JP5610245B2 - Image composition apparatus, image composition method, image composition program, and recording medium - Google Patents

Description

  The present invention relates to an image composition device, an image composition method, an image composition program, and a recording medium.

  2. Description of the Related Art Conventionally, an image synthesizing apparatus that performs high dynamic range synthesis (HDR (High Dynamic Range) synthesis) is known (see Patent Document 1). This apparatus apparently expands the dynamic range of the video signal by synthesizing a plurality of screens sequentially captured under different exposure conditions. This eliminates “overexposure” or “blackout” (part where the luminance level is extremely high or low) that has occurred during backlighting. Further, in this apparatus, HDR synthesis is performed after coordinate conversion of each of the plurality of screens is performed in response to a positional shift with time during imaging between the plurality of screens caused by camera shake. Specifically, HDR synthesis is performed using the common area portion of the two screens using the motion information of the image. This eliminates the positional deviation (screen blur) of the screen (image sensor) with respect to the subject.

Japanese Patent No. 3110797

  By the way, when the subject is a moving object, the subject positions are different in a plurality of images sequentially captured. For this reason, in the image composition device described in Patent Document 1, even when the subject is moving and the color is changed, the image is synthesized as if the color has changed due to different exposure. Therefore, there is a possibility that an appropriate composite image cannot be generated. In this technical field, an image composition device, an image composition method, an image composition program, and a recording medium storing the image composition program that can generate an appropriate composite image even when the subject moves are desired. Yes.

That is, an image composition device according to one aspect of the present invention is a device that generates a composite image using a first image and a second image having different exposure conditions. This apparatus includes an input unit, a likelihood calculation unit, an exposure estimation unit, and a synthesis unit. The input unit inputs the first image and the second image. The likelihood calculating unit calculates the moving subject likelihood in each pixel based on the difference between the first image and the second image before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image. To do. The exposure estimation unit estimates an exposure conversion function based on the moving subject likelihood. The combining unit combines the first image and the second image using the exposure conversion function. Further, the exposure estimation unit estimates the exposure conversion function by setting the weight of the pixel to be lower as the moving subject likelihood is higher.

  In this image construction device, the moving subject likelihood at each pixel is calculated based on the difference between the first image and the second image before matching the exposures of the first image and the second image. Based on the moving subject likelihood, an exposure conversion function that matches the exposure conditions of the first image and the second image is estimated. Thus, since the moving subject likelihood is taken into consideration when adjusting the exposure, for example, the exposure can be adjusted except for a region where the color may change due to the movement of the subject. Therefore, it is possible to generate an appropriate composite image.

In one embodiment, the likelihood calculating unit normalizes the neighboring pixels of the pixels of the first image and the corresponding neighboring pixels of the second image, and based on the difference between the normalized neighboring pixels, it may calculate the moving subject likelihood that put to each pixel. With this configuration, the moving subject likelihood at each pixel can be calculated appropriately.

  In one embodiment, the likelihood calculating unit changes the resolution of the plurality of first processed images and the second image obtained by changing the resolution of the first image in stages. The moving subject likelihood of each pixel may be calculated by calculating the difference of each pixel for each resolution using the plurality of second processed images obtained in step S1 and weighting the difference obtained for each resolution. With this configuration, the moving subject likelihood of each pixel can be calculated with high accuracy.

  In one embodiment, the likelihood calculation unit is obtained for each resolution based on the reliability of the difference between the first image and the second image and the image size or resolution of the first processed image or the second processed image. The differences may be weighted. With this configuration, the moving subject likelihood of each pixel can be calculated with higher accuracy.

  In one embodiment, the synthesizing unit calculates a moving subject likelihood in each pixel based on a difference between the first image and the second image, and uses the moving subject likelihood and the exposure conversion function to perform the first image and the second image. Two images may be combined. By configuring in this way, it is possible to synthesize in consideration of the movement of the subject, so that an appropriate synthesized image can be generated.

  In one embodiment, the synthesizing unit generates a luminance base mask indicating a synthesis ratio of the pixel values of the first image and the second image based on the original luminance value of the first image or the second image. May be. The synthesizing unit may generate a subject blur mask indicating a synthesis ratio of the pixel values of the first image and the second image based on the difference between the first image and the second image. Further, the combining unit may combine the luminance base mask and the subject blur mask to generate a combined mask that combines the pixel values of the first image and the second image.

  With this configuration, it is possible to generate a subject blur mask that is different from the luminance base mask that is synthesized based on the luminance value, based on the difference between the first image and the second image, with the exposure adjusted. . For this reason, it is possible to combine only areas where subject blurring occurs by different processes. Therefore, it is possible to generate a composite image in which subject blur is suppressed.

  In one embodiment, the synthesizing unit may calculate a moving subject likelihood in each pixel based on the difference between the first image and the second image, and generate a subject blur mask based on the moving subject likelihood. Good. With this configuration, a subject blur mask can be generated by specifying a region where subject blur occurs based on the moving subject likelihood.

  In one embodiment, the synthesizing unit obtains the plurality of first processed images obtained by changing the resolution of the first image in stages, and the resolution of the second image, respectively. The difference between each pixel is calculated for each resolution using a plurality of second processed images, the moving subject likelihood of each pixel is calculated by weighting the difference obtained for each resolution, and the moving subject likelihood A subject blur mask may be generated based on the above. With this configuration, the moving subject likelihood of each pixel can be calculated with high accuracy.

  In one embodiment, the synthesizing unit detects a region where pixels whose moving subject likelihood is equal to or less than a predetermined threshold is adjacent, assigns an identification label to each region, and generates a subject blur mask for each region. May be. With such a configuration, even when there is a moving body that moves differently in the image, it can be appropriately combined.

