JP5028138B2 - Image processing apparatus and image processing program - Google Patents

Description

  The present invention relates to an image processing apparatus and an image processing program for detecting a specific area from an observation image, and more particularly to an image processing apparatus and image processing for detecting a specific area from a series of observation images captured at different imaging distances. It is about the program.

  As a system for obtaining a series of observation images in which at least one observation object is individually imaged among a plurality of types of observation objects, for example, a capsule endoscope that observes the inside of a subject has been developed. Capsule endoscopes have an imaging function and a wireless communication function. After being swallowed by a subject and before being spontaneously discharged, the inside of the esophagus, stomach, small intestine, and large intestine is subjected to peristalsis. The image is captured while moving sequentially, and image data generated by the imaging is sequentially transmitted to an external receiver by wireless communication. A doctor, nurse, or the like can display the image data acquired by the external receiver as an observation image and observe the inside of the subject based on the observation image.

  Usually, the number of series of observation images acquired by a capsule endoscope is enormous, and doctors, nurses, etc. need a lot of time and labor to perform observations based on the series of observation images. Is done. On the other hand, an abnormality detection system has been developed that can efficiently observe a lesioned part by detecting and displaying the abnormal part in the observation target (for example, see Patent Document 1). In this abnormality detection system, an observed image is divided into a plurality of unit blocks, and the abnormal information is detected by comparing the color information of each unit block with the color reference of the abnormal part and the color reference of the normal part prepared in advance. I am doing so.

JP-T-2004-521893 JP 2005-192880 A

  By the way, in a series of observation images acquired by the capsule endoscope, the capsule endoscope as an imaging device performs imaging while moving in the organ, so the same abnormal part in the organ is detected for each observation image. Images are taken at different imaging distances. For this reason, even if it is an image area | region corresponding to the same abnormal part, the area | region size (size) and color will show the characteristic which changes according to imaging distance for every observation image.

  However, in the above-described abnormality detection system, there is a problem in that it is not possible to reliably detect an abnormal part that shows different characteristics depending on the imaging distance for each observation image.

  The present invention has been made in view of the above, and is an image processing apparatus capable of improving detection accuracy when reliably detecting an image region having different characteristics depending on an imaging distance from a series of observation images. An object of the present invention is to provide an image processing program.

  In order to achieve the above object, an image processing apparatus according to the present invention is the image processing apparatus for detecting a characteristic image area that is a unique area from a series of observation images captured at different imaging distances. The feature image region is extracted from the processing target image using distance estimation means for estimating an imaging distance when the processing target image of the observed images is captured, and a processing parameter for detecting the feature image region. An area detection means for detecting, a parameter value corresponding to an estimation result of the distance estimation means is set as the processing parameter, and a setting control means for causing the area detection means to detect the feature image area using the processing parameter; It is provided with.

  In addition, the image processing program according to the present invention provides an image processing apparatus for detecting a characteristic image region, which is a unique region, from a series of observation images captured at different imaging distances. A distance estimation procedure for estimating an imaging distance when a target image is captured; a region detection procedure for detecting the feature image region from within the processing target image using a processing parameter for detecting the feature image region; Setting a parameter value corresponding to the estimation result of the distance estimation unit as the processing parameter, and causing the region detection unit to detect the feature image region using the processing parameter; To do.

  According to the image processing apparatus and the image processing program according to the present invention, it is possible to improve the detection accuracy when reliably detecting an image region having different characteristics depending on the imaging distance from a series of observation images.

  Exemplary embodiments of an image processing apparatus and an image processing program according to the present invention are explained in detail below with reference to the accompanying drawings. In this embodiment, the image processing apparatus according to the present invention provides a series of observation images obtained by individually capturing at least one observation object among a plurality of types of observation objects, for example, an esophagus by a capsule endoscope, In the following description, it is assumed that a series of observation images in which at least one of the stomach, the small intestine, and the large intestine is sequentially captured is processed. However, observation images that can be processed by the image processing apparatus according to the present invention are not limited to observation images obtained by imaging such organs (digestive organs) as observation targets, and the present invention is limited by this embodiment. It is not something. In the description of the drawings, the same portions are denoted by the same reference numerals.

(Embodiment)
First, an image processing apparatus according to an embodiment of the present invention will be described. FIG. 1 is a block diagram showing a main configuration of an image processing apparatus 1 according to the present embodiment. As shown in FIG. 1, the image processing apparatus 1 processes an image stored in the storage unit 3 and an input unit 2, a storage unit 3 and an output unit 5 that respectively input, store, and output various types of information including images. And an image processing unit 4 that is electrically connected to each of these units, and a control unit 6 that controls processing and operations of the connected units.

