality of binarized images by binarizing a plurality of second images generated based on the luminance information and color information of the first image using the luminance information and the color information. When,
A subject estimation procedure for estimating a subject included in the first image from the estimated position of the attention point and the plurality of binarized images;
Is a subject specifying program for causing a computer to execute.