  In one embodiment, the synthesis unit forcibly selects a pixel value having a low luminance value from the first image and the second image as the subject blur mask, or the first image and the second image. A second mask for forcibly selecting a pixel value having a high luminance value may be generated. With this configuration, it is possible to forcibly select either the first image or the second image for an area where the subject may be moving. Therefore, it is possible to avoid the subject from being displayed in a double or triple shift in the combined image due to the movement of the subject.

  In one embodiment, the synthesis unit may generate a synthesis mask by multiplying the luminance base mask by a mask obtained by inverting the first mask or by adding a second mask. With this configuration, it is possible to generate a composite mask that appropriately corrects subject blurring.

  In one embodiment, the image processing apparatus may further include a motion information acquisition unit that acquires motion information of pixels between the first image and the second image. The likelihood calculating unit may correct the first image and the second image based on the motion information, and calculate the moving subject likelihood of each pixel using the corrected first image and second image. . With this configuration, even when the image sensor moves relative to the subject, the motion of the image sensor can be corrected and the moving subject likelihood of each pixel can be calculated.

  In one embodiment, the first image may be an image in which images having different exposure conditions are combined. With this configuration, a final composite image can be generated by sequentially combining a plurality of images having different exposure conditions.

An image processing method according to another aspect of the present invention is a method for generating a composite image using a first image and a second image having different exposure conditions. In this method, a first image and a second image are input. Then, before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image, the moving subject likelihood at each pixel is calculated based on the difference between the first image and the second image. Then, the pixel with the higher moving subject likelihood is set to a smaller weight for the pixel, and the exposure conversion function is estimated. Further, the first image and the second image are synthesized using the exposure conversion function.

An image composition program according to still another aspect of the present invention is a program that causes a computer to operate so as to generate a composite image using a first image and a second image having different exposure conditions. This program causes the computer to operate as an input unit, a likelihood calculation unit, an exposure estimation unit, and a synthesis unit. The input unit inputs the first image and the second image. The likelihood calculating unit calculates the moving subject likelihood in each pixel based on the difference between the first image and the second image before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image. To do. The exposure estimation unit estimates the exposure conversion function by setting the weight of the pixel to be smaller as the moving subject likelihood is higher . The combining unit combines the first image and the second image using the exposure conversion function.

A recording medium according to still another aspect of the present invention is a recording medium on which an image composition program for operating a computer to generate a composite image using a first image and a second image having different exposure conditions is recorded. . This program causes the computer to operate as an input unit, a likelihood calculation unit, an exposure estimation unit, and a synthesis unit. The input unit inputs the first image and the second image. The likelihood calculating unit calculates the moving subject likelihood in each pixel based on the difference between the first image and the second image before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image. To do. The exposure estimation unit estimates the exposure conversion function by setting the weight of the pixel to be smaller as the moving subject likelihood is higher . The combining unit combines the first image and the second image using the exposure conversion function.

  According to the image synthesizing method, the image synthesizing program, and the recording medium according to the other aspects of the present invention, the same effects as those of the above-described image synthesizing apparatus are obtained.

  According to various aspects and embodiments of the present invention, an image synthesizing apparatus, an image synthesizing method, an image synthesizing program, and an image synthesizing program that can generate an appropriate synthesized image even when the subject moves are stored. A recording medium is provided.

It is a functional block diagram of the portable terminal carrying the image composition apparatus concerning one embodiment. It is a hardware block diagram of the portable terminal in which the image composition apparatus of FIG. 1 is mounted. 3 is a flowchart showing a preprocessing operation of the image composition device shown in FIG. 1. It is a schematic diagram explaining motion detection. It is a schematic diagram explaining a difference image. It is a schematic diagram explaining the example which derives | leads-out a difference image using multiresolution. It is a graph which shows an example of an exposure conversion function. It is a schematic diagram explaining a luminance conversion function. 3 is a flowchart illustrating a composition operation of the image composition apparatus illustrated in FIG. 1. It is a schematic diagram explaining the flow of a synthetic | combination process. It is a schematic diagram explaining a synthetic mask. (A) is a graph which shows an example of an exposure conversion function. (B) is a graph which shows an example of the weight at the time of connecting exposure conversion functions. It is a schematic diagram explaining a brightness | luminance base mask. (A) is an example of an input image. (B) is an example of a luminance base mask. It is a schematic diagram explaining the subject blur region labeling of a difference image. (A) is an example of a difference image. (B) is an example of a labeled difference image. It is a schematic diagram explaining the flow of a subject blur mask generation process. It is a schematic diagram explaining the flow of a synthetic | combination mask production | generation process.

  Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In addition, in each figure, the same code | symbol is attached | subjected to the same or an equivalent part, and the overlapping description is abbreviate | omitted.

  The image composition device according to the present embodiment is a device that generates a single composite image by combining a plurality of images having different exposure conditions. This image synthesizing apparatus is employed, for example, when HDR synthesis is performed in which a plurality of images sequentially captured under different exposure conditions are synthesized and the dynamic range of the video signal is apparently enlarged. The image composition device according to the present embodiment is suitably mounted on a mobile terminal with limited resources such as a mobile phone, a digital camera, and a PDA (Personal Digital Assistant), but is not limited thereto. For example, it may be mounted on a normal computer system. In the following, considering the ease of understanding, an image composition device mounted on a portable terminal having a camera function will be described as an example of an image composition device according to the present invention.