  The input unit 2 is configured using a data communication interface, and inputs image data of a series of observation images as a processing target from the data communication interface to the control unit 6. The input unit 2 includes various input devices, and inputs various information such as parameter values of processing parameters used by the control unit 6 for processing.

  The storage unit 3 is configured by using a hard disk, ROM, RAM, and the like, and various information such as various processing programs executed by the control unit 6, various processing parameters used by the control unit 6 for processing, various processing results by the control unit 6, and the like. Remember. In particular, the storage unit 3 includes an observation image storage unit 3a that stores a series of observation images input from the input unit 2, an organ determination reference data storage unit 3b that stores reference data used for organ determination processing described later, and a later described A distance estimation reference data storage unit 3c that stores reference data used in the imaging distance estimation process, and an abnormality detection parameter storage unit 3d that stores parameter values of parameters used in an abnormal area detection process described later. The storage unit 3 includes a portable storage medium that can be attached to and detached from the image processing apparatus 1. The storage unit 3 acquires image data via the portable storage medium without using the input unit 2 and stores a series of observation images. be able to.

  The image processing unit 4 is realized by, for example, a CPU, and performs various image processing on a series of observation images stored in the observation image storage unit 3a based on a predetermined image processing program executed by the control unit 6. In particular, the image processing unit 4 includes an organ determination unit 4a that determines a predetermined organ as an observation target captured in each observation image, and an imaging distance estimation unit 4b that estimates an imaging distance when the observation image is captured. And an abnormal area detecting unit 4c for detecting an abnormal area as a characteristic image area having a predetermined characteristic from the observed image. Specifically, the organ determination unit 4a determines that the observation target captured in each observation image is one of the esophagus, stomach, small intestine, and large intestine, and the abnormal region detection unit 4c is determined by the organ determination unit 4a. An abnormal region that is an image region corresponding to the determined abnormal portion in the organ is detected.

  The output unit 5 is configured using various displays such as a liquid crystal display, and includes a series of observation images, determination results of the organ determination unit 4a, estimation results of the imaging distance estimation unit 4b, detection results of the abnormal region detection unit 4c, and the like. Various information is displayed. The output unit 5 includes a data communication interface, and can output various types of information from the data communication interface to an external device.

  The control unit 6 is realized by a CPU, and controls processing and operation of each unit included in the image processing apparatus 1 by executing a predetermined processing program stored in the storage unit 3. In particular, the control unit 6 executes an image processing control unit 6a that executes a predetermined image processing program stored in the storage unit 3 and causes the image processing unit 4 to process a series of observation images stored in the observation image storage unit 3a. Prepare. In addition, the control unit 6 causes the output unit 5 to output the processing result by the image processing unit 4 and the like.

  Subsequently, an image processing procedure performed by the image processing apparatus 1 will be described. FIG. 2 is a flowchart showing a processing procedure for processing a series of observation images stored in the observation image storage unit 3a by causing the control unit 6 to execute a predetermined image processing program. As shown in this figure, the image processing control unit 6a first reads a series of observation images from the observation image storage unit 3a (step S101), and determines an organ imaged in each observation image by the organ determination unit 4a. In addition to performing the process (step S102), an imaging distance estimation process is performed in which the imaging distance estimation unit 4b estimates the imaging distance when each observation image is captured (step S103). Thereafter, the image processing control unit 6a performs a parameter setting process for setting a parameter value to an abnormality detection parameter, which will be described later, based on the organ determination result of the organ determination process and the imaging distance estimation result of the imaging distance estimation process (step (S104) An abnormal area detection process is performed to cause the abnormal area detection unit 4c to detect an abnormal area using the abnormality detection parameter (step S105). Then, the image processing control unit 6a causes the output unit 5 to output the detection result of the abnormal area detection processing (step S106), and ends the series of processing.

  In the organ determination process in step S102, the organ determination unit 4a determines the organ imaged in each observation image based on the frequency component information of the observation image. For example, the esophagus and stomach are flat with less irregularities on the mucosal surface than the small intestine. Conversely, the small intestine has many irregularities on the surface due to villi and the like. For this reason, the low frequency component is dominant in the observation image in which the esophagus and the stomach are imaged, and the high frequency component is dominant in the observation image in which the small intestine is imaged. The organ determination unit 4a uses such properties to determine whether the organ imaged in the observation image is the esophagus or stomach, the small intestine, or the large intestine. Specifically, the organ determination unit 4a performs organ determination using, for example, a power spectrum obtained by Fourier transform as frequency component information.