  FIG. 1 is a functional block diagram of a mobile terminal 2 including an image composition device 1 according to this embodiment. A mobile terminal 2 shown in FIG. 1 is a mobile terminal carried by a user, for example, and has a hardware configuration shown in FIG. FIG. 2 is a hardware configuration diagram of the mobile terminal 2. As shown in FIG. 2, the portable terminal 2 physically includes a main storage device such as a CPU (Central Processing Unit) 100, a ROM (Read Only Memory) 101, and a RAM (Random Access Memory) 102, a camera, a keyboard, and the like. The input device 103, the output device 104 such as a display, the auxiliary storage device 105 such as a hard disk, and the like are configured as a normal computer system. Each function of the portable terminal 2 and the image composition device 1 described later is configured to load the input device 103 and the output device 104 under the control of the CPU 100 by reading predetermined computer software on hardware such as the CPU 100, the ROM 101, and the RAM 102. This is realized by operating and reading and writing data in the main storage device and the auxiliary storage device 105. Although the above description has been given as the hardware configuration of the mobile terminal 2, the image composition apparatus 1 normally includes a CPU 100, a main storage device such as the ROM 101 and RAM 102, an input device 103, an output device 104, an auxiliary storage device 105, and the like. It may be configured as a computer system. The mobile terminal 2 may include a communication module or the like.

  As shown in FIG. 1, the mobile terminal 2 includes a camera 20, an image composition device 1, and a display unit 21. The camera 20 has a function of capturing an image. For example, a CMOS pixel sensor or the like is used as the camera 20. The camera 20 has a continuous imaging function that repeatedly captures images at a predetermined interval from a timing specified by a user operation or the like, for example. That is, the camera 20 has a function of acquiring not only one still image but also a plurality of still images (continuous frame images). Further, the camera 20 has a function of changing the exposure condition of each successive frame image and taking an image. That is, each image continuously captured by the camera 20 is an image with different exposure conditions. For example, the camera 20 has a function of outputting a captured frame image to the image composition device 1 every time it is captured.

  The image composition device 1 includes an image input unit 10, a preprocessing unit 11, a motion correction unit 15, and a composition unit 16.

  The image input unit 10 has a function of inputting a frame image captured by the camera 20. The image input unit 10 has a function of inputting, for example, a frame image captured by the camera 20 every time it is captured. In addition, the image input unit 10 has a function of saving an input frame image in a storage device provided in the mobile terminal 2.

  The preprocessing unit 11 performs preprocessing before HDR synthesis. The preprocessing unit 11 includes a motion information acquisition unit 12, a likelihood calculation unit 13, and an exposure estimation unit 14.

  The motion information acquisition unit 12 has a function of acquiring pixel motion information between images. For example, when the input frame image is a first image and a second image, pixel motion information between the first image and the second image is acquired. For example, a motion vector is used as the motion information. Further, when three or more input images are input by the image input unit 10, the motion information acquisition unit 12 sorts the input images in the order of exposure, and acquires motion information between input images having similar exposure conditions. Good. By detecting motion by comparing images with similar exposure conditions, it is possible to avoid a decrease in motion detection accuracy due to a difference in exposure between images. Then, the motion information acquisition unit 12 may select a reference image that matches the motion information from among a plurality of input images. As the reference image, for example, an image having the most effective pixels among a plurality of input images is employed. Here, an effective pixel is a pixel that is not crushed or blown out. Blackout or whiteout is determined based on the luminance value. When the motion information acquisition unit 12 acquires motion information using two input images, the motion information acquisition unit 12 extracts a feature point from an input image with high exposure among the two input images, and sets a corresponding point for the feature point as an exposure point. You may obtain | require from a low image. By operating in this way, it is possible to avoid that the motion information cannot be acquired because the points extracted as feature points in the low-exposure image are whiteout in the high-exposure image. Note that movement information may be acquired from a gyro sensor or the like. The motion information acquisition unit 12 has a function of outputting motion information to the likelihood calculation unit 13.

  The likelihood calculating unit 13 has a function of calculating the likelihood of the movement of the subject in each pixel (moving subject likelihood). The larger the moving subject likelihood, the higher the possibility that the subject is moving and the higher the possibility that the composite image will be blurred. The likelihood calculating unit 13 corrects the screen motion between the input images using the motion information. Thereafter, the likelihood calculating unit 13 normalizes the pixel values of the corresponding pixels in the two input images. For example, the likelihood calculating unit 13 obtains Local Territory Patterns (LTP) based on pixel values of neighboring pixels. As the pixel value, RGB three colors are used, and as the neighboring pixels, 24 neighborhoods are used. Then, the likelihood calculating unit 13 calculates the moving subject likelihood by using the difference between the normalized images. For example, the normalized pixel value difference, that is, the sign mismatch rate in the LTP of the target pixel is calculated as the moving subject likelihood of the target pixel.

  In addition, the likelihood calculating unit 13 may calculate the moving subject likelihood by converting two input images into multiple resolutions. For example, the likelihood calculating unit 13 changes the resolution of each input image (the first image and the second image) in stages, thereby a plurality of images having different resolutions (the first processed image and the second processed image). Create Then, the likelihood calculating unit 13 creates a difference image between the first processed image and the second processed image at the same resolution. This difference image is a difference between the first processed image and the second processed image, and specifically, is a difference between pixel values. And the likelihood calculation part 13 calculates the moving subject likelihood of each pixel by weighting the difference image obtained for every resolution. As the weight (reliability), a code mismatch rate in the LTP of each pixel is used. For example, in LTP, the number of pairs that are significantly different is used. The weight may be further weighted according to the image size or resolution of the first processed image or the second processed image. That is, the weight may be increased as the image size is larger or the resolution is larger. The likelihood calculating unit 13 has a function of outputting the moving subject likelihood of each pixel to the exposure estimating unit 14.

  The exposure estimation unit 14 has a function of estimating an exposure conversion function that matches exposure conditions between input images. The exposure conversion function is a function for converting the exposure of each input image to an exposure equivalent to the reference image. When three or more input images are input, the exposure estimation unit 14 may match the exposure conditions between input images having similar exposure conditions. By matching the exposures by comparing the images having similar exposure conditions, it is possible to avoid the estimation accuracy from being lowered due to the difference in exposure between the images.

For example, the exposure estimation unit 14 corrects the motion between the input screens using the motion information. Then, in the two input images after motion correction, the luminance values are sampled from the same location as a set, and the relationship is plotted. For example, a Halton number sequence is used as the coordinates of the input image. Note that the exposure estimation unit 14 does not have to adopt a luminance value greater than or equal to a predetermined value or a luminance value less than or equal to a predetermined value as a sampling point. For example, the luminance value included in the range of 10 to 245 is employed as the sampling point. The exposure estimation unit 14 estimates the exposure conversion function by fitting the plot result, for example. For example, when the original luminance value at the sampling point i of the first image is K _i , the exposure conversion function is f (K _i ), and the original luminance value at the sampling point i of the second image is U _i , the following error The function e may be used for fitting by the Gauss-Newton method.

The exposure estimation unit 14 performs sampling for deriving an exposure conversion function based on the moving subject likelihood of each pixel. For example, the exposure estimation unit 14 selects a sampling point based on the moving subject likelihood of each pixel. For example, the exposure estimation unit 14 samples the luminance value from pixels with a small moving subject likelihood by providing threshold values in stages. Further, the exposure estimation unit 14 may weight the sampling points based on the moving subject likelihood. For example, the following error function e may be minimized for fitting.

In Equation 2, w _i is a weight. Here, the weight w _i is set to be smaller as the moving object likelihood becomes higher. As described above, the exposure estimation unit 14 calculates the exposure conversion function based on the moving subject likelihood of each pixel, so that the data of the sampling point with lower reliability does not affect the derivation of the exposure conversion function. be able to. Note that the exposure conversion function may be changed so that the input image after conversion falls within a representable range.

  The motion correction unit 15 has a function of correcting motion between input screens using motion information. The synthesizing unit 16 synthesizes the input images or the already synthesized image and the input image using the synthesis mask. The combination mask is an image of the combination ratio (weight) when combining images (α blend). When there are three or more input images, the synthesizing unit 16 first synthesizes the two input images according to the synthesis mask, generates a synthesis mask of the synthesized image and the remaining input images, and performs synthesis. The combining unit 16 combines the luminance base mask and the subject blur mask to generate a combined mask. The luminance base mask is a mask for avoiding the use of an overexposure or underexposure region for synthesis by determining a weight for combining images based on a luminance value. The subject blur mask is a mask for avoiding a phenomenon (ghost phenomenon) in which a subject is displayed in a double or triple manner when an image in which the subject moves is combined.

The synthesizer 16 calculates a weight based on the original luminance value of the input image and generates a luminance base mask. The weight is obtained by the following calculation formula, for example.

According to the above calculation formula, the weight is appropriately determined and the luminance discontinuity is reduced. In order to reduce spatial discontinuity, the synthesis mask may be subjected to blurring processing.

  The synthesizer 16 calculates a weight based on the difference between the input images and generates a subject blur mask. The synthesizer 16 calculates the moving subject likelihood from the difference in pixel values between the input images. The pixel value difference between the input images and the moving subject likelihood can be obtained by operating in the same manner as the likelihood calculating unit 13 described above. Then, the likelihood calculating unit 13 detects a subject blur region adjacent to a pixel whose moving subject likelihood is equal to or less than a predetermined threshold, assigns an identification label to each subject blur region, and subjects each subject blur region. Generate a blur mask. The predetermined threshold value can be changed as appropriate according to the required specifications. If the threshold value is set large, it is possible to easily extract the continuous area. By generating a mask for each subject blur area, pixels can be selected from an image with a large amount of information so as to avoid a whiteout area or a blackout area for each subject blur area. That is, as this subject blur mask, lo_mask (first mask) for forcibly selecting a pixel value having a low luminance value among the images to be combined, or a pixel forcibly having a high luminance value among the images to be combined. There is a hi_mask (second mask) that selects a value. The synthesizing unit 16 basically generates a second mask that selects pixel values from a high-exposure image with a large amount of information. However, the composition unit 16 generates the first mask when the subject blur area is affected by the whiteout area in the high-exposure image. Specifically, the first mask is generated when any of the following conditions is satisfied. The first condition is a case where, among the two images to be combined, the overexposure area of the high exposure image is larger than the area of the blackout area of the low exposure image. The second condition is a case where, in a high-exposure image of the two images to be combined, the area of the whiteout region in the subject blur region is 10% or more. It should be noted that, in the high-exposure image of the two images to be combined, a condition may be set where the region adjacent to the subject blur region is a whiteout region.

  The combining unit 16 combines the luminance base mask and the subject blur mask to generate a combined mask. For example, the synthesis unit 16 multiplies the luminance base mask by a mask obtained by inverting the first mask. The synthesizing unit 16 adds the second mask to the luminance base mask. The synthesizer 16 synthesizes all the input images and outputs the final synthesized image to the display unit 21. The display unit 21 displays a composite image. For example, a display device is used as the display unit 21.

  Next, the operation of the image composition device 1 will be described. FIG. 3 is a flowchart for explaining preprocessing for HDR synthesis. The control process illustrated in FIG. 3 starts when the HDR synthesis mode is selected by the user and the camera 20 continuously captures a plurality of images, for example.

First, the image input unit 10 inputs an image frame (S10). In the following description, it is assumed that five input images I _{0 to} I ₄ are input in consideration of easy understanding. When the process of S10 ends, the process proceeds to the exposure order sort process (S12).