  FIG. 3 is a flowchart showing an organ determination processing procedure by the organ determination unit 4a. As shown in this figure, the organ determination unit 4a calculates a feature vector based on the power spectrum for each observation image (step S111), reads the organ determination reference data from the organ determination reference data storage unit 3b (step S112), Based on the calculated feature vector and the read organ determination reference data, the organ imaged in each observation image is determined (step S113). Thereafter, the organ determination unit 4a ends the organ determination process and returns to step S102.

  In step S111, the organ determination unit 4a first calculates a power spectrum by Fourier transform for the processing target image in the series of observation images, and calculates a high frequency component, an intermediate frequency component, and a low frequency component from the calculated power spectrum. Extracted as feature amounts A to C, respectively. Then, the vector on the feature space indicated by the feature amounts A to C is associated with the processing target image as the feature vector indicating the frequency component information of the processing target image. Further, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and performs the same processing for each switched processing target image, thereby calculating the feature vector of each observation image.

  In step S112, the organ determination unit 4a reads organ determination reference data as a class dictionary in which each organ is classified in advance in the feature space, for example, as shown in FIG. In step S113, the organ determination unit 4a uses a known determination method such as the kNN method (k-Nearest Neighbor Method) or the subspace method, for example, based on the organ determination reference data read in step S112. The type of organ to which the feature vector of each observation image calculated in S111 belongs is determined. At that time, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and determines the type of organ to which the feature vector belongs for each switched processing target image. Thus, the organ determination unit 4a determines that the organ imaged in each observation image is one of the esophagus or stomach, the small intestine, and the large intestine, and associates the determination result with each observation image.

  Note that, here, the organ is determined based on the feature vector indicated by the three frequency components in the power spectrum. However, the number of frequency components used as the feature amount, that is, the number of dimensions of the feature amount is limited to three. Alternatively, two or four or more can be used. By using four or more dimensions, it is possible to perform organ determination with higher accuracy. However, since the processing time required for organ determination increases when the number of dimensions is four or more, it is preferable to appropriately set the number of dimensions according to desired determination accuracy. Here, the organ determination is performed by characterizing the power spectrum with the feature vector. However, the present invention is not limited to the feature vector. For example, the distribution pattern of the power spectrum is based on a reference distribution pattern obtained for each organ in advance. It is also possible to determine an organ by performing pattern matching.

  Next, in the imaging distance estimation process in step S103, the imaging distance estimation unit 4b captures the imaging distance when each observation image is captured, that is, the observation image, based on the luminance information or gain information of the observation image. A distance from a capsule endoscope or the like as an imaging device to an organ inner wall as a subject is estimated.

  Generally, in the observation image, the bounce brightness changes according to the imaging distance. The bounce luminance means the luminance when the illumination light emitted from the imaging device or the like is reflected by the subject and then received as observation light by the imaging device. The bounce brightness increases as the subject approaches the imaging device and decreases as the subject moves away. As shown in FIG. 5, the relationship between the imaging distance and the bounce luminance is known as the bounce luminance being inversely proportional to the square of the imaging distance. When the bounce luminance is high, the average luminance of the observation image is high, and when the bounce luminance is low, the luminance average is low. The imaging distance estimation unit 4b uses this property to estimate the imaging distance based on the average brightness of the observation image. Here, the average luminance of the observation image is an average value of the luminance values (pixel values) of all the pixels or a predetermined plurality of pixels in the observation image.

  On the other hand, when the imaging apparatus has an AGC (Auto Gain Control) function, generally, the gain is corrected according to the rebound brightness when the observation image is captured. That is, gain correction is performed so that the gain is lowered when the imaging distance is small and the rebound brightness is high, and the gain is increased when the imaging distance is large and the rebound brightness is low. The capsule endoscope is normally provided with an AGC function, and an AGC correction value that is a gain correction value is recorded as additional information for each captured observation image. The imaging distance estimation unit 4b uses this AGC function characteristic to estimate the imaging distance based on the AGC correction value recorded for each observation image.

  FIG. 6 is a flowchart showing a processing procedure of imaging distance estimation processing by the imaging distance estimation unit 4b. As shown in this figure, the imaging distance estimation unit 4b first determines whether or not an AGC correction value is recorded in the observation image (step S121). If the AGC correction value is recorded, the AGC correction value is also corrected. When the AGC correction value is not recorded, the imaging distance is estimated based on the luminance average.