In the processing in S12, the motion information obtaining section 12 sorts the input image _I 0 ~I ₄ in display order. The motion information acquisition unit 12 sorts using, for example, the average value of the luminance values. Here, it is assumed that the luminance value is smaller as the numbers of the input images I _{0 to} I ₄ are smaller. In this case, the input images I _{0 to} I ₄ are sorted in numerical order. When the process of S12 ends, the process proceeds to a motion information acquisition process (S14).

In the processing in S14, the motion information obtaining section 12 acquires the motion information between each image of the input image _I 0 ~I _4. FIG. 4 is a schematic diagram illustrating motion information acquisition processing. As illustrated in FIG. 4, it is assumed that the input images I _{0 to} I ₄ are arranged so that the average luminance value increases in order from the left to the right. First, the motion information acquisition unit 12 sets a reference image from among the input images I _{0 to} I ₄ . Here, the reference image input image I _2. Next, motion information between input images having similar exposure conditions is acquired (for example, input image I ₀ and input image I ₁ , input image I ₁ and input image I _2, etc.). The motion information acquisition unit 12 extracts a feature point from the input image with high exposure out of the two input images, and extracts a corresponding point for the extracted feature point from the input image with low exposure. Based on this motion information, it is possible to obtain a conversion matrix for converting input images having similar exposure conditions into coordinates of the same dimension. FIG. 4 shows conversion matrices m10, m21, m32, and m43 for matching a low-exposure image to a high-exposure image among input images with similar exposure conditions. Next, using the transformation matrix m10, m21, m32, m43, deforming the reference image _{I 2} other input image _I 0 other _than, I _1, I 3, the coordinates of the _{I 4} to the reference image _{I 2} corresponding coordinates A transformation matrix is calculated. As shown in FIG. 4, the conversion matrix for converting the input image I ₀ into the reference image I ₂ is m10 * m21. Transformation matrix to convert the input image _{I 1} to the reference image _{I 2} is m10. A conversion matrix for converting the input image I ₃ to the reference image I ₂ is (m32) ⁻¹ . A conversion matrix for converting the input image I ₄ to the reference image I ₂ is (m32 * m43) ⁻¹ . Hereinafter, the converted input image will be described as I _{0 ′ to} I _{4 ′} . When the process of S14 ends, the process proceeds to a moving subject likelihood calculation process (S16).

In the process of S16, the likelihood calculating unit 13 calculates the moving subject likelihood between the respective images of the input images I0 _'to I4 _' . FIG. 5 is an example of calculating the moving subject likelihood between the images of the input image I _{0 ′} and the input image I _{1 ′} . FIG. 5 shows a case where the R value is used as the pixel value. As shown in FIG. 5, the likelihood calculating unit 13 obtains pixel values (R values) near 8 of the target pixel (R value = 42) of the input image I _{0 ′} . Then, normalization is performed using the pixel value of the target pixel and the pixel values in the vicinity of 8. For example, LTP is used. If the difference between the pixel value of the pixel of interest and the pixel values in the vicinity of 8 is within a range of ± 5, 0 is set when it is larger than +5, and -1 when it is smaller than −5. The likelihood calculating unit 13 also normalizes the input image I _{1 ′} in the same manner. In the figure, the pixels of the input image I _{1 ′} corresponding to the target pixel of the input image I _{0 ′} are normalized. Next, when the pixel values of the normalized pixels are compared, it can be seen that a difference has occurred. The difference image X is represented as an image in which the color of the pixel is changed from black to white in accordance with the magnitude of the difference (the degree of mismatch of the signs). This difference image is an image of the moving subject likelihood of each pixel. In addition, it is not limited to 8 vicinity, 24 vicinity may be sufficient. Further, not only the R value but also the G value and the B value may be processed similarly.

In order to improve the moving subject likelihood of accuracy of the smoothing region shown in region C ₁ of the difference image X, the likelihood calculating unit 13 may obtain the moving subject likelihood using multiresolution. FIG. 6 is an example of obtaining the moving subject likelihood using multi-resolution. First, the likelihood calculating unit 13 generates a plurality of images in which the resolutions of the input image I _{0 ′} and the input image I _{1 ′} are changed stepwise. Then, difference images are generated with the same resolution. This difference image is obtained by simply subtracting pixel values. FIG. 6 shows a case where the input image I _{0 ′} and the input image I _{1 ′} are multiplexed in six stages. The difference images are X _{1 to} X ₆ , and the difference image has a lower resolution as the number increases. Also, the lower the resolution, the smaller the image size. The difference image is weighted with reliability to calculate a final difference image. As the reliability, for example, a value obtained by multiplying the number of pairs having a significant difference in the above-described LTP difference by the image size (or resolution) is used. For example, in the case of LTP shown in FIG. 5, the number of pairs having a significant difference is 1. In this way, the weight image (the image of the weights) corresponding to the difference images X _{1 to} X ₆ is calculated by multiplying the number of pairs by the image size for each pixel. Then, a final difference image is calculated using the difference images X _{1 to} X ₆ and the weight image. The likelihood calculating unit 13 calculates a difference image from the input images I _{1 ′ to} I _{4 ′} by a method similar to the method described above. When the process of S16 is completed, the process proceeds to an exposure conversion function estimation process (S18).

In the process of S18, the exposure estimation unit 14 estimates an exposure conversion function. The exposure estimation unit 14 can express the exposure conversion function by the following formula, where x is the luminance value before conversion and y is the luminance value after conversion.

Here, (a, b) are exposure conversion parameters. An exposure conversion function can be obtained by deriving the exposure conversion parameters (a, b). Hereinafter, a case where an exposure conversion function between the input image I _{0 ′} and the input image I _{1 ′} after motion correction is obtained will be described. The exposure estimation unit 14 samples several sets of luminance values of the low-exposure input image I _{0 ′} and the low-exposure input image I _{1 ′} at the point (x, y) of the input image. Plot the relationship. Here, a point to be sampled is selected based on the difference image acquired in the process of S16. For example, the setting is made so that sampling is not performed from an area where the moving subject likelihood is high. In other words, the sampling is set so as to sample from a moving object with a low likelihood. Then, for example, a lower weight is assigned as the moving subject likelihood is higher, and the exposure conversion function is estimated using Expression 2. Thereby, for example, fitting as shown in FIG. 7 is performed. The likelihood calculating unit 13 estimates an exposure conversion function between the input images I _{1 ′ to} I _{4 ′} by a method similar to the method described above. Note that data with luminance values close to 0 or data close to 255 may be excluded.

FIG. 8 is a schematic diagram illustrating the exposure conversion function estimation process. In FIG. 8, exposure conversion parameters (a10, b10), (a21, b21), (a32, b32) for matching a low-exposure image to a high-exposure image among input images with similar exposure conditions. (A43, b43) are shown. The conversion result is obtained by setting A ₀ of the exposure conversion parameter (A ₀ , B ₀ ) of the input image I _{0 ′ having} the lowest exposure to 1.0 so that the final composite image can be expressed. May not exceed 1.0. Here, _'the image after exposure conversion of the input image I _0' input image I ₀ is displayed as a _'. If the exposure conversion parameter of the reference image I _{2 ′ with} respect to the input image I _{0 ′ having} the lowest exposure is (A ₂ , B ₂ ), A ₀ is set to 1.0 and B ₂ is set to 1.0. Thus, the color may be set to be equal to that of the input image when the gain is 1 / A ₂ . The likelihood calculating unit 13 performs the above-described processing separately for each RGB channel. When the process of S18 ends, the pre-process shown in FIG. 3 ends.

  Thus, the control process shown in FIG. 3 is finished. By performing the control process shown in FIG. 3, by detecting the subject blur before estimating the exposure conversion function, it is possible to avoid sampling from the subject blur region, or to change the data sampled from the subject blur region. The influence can be reduced by weighting. For this reason, the exposure conversion function can be estimated with high accuracy. Also, with the conventional HDR technology, correction of subject blur cannot be performed accurately unless exposure adjustment is performed, and conversely, exposure correction is accurate unless subject blur correction is performed. Can not. However, the deadlock relationship can be resolved by simply detecting subject shake (subject movement) before estimating the exposure conversion function.

  Next, the composition operation of the image composition device 1 will be described. FIG. 9 is a flowchart for explaining HDR synthesis. The control process shown in FIG. 9 starts when the control process shown in FIG.

As shown in FIG. 9, the motion correction unit 15 actually corrects the motion (S20). In this process, similarly to the process of S14 of FIG. 3, the motion correction unit 15 corrects the motion of the input images I _{0 ″ to} I _{4 ″} after exposure conversion using the conversion matrix. Note that a subpixel interpolation algorithm or the like may be used according to the required accuracy. When the process of S20 ends, the process proceeds to a luminance base mask generation process and a subject blur area extraction process (S22 and S24).

In the process of S22, the synthesis unit 16 generates a luminance base mask. FIG. 10 is a schematic diagram illustrating the flow of the synthesis process. As shown in FIG. 10, the images are synthesized by replacing the input images I _{1 ″ to} I _{4 ″ in} order from the input image I _{0 ″ having} the lowest exposure. That is, first, a luminance base mask that determines how much the input image I _{1 ″} is to be combined with the input image I _{0 ″} is generated. This luminance base mask calculates the weight from the original luminance value of the input image I _{1 ″} . For example, the weight near the whiteout area is set to zero. By setting the weights in this way and combining the images with high exposure on the images with low exposure, it is possible to always select an input image with a large amount of information for the target pixel. FIG. 11A is a graph showing the relationship of the pixel value with respect to the input luminance. As shown in FIG. 11A, the functions f _{0 to} f ₃ are graphs showing which image pixel value is adopted based on the luminance value. The functions f _{0 to} f ₃ are applied to an image having a larger exposure as the number increases. For example, when the input image I _{0 ″ having} the lowest exposure is input, the function f ₀ is applied and all pixel values are adopted. Next, when the input image I _{1 ″} is input, the function f ₀ and the function f ₁ are applied. For this reason, the input image I _{1 ″} is adopted in the range of the luminance values of S0 to S5, and the input image I _{0 ″} is adopted in the range of the luminance value of S6 or more. The range of the luminance values of S5 to S6 is adopted as a composite value blended with the weight shown in (B). For convenience, γ correction is omitted. Next, when the input image I _{2 ″} is input, the functions f _{0 to} f ₂ are applied. For this reason, the input image I _{2 ″} is adopted in the range of luminance values of S0 to S3, the input image I _{1 ″} is adopted in the range of luminance values of S4 to S5, and the range of luminance values of S6 or more. Then, the input image I _{0 ″} is adopted. The range of the luminance values of S3 to S4 and S5 to S6 is adopted as a composite value blended with the weight shown in (B). Next, when the input image I _{3 ″} is input, the functions f _{0 to} f ₃ are applied. For this reason, the input image I _{3 ″} is adopted in the range of luminance values of S0 to S1, the input image I _{2 ″} is adopted in the range of luminance values of S2 to S3, and the luminance values of S4 to S5. In the range, the input image I _{1 ″} is adopted, and in the range of the luminance value of S6 or more, the input image I _{0 ″} is adopted. The range of the luminance values of S1 to S2, S3 to S4, and S5 to S6 is adopted as a composite value blended with the weight shown in (B). In this way, images with high exposure are preferentially adopted. In addition, an image with low exposure is adopted for the overexposed region portion, and the boundary portion is blended smoothly. FIG. 12 shows an example of a luminance base mask obtained by imaging the graph shown in FIG. 12A shows an input image, and FIG. 12B shows a luminance base mask of the input image. In FIG. 12B, white is used when the pixel value of the 100% input image is used, and black is displayed when the pixel value of the 100% input image is not used. When the process of S22 ends, the process proceeds to a synthesis mask generation process (S32).