  Specifically, when the AGC correction value is recorded in the observation image (step S121: Yes), the imaging distance estimation unit 4b acquires the AGC correction value from each observation image (step S122), and in all the observation images. The overall average of the AGC correction values is calculated (step S123), and the imaging distance of each observation image is estimated based on the AGC correction value and the overall average (step S124).

  On the other hand, when the AGC correction value is not recorded in the observation image (step S121: No), the imaging distance estimation unit 4b calculates the luminance average for each observation image (step S125), and the luminance average in all the observation images. Is calculated (step S126), and the imaging distance of each observation image is estimated based on the luminance average and the overall average (step S127). Then, the imaging distance estimation unit 4b ends the imaging distance estimation process after step S124 or S127, and returns to step S103.

In step S124, the imaging distance estimation unit 4b calculates the imaging distance X for the processing target image in the series of observation images based on the AGC correction value C and the overall average _Cave by the following equation (1). presume. Furthermore, the imaging distance estimation unit 4b sequentially switches the processing target images in the series of observation images, and performs the same calculation for each switched processing target image, thereby estimating the imaging distance of each observation image. Note that the coefficient f (C _ave ) in Equation (1) is a coefficient determined by a predetermined calculation based on the overall average C _ave .

In step S127, the imaging distance estimation unit 4b estimates the imaging distance X by the following equation (2) based on the luminance average E and the overall average E _ave for the processing target image in the series of observation images. To do. Then, the imaging distance estimation unit 4b sequentially switches the processing target images in the series of observation images, and performs the same calculation for each switched processing target image, thereby estimating the imaging distance of each observation image. Note that the coefficient f (E _ave ) in Equation (2) is a coefficient determined by a predetermined calculation based on the overall average E _ave .

  In addition, when the AGC correction value is recorded in the observation image, the imaging distance estimation unit 4b uses the AGC correction value and the luminance average in combination instead of estimating the imaging distance based only on the AGC correction value. An imaging distance can also be estimated. In this case, the imaging distance estimation unit 4b performs both the processing of steps S122 to S124 and the processing of steps S125 to 127, and can use, for example, an average value of imaging distances estimated by each processing as a final estimation result. .

  Next, in the parameter setting process in step S104, the image processing control unit 6a determines, for each observation image, the organ determined by the organ determination process in step S102 and the imaging distance estimated by the imaging distance estimation process in step S103. First, parameter values are set for the abnormality detection parameters. Here, the abnormality detection parameter is one of the processing parameters used by the abnormal region detection unit 4c for the abnormal region detection processing in step S105, and processes a parameter indicating a predetermined feature amount of the abnormal region or an observation image. In this case, a parameter indicating a pixel block size (processing unit pixel block size) as a processing unit is included.

  In general, the characteristics of an image region to be detected as an abnormal region differ depending on the types of organs such as the stomach, small intestine, and large intestine. For example, the region size (size) of an abnormal region to be detected in the small intestine is detected in the stomach. Smaller than the region size For this reason, when an abnormal region is detected, if the lower limit threshold as a criterion for determining the region size as the feature amount is set equal to the stomach and the small intestine, an image of the region size to be detected in the small intestine An area is not detected as an abnormal area, and an image area having an area size that does not need to be detected in the stomach may be erroneously detected as an abnormal area. In contrast, by setting appropriate lower thresholds for the stomach and the small intestine, it is possible to reliably detect abnormal areas with characteristic sizes corresponding to the stomach and small intestine, and to detect them as abnormal areas. Image regions that do not need to be performed can be excluded from detection targets.

  On the other hand, for example, even in the case of observation images obtained by imaging the same abnormal portion, if the imaging distances are different, the area sizes of the abnormal areas on the observation image are different. Specifically, the area size of the abnormal area decreases as the imaging distance increases. Therefore, when detecting an abnormal area, if the lower limit threshold of the area size is set equal regardless of the imaging distance, the area size of the abnormal area to be detected is too small to be detected when the imaging distance is large. When the imaging distance is small, a feature image area having an area size that does not need to be detected may be erroneously detected as an abnormal area. On the other hand, by setting an appropriate lower limit threshold according to the imaging distance, it is possible to reliably detect an abnormal area to be detected regardless of the imaging distance, and a feature image area that does not need to be detected as an abnormal area Can be excluded from detection targets.

  From the above, in the image processing apparatus 1, for example, as shown in FIG. 7, the lower limit threshold value of the appropriate region size for each organ according to the imaging distance is abnormal in advance as the parameter value of the size parameter that is an abnormality detection parameter. It is stored in the detection parameter storage unit 3d. Then, the image processing control unit 6a sets parameter values corresponding to the organ determined by the organ determination unit 4a and the imaging distance estimated by the imaging distance estimation unit 4b for each observation image in the parameter setting process of step S104. The data is read from the abnormality detection parameter storage unit 3d, and the read parameter value is set as an abnormality detection parameter.

  Here, the parameter value indicating the region size of the abnormal region has been described as the parameter value of the abnormality detection parameter stored in advance in the abnormality detection parameter storage unit 3d. However, it is not necessary to interpret the parameter value limited to the region size. It may be a parameter value indicating another feature amount of the region, for example, a color. Further, the parameter value is not limited to the feature amount of the abnormal region, and may be a parameter value indicating the processing unit pixel block size of the observation image, for example.

  Furthermore, the parameter values stored in advance are not limited to one type, and parameter values for a plurality of abnormality detection parameters can be stored. In that case, for example, the abnormality detection parameter storage unit 3d stores the parameter value for each of a plurality of abnormality detection parameters in the storage table as shown in FIG. 7, and the image processing control unit 6a performs abnormality detection in the parameter setting process. It is preferable to read the parameter value from the corresponding storage table for each parameter and set the read parameter value.

  Next, in the abnormal region detection process in step S105, the abnormal region detection unit 4c sets a parameter value corresponding to the processing target image in the parameter setting processing in step S104 for the processing target image in the series of observation images. A predetermined abnormality area is detected from the processing target image using the abnormality detection parameter thus set. Specifically, the abnormal area detection unit 4c uses, for example, a size parameter in which a parameter value stored in the storage table shown in FIG. 7 is set, and is an area smaller than the area size indicated by the parameter value. An image area having a predetermined characteristic (for example, color) as an abnormal area is detected as an abnormal area. Furthermore, the abnormal area detection unit 4c detects a desired abnormal area from each observation image by sequentially switching the processing target images in the series of observation images and performing the same processing for each switched processing target image.

The abnormal area detection unit 4c is not limited to the process using the size parameter, and can perform the abnormal area detection process by various processes. For example, as disclosed in Patent Document 2, each pixel in the processing target image or the averaged pixel is mapped to a feature space based on the color information, and a normal cluster and an abnormal cluster are specified after clustering. Thus, the pixel area belonging to the abnormal part cluster can be detected as an abnormal area. In this case, the abnormal area detection unit 4c performs an abnormal area detection process using a color parameter that is an abnormality detection parameter in which a parameter value indicating color information (for example, color or chromaticity) constituting the feature space is set. Good. And the abnormality detection parameter memory | storage part 3d is good to memorize | store the parameter value set to the color parameter beforehand for every organ according to an imaging distance.

  In addition, as disclosed in Patent Document 1, for example, the abnormal region detection unit 4c divides a processing target image into a plurality of processing unit pixel blocks, and previously creates color information for each processing unit pixel block. The abnormal region can be detected by comparing the color reference of the abnormal part and the color reference of the normal part. In this case, the abnormal area detection unit 4c may perform the abnormal area detection process using at least one of the color parameter and the block size parameter. Here, the color parameter is a parameter in which a parameter value indicating the hue or chromaticity of each color reference is set, and the block size parameter is an abnormality detection in which a parameter value indicating the processing unit pixel block size is set It is a parameter. Then, the abnormality detection parameter storage unit 3d may store in advance the parameter value set for at least one of the color parameter and the block size parameter for each organ according to the imaging distance.

  When the imaging magnification is not constant for each image, the capsule endoscope records magnification information as additional information in each observation image. When the abnormal area detection process is performed using the size parameter or the block size parameter described above, the abnormal area may be detected in consideration of the magnification.

  As described above, in the image processing apparatus 1 according to the present embodiment, an organ as an observation target captured in a processing target image in a series of observation images is an esophagus or stomach that is a predetermined one or more observation targets. An abnormality that detects an abnormal region as a feature image region having a predetermined feature from the processing target image using the organ determination unit 4a that determines one of the small intestine and the large intestine and the abnormality detection parameter as the predetermined processing parameter An area detection unit 4c, an image processing control unit 6a that sets a parameter value corresponding to the determination result of the organ determination unit 4a as an abnormality detection parameter, and causes the abnormal region detection unit 4c to detect an abnormal region using the abnormality detection parameter; Therefore, it is possible to reliably detect an abnormal area as an image area corresponding to an abnormal part or the like showing different characteristics depending on the type of organ from the processing target image. It is possible to improve the detection accuracy of Rutoki.