  On the other hand, in the process of S24, the synthesizer 16 extracts a subject blur area. For example, the synthesizing unit 16 calculates a difference image in the same manner as the process of S16 in FIG. 3, and extracts a region where the moving subject likelihood is a predetermined value or more as a subject blur region. FIG. 13A is an example of a difference image including a subject blur area. When the process of S24 is completed, the process proceeds to a labeling process (S26).

In the process of S26, the synthesizer 16 labels the subject blur area. The combining unit 16 sets one of the label R _n with respect to subject shake a continuous area. FIG. 13B shows an example in which continuous regions are labeled. When the process of S26 is completed, the process proceeds to a reference image selection process for each region (S28).

In the process of S28, the combining unit 16 sets a reference image for each subject blur area. The synthesizing unit 16 gives priority to an image with high exposure basically as a reference image. For example, when the input image I _{0 ″} and the input image I _{1 ″} are combined, the input image I _{1 ″} is selected as the reference image. However, when the subject blur area is affected by the whiteout area in the input image I _{1 ″} , the input image I _{0 ″} is selected as the reference image. When the processing of S28 is completed, the process proceeds to subject blur mask generation processing (S30).

In the process of S30, the synthesis unit 16 generates a subject blur mask for each subject blur region. The synthesizing unit 16 generates the second mask when giving priority to the high exposure image as the reference image. On the other hand, the first mask is generated when priority is given to the low-exposure image as the reference image. FIG. 14 is a schematic diagram illustrating a series of processes from S24 to S30. As shown in FIG. 14, when the input image I _{0 ″} and the input image I _{1 ″} are combined, a difference image X is obtained, and a first mask (lo_mask) or a second mask is obtained for each area of the difference image. (Hi_mask) is generated. That is, for the region where the subject moves, the ghost phenomenon described above can be avoided by inputting the pixel value from only one image by using the subject blur mask. When the process of S30 ends, the process proceeds to a synthesis mask generation process (S32).

  In the process of S32, the synthesis unit 16 generates a synthesis mask based on the luminance base mask and the subject blur mask. FIG. 15 is a schematic diagram for explaining the synthesis mask generation process. As shown in FIG. 15, the luminance base mask is multiplied by an image obtained by inverting lo_mask. Also, hi_mask is added to the luminance base mask. By combining in this way, a composite mask is generated. When the process of S32 is completed, the process proceeds to the composition process (S34).

In the process of S34, the synthesis unit 16 performs a synthesis process according to the synthesis mask created in the process of S32. When the luminance value P ₀ of the combined image and the luminance value P ₁ of the input image to which the exposure conversion function is applied are combined with the weight a, the combined luminance value P ₂ can be obtained by the following formula.

At this time, the entire region is synthesized as it is for the image with the lowest exposure. When the process of S34 is completed, the process proceeds to an input image confirmation process (S36).

In the process of S36, it is determined whether or not the combining unit 16 combined all input images. If all input images have not been combined, the process proceeds to S22 and S24. Then, for example, as shown in FIG. 10, a synthesis process of a composite image O ₀ of the input image I _{0 ″} and the input image I _{1 ″} and a new input image I _{0 ″} is performed. On the other hand, when all the input images have been combined, the control process shown in FIG. 9 ends.

  By executing the control processing shown in FIG. 9, an HDR composite image in which subject blur is corrected is generated.

  Next, an image composition program for causing the portable terminal (computer) 2 to function as the image composition apparatus 1 will be described.

  The image composition program includes a main module, an input module, and an arithmetic processing module. The main module is a part that comprehensively controls image processing. The input module operates the mobile terminal 2 so as to acquire an input image. The arithmetic processing module includes a motion information acquisition module, a likelihood calculation module, an exposure estimation module, a motion correction module, and a synthesis module. The functions realized by executing the main module, the input module, and the arithmetic processing module are the image input unit 10, the motion information acquisition unit 12, the likelihood calculation unit 13, the exposure estimation unit 14, the motion of the image synthesis apparatus 1 described above. The functions of the correction unit 15 and the synthesis unit 16 are the same.

  The image composition program is provided by a recording medium such as a ROM or a semiconductor memory, for example. The image composition program may be provided as a data signal via a network.

  As described above, according to the image synthesizing apparatus 1, the image synthesizing method, and the image synthesizing program according to the present embodiment, before matching the exposure of the first image and the second image, based on the difference between the first image and the second image. Thus, the likelihood of movement of the subject in each pixel is calculated. Then, an exposure conversion function that matches the exposure conditions of the first image and the second image is estimated based on the likelihood of the movement of the subject. Thus, since the likelihood of the movement of the subject is taken into account when adjusting the exposure, for example, the exposure can be adjusted except for a region where the color may change due to the movement of the subject. Therefore, it is possible to generate an appropriate composite image. Furthermore, the use of the subject blur mask prevents the occurrence of subject blur (ghost-like display), and a clear image can be obtained.

  The embodiment described above shows an example of an image composition apparatus according to the present invention. The image synthesizing apparatus according to the present invention is not limited to the image synthesizing apparatus 1 according to the embodiment, and the image synthesizing apparatus according to the embodiment may be modified or otherwise changed without changing the gist described in each claim. It may be applied to the above.