  Further, the image processing apparatus 1 includes an imaging distance estimation unit 4b that estimates an imaging distance when a processing target image is captured, and the image processing control unit 6a includes the determination result of the organ determination unit 4a and the imaging distance estimation unit 4b. Since the parameter value corresponding to the estimation result is set as an abnormality detection parameter, and the abnormal region detection unit 4c is used to detect the abnormal region using this abnormality detection parameter, it differs depending on the type of organ and the imaging distance. It is possible to improve the detection accuracy when the abnormal region indicating the feature is reliably detected from the processing target image.

  Further, since the image processing apparatus 1 sequentially switches the processing target images in the series of observation images and detects the abnormal region for each of the switched processing target images, the abnormal region is detected from within each observation image in the series of observation images. Can be detected.

(Modification 1)
Next, Modification Example 1 of the image processing apparatus according to the present embodiment will be described. In the organ determination process described above, the organ determination unit 4a performs the organ determination based on, for example, the frequency component information of the observation image indicated by the power spectrum. The organ is determined based on this.

  Usually, since there are relatively few irregularities inside the esophagus and stomach, the observation image obtained by imaging the esophagus or stomach has a higher correlation between each pixel and its surrounding pixels than the observation image obtained by imaging the small or large intestine. There is. Therefore, the organ determination unit 4a can determine whether the organ imaged in the observation image is the esophagus or stomach, the small intestine, or the large intestine by obtaining the height of the correlation.

In general, it is known that the level of correlation between each pixel and its surrounding pixels is indicated by the amount of image information, and the amount of image information is indicated by entropy. The entropy H (f) is obtained by the following equation (3) using the bit string r of the surrounding pixels of the target pixel and the probability p (r; f) that the target pixel has the pixel value f. Here, the entropy H (f) shown by the equation (3) is the entropy of the Markov information source.
H (f) = − log ₂ (p (r; f)) (3)

  The entropy H (f) corresponding to each pixel can be obtained by performing the calculation according to the equation (3) on the entire image. When the value of entropy H (f) is large as a tendency of the entire image, the image is an image having a low correlation between each pixel and its surrounding pixels, and when the value of entropy H (f) is small, It can be said that the image has a high correlation between each pixel and its surrounding pixels. The organ determination unit 4a according to the first modification uses this characteristic to calculate the entropy H (f) for each observation image, and based on the calculation result, the organ imaged in the observation image Whether the stomach is the small intestine or the large intestine is determined.

  FIG. 8 is a flowchart showing a processing procedure of an organ determination process by the organ determination unit 4a according to the first modification. As shown in this figure, the organ determination unit 4a calculates entropy according to equation (3) for each observation image (step S211), calculates the overall average of entropy in all observation images (step S212), Based on the overall average and the organ determination reference data stored in the organ determination reference data storage unit 3b in advance, the organ imaged in each observation image is determined (step S213). Thereafter, the organ determination unit 4a ends the organ determination process and returns to step S102.

  Here, as shown in FIG. 9, for example, as shown in FIG. 9, a storage table in which an organ is associated with each entropy and its overall average is stored as organ determination reference data in the organ determination reference data storage unit 3b. The organ determination reference data can be created in advance based on knowledge about each organ.

  In step S213, the organ determination unit 4a determines that the organ associated with the entropy and the overall average is captured by the organ determination reference data with respect to the processing target image in the series of observation images. For example, when the entropy of the processing target image and the overall average of the entropy are both 0.2, the organ determination unit 4a uses the organ determination reference data shown in FIG. 9 to capture the organ imaged in the processing target image. Is determined to be the small intestine. Furthermore, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and performs the same processing for each switched processing target image, thereby determining the organ imaged in each observation image.

(Modification 2)
Next, a second modification of the image processing apparatus according to this embodiment will be described. In the second modification, the organ determination unit 4a performs organ determination based on the texture information of the observation image. Usually, since there are relatively few irregularities in the esophagus or stomach, there is little texture information in an observation image obtained by imaging the esophagus or stomach. On the other hand, in the small intestine, since many uneven patterns such as villi are observed on the surface thereof, a lot of texture information is obtained in an observation image obtained by imaging the small intestine. Therefore, the organ determination unit 4a can determine whether the organ imaged in the observation image is the esophagus or stomach, the small intestine, or the large intestine based on the texture information of the observation image.

  In general, as a texture information, a method for obtaining a statistical feature value of a texture using a co-occurrence matrix is known. The co-occurrence matrix is used to obtain a feature amount indicating properties such as uniformity, directionality, and contrast of pixel values from pixel values of pixel pairs at two distant points. In the second modification, a case will be described in which correlation and entropy are obtained by using a co-occurrence matrix as texture information.

  FIG. 10 is a flowchart illustrating a processing procedure of an organ determination process performed by the organ determination unit 4a according to the second modification. As shown in this figure, the organ determination unit 4a calculates the correlation and entropy by the co-occurrence matrix for each observation image (step S311), calculates the overall average of the correlation and entropy in all the observation images (step S312). ), And a feature vector based on the calculated correlation and entropy and their respective averages is calculated (step S313). Then, the organ determination unit 4a reads the organ determination reference data from the organ determination reference data storage unit 3b (step S314), and is captured in each observation image based on the calculated feature vector and the read organ determination reference data. The determined organ is determined (step S315). Thereafter, the organ determination unit 4a ends the organ determination process and returns to step S102.

  In step S311, the organ determination unit 4a first calculates correlation and entropy using a co-occurrence matrix for the processing target images in the series of observation images, and sequentially switches the processing target images within the series of observation images. By performing the same processing for each switched processing target image, the feature vector of each observation image is calculated.

  In step S313, the organ determination unit 4a first four-dimensionally calculates the correlation and entropy calculated in step S311 and the overall averages of correlation and entropy calculated in step S312 for the processing target image in the series of observation images. Feature amount. Then, a vector on the feature space indicated by the four-dimensional feature amount is calculated as a feature vector indicating the texture information of the processing target image, and is associated with the processing target image. Further, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and performs the same processing for each switched processing target image, thereby calculating the feature vector of each observation image.

  In step S314, the organ determination unit 4a reads organ determination reference data as a class dictionary in which each organ is classified in advance in a four-dimensional feature space. In step S315, the organ determination unit 4a uses, for example, a known determination method such as the kNN method or the subspace method, and each observation image calculated in step S313 based on the organ determination reference data read in step S314. The type of organ to which the feature vector belongs is determined. At that time, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and determines the type of organ to which the feature vector belongs for each switched processing target image. Thus, the organ determination unit 4a determines that the organ imaged in each observation image is one of the esophagus or stomach and the small intestine or large intestine, and associates the determination result with each observation image.

  Note that, here, it has been described that the organ is determined based on the feature vector having the correlation and power spectrum calculated from the co-occurrence matrix and the respective overall average as the feature amount. However, the feature amount constituting the feature vector Is not limited to correlation and entropy, and other feature quantities that can be calculated from the co-occurrence matrix can also be used. Further, the feature quantity constituting the feature vector is not limited to four dimensions, but may be two dimensions or five dimensions or more. When the feature quantity is five dimensions or more, organ determination can be performed with higher accuracy. However, since the processing time for organ determination increases when the number of dimensions is five or more, it is preferable to set the number of dimensions appropriately according to the desired determination accuracy. Note that the overall average is used as the feature quantity constituting the feature vector as described above, because there is an individual difference in organ characteristics for each subject, and the influence of the individual difference is reduced.

(Modification 3)
Next, Modification Example 3 of the image processing apparatus according to the present embodiment will be described. In the third modification, the organ determination unit 4a determines whether the organ imaged in each observation image is the esophagus or the stomach, the small intestine or the large intestine based on the file size of the compressed image data obtained by compression-coding the observation image. It is determined whether it is. Usually, since there are relatively few irregularities inside the esophagus and stomach, the observation image obtained by imaging the esophagus or stomach has a higher correlation between each pixel and its surrounding pixels than the observation image obtained by imaging the small or large intestine. There is. The organ determination unit 4a determines whether the organ captured in the observation image is the esophagus or stomach, the small intestine, or the large intestine by determining the height of the correlation based on the file size of the compressed image data.

  The organ determination unit 4a determines whether the organ imaged in the observation image is the small intestine or the large intestine based on the amount of change in the file size of the compressed image data of the observation image. Usually, in the large intestine, stool etc. are clogged inside, so when acquiring observation images with a capsule endoscope, for example, the movement of the capsule endoscope is stagnant, and files between observation images that are continuous in time series The size hardly changes. On the other hand, in the small intestine, since the capsule endoscope can move more smoothly than in the large intestine, a significant change is seen in the file size between observation images that are continuous in time series. The organ determination unit 4a uses this characteristic to determine whether the observation target imaged in the observation image is the small intestine based on the amount of change in the file size between the observation images continuous in time series. Determine if it is the large intestine.