  For example, in each of the above-described embodiments, an example in which the camera 20 acquires a frame image has been described. However, an image transmitted from another device via a network may be used. Further, when only recording without displaying the composite image, the display unit 21 may not be provided.

  In addition, the image composition device 1 according to each embodiment described above may be operated together with the camera shake correction device.

  DESCRIPTION OF SYMBOLS 1 ... Image composition apparatus, 10 ... Image input part (input part), 12 ... Motion information acquisition part, 13 ... Likelihood calculation part, 14 ... Exposure estimation part, 15 ... Motion correction part, 16 ... Composition part.

Claims (16)

An image composition device that generates a composite image using a first image and a second image having different exposure conditions,
An input unit for inputting the first image and the second image;
The likelihood of calculating the moving subject likelihood in each pixel based on the difference between the first image and the second image before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image. A degree calculator,
An exposure estimation unit that estimates the exposure conversion function based on the moving subject likelihood;
A combining unit that combines the first image and the second image using the exposure conversion function;
Equipped with a,
The exposure estimation unit estimates the exposure conversion function by setting the weight of the pixel to be smaller as the moving subject likelihood is higher.
Image composition device. The likelihood calculation unit normalizes the neighboring pixels of the pixels of the first image and the corresponding neighboring pixels of the second image, and based on the difference between the normalized neighboring pixels, The image composition device according to claim 1, wherein the moving subject likelihood is calculated. The likelihood calculating unit
A plurality of first processed images obtained by changing the resolution of the first image in stages, and a plurality of second processed images obtained by changing the resolution of the second image in stages. The image synthesizing apparatus according to claim 1 or 2, wherein a difference between the pixels is calculated for each resolution, and the moving subject likelihood of each pixel is calculated by weighting the difference obtained for each resolution.   The likelihood calculation unit is obtained for each resolution based on the reliability of the difference between the first image and the second image and the image size or resolution of the first processed image or the second processed image. The image composition device according to claim 3, wherein the difference is weighted. The synthesis unit is
A moving subject likelihood at each pixel is calculated based on the difference between the first image and the second image, and the first image and the second image are synthesized using the moving subject likelihood and the exposure conversion function. The image composition device according to any one of claims 1 to 4. The synthesis unit is
Generating a luminance base mask indicating a synthesis ratio of pixel values of the first image and the second image based on the magnitude of the original luminance value of the first image or the second image;
Based on the difference between the first image and the second image, generate a subject blur mask indicating a synthesis ratio of pixel values of the first image and the second image,
The image composition device according to any one of claims 1 to 5, wherein the luminance base mask and the subject blur mask are combined to generate a composite mask that combines pixel values of the first image and the second image. . The synthesis unit is
Based on the difference between the first image and the second image, the likelihood of a moving subject at each pixel is calculated,
The image composition device according to claim 6, wherein the subject blur mask is generated based on the moving subject likelihood. The synthesis unit is
A plurality of first processed images obtained by changing the resolution of the first image in stages, and a plurality of second processed images obtained by changing the resolution of the second image in stages. And calculating a moving subject likelihood of each pixel by calculating a difference of each pixel for each resolution, weighting the difference obtained for each resolution, and applying the subject blur mask based on the moving subject likelihood. The image composition device according to claim 7 which generates. The synthesis unit is
The image composition according to claim 8, wherein an area adjacent to a pixel whose moving subject likelihood is equal to or less than a predetermined threshold is detected, an identification label is assigned to each area, and the subject blur mask is generated for each area. apparatus. The synthesis unit is
As the subject blur mask, a first mask that forcibly selects a pixel value having a low luminance value from the first image and the second image, or a luminance that is forcibly selected from the first image and the second image. The image composition device according to any one of claims 6 to 9, wherein a second mask for selecting a pixel value having a high value is generated. The synthesis unit is
The image composition device according to claim 10, wherein the composite mask is generated by multiplying the luminance base mask by a mask obtained by inverting the first mask, or adding the second mask. A movement information acquisition unit that acquires movement information of pixels between the first image and the second image;
The likelihood calculating unit corrects the first image and the second image based on the motion information, and calculates the moving subject likelihood of each pixel using the corrected first image and second image. The image composition device according to any one of claims 1 to 11 to calculate.   The image synthesizing apparatus according to claim 1, wherein the first image is an image obtained by synthesizing images having different exposure conditions. An image composition method for generating a composite image using a first image and a second image having different exposure conditions,
Input the first image and the second image;
Before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image, a moving subject likelihood in each pixel is calculated based on the difference between the first image and the second image,
Estimate the exposure conversion function by setting the weight of the pixel to be lower as the moving subject likelihood is higher,
Combining the first image and the second image using the exposure conversion function;
Image composition method. An image composition program for operating a computer to generate a composite image using a first image and a second image having different exposure conditions,
An input unit for inputting the first image and the second image;
The likelihood of calculating the moving subject likelihood in each pixel based on the difference between the first image and the second image before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image. Degree calculator,
The higher the moving subject likelihood is, the smaller the weight of the pixel is set , and the exposure estimation unit that estimates the exposure conversion function, and the first image and the second image are synthesized using the exposure conversion function. An image composition program for operating the computer as a composition unit. A recording medium recording an image composition program for operating a computer to generate a composite image using a first image and a second image having different exposure conditions,
An input unit for inputting the first image and the second image;
The likelihood of calculating the moving subject likelihood in each pixel based on the difference between the first image and the second image before estimating an exposure conversion function that matches the exposure conditions of the first image and the second image. Degree calculator,
The higher the moving subject likelihood is, the smaller the weight of the pixel is set , and the exposure estimation unit that estimates the exposure conversion function, and the first image and the second image are synthesized using the exposure conversion function. A recording medium on which an image composition program for operating the computer as a composition unit is recorded.