  FIG. 11 is a flowchart showing the organ determination processing procedure. As shown in this figure, the organ determination unit 4a first calculates a moving average of the file size based on the file size of the compressed image data in the series of observation images (step S411), and calculates the overall average of the file size. Calculate (step S412). Furthermore, the organ determination unit 4a calculates the change amount of the file size between each successive observation image in the series of observation images, calculates the moving average of the change amount of the file size (step S413), and also determines the file size. An overall average of the amount of change is calculated (step S414). Thereafter, the organ determination unit 4a determines the organ imaged in each observation image based on the calculation results in steps S411 to S413 (step S415), ends the organ determination process, and returns to step S102.

  In step S411, the organ determination unit 4a calculates a size average, which is an average of file sizes of a plurality of observation images close to the time series including the processing target image, with respect to the processing target image among the series of observation images. Then, the calculated size average is associated with the processing target image. In the third modification, the average size is calculated using, for example, 100 observation images close to the time series in the series of observation images. However, the number of observation images for calculating the average size may be set to an appropriate number depending on the imaging interval when a series of observation images are captured. The organ determination unit 4a sequentially switches the processing target images in the series of observation images, and calculates a size average for each of the switched processing target images, thereby obtaining a moving average of file sizes over the entire series of observation images. Specifically, the organ determination unit 4a obtains a moving average of file sizes as shown in FIG. 12-2 based on the file size information of a series of observation images shown in FIG. .

  In step S413, the organ determination unit 4a averages the amount of change in file size between the observation images in a plurality of observation images close to the time series including the processing target image with respect to the processing target image among the series of observation images. The average amount of change is calculated. Then, the calculated change amount average is associated with the processing target image. In the third modification, the change amount average is calculated using, for example, 100 observation images close to the time series in the series of observation images. However, the number of observation images for calculating the average amount of change may be set to an appropriate number depending on the imaging interval when a series of observation images are captured. The organ determination unit 4a sequentially switches the processing target images within the series of observation images, and calculates the average amount of change for each switched processing target image, thereby calculating a moving average of the amount of change in file size over the entire series of observation images. obtain. Specifically, in step S413, the organ determination unit 4a moves the change amount of the file size as shown in FIG. 12-3 based on the file size information of a series of observation images shown in FIG. Get the average.

In step S415, the organ determination unit 4a first captures the processing target image of the processing target image in the series of observation images according to the size relationship between the size average calculated in step S411 and a predetermined size determination criterion. It is determined whether the organ formed is the esophagus or stomach, small intestine or large intestine. Specifically, the organ determination unit 4a calculates a threshold value T _Fsize as a size determination criterion by the following equation (4) based on the overall average F _sizeAve calculated in step S412 and a preset variable M. Then (see FIG. 12-2), it is determined whether or not the average _size F _size satisfies the following equation (5) with respect to the threshold value T _Fsize .
T _Fsize = F _sizeAve + M (4)
F _size <T _Fsize (5)

  When the expression (5) is satisfied, the organ determination unit 4a determines that the organ imaged in the processing target image is the esophagus or the stomach. When the expression (5) is not satisfied, the organ determination unit 4a determines that the organ is the small intestine or the large intestine. To do. Then, this determination result is associated with the processing target image. Furthermore, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and performs the same determination for each of the switched processing target images, so that the organs captured in each observation image in the series of observation images are displayed. Determine if it is the esophagus or stomach, small intestine or large intestine.

  In addition, when it is clear that the stomach, the small intestine, and the large intestine are captured in this order in the series of observation images, the organ determination unit 4a satisfies the formula (5) by switching the processing target images in order from the top image. Otherwise, all observation images after the first observation image determined are determined as images in which the small intestine or the large intestine is imaged. As a result, an observation image obtained by imaging the esophagus or stomach and an observation image obtained by imaging the small intestine or the large intestine can be discriminated more quickly.

Next, the organ determination unit 4a performs an average change amount calculated in step S413 on a processing target image among a series of observation images determined that the small intestine or the large intestine is imaged, and a predetermined change amount determination criterion. According to the size relationship, it is determined whether the organ imaged in the processing target image is the small intestine or the large intestine. Specifically, the organ determination unit 4a calculates a threshold value T _FsizeDiff as a variation determination criterion by the following equation (6) based on the overall average F _sizeDiffAve calculated in step S414 and a preset variable N. It is calculated (see FIG. 12-3), and it is determined whether or not the variation average F _sizeDiff satisfies the following equation (7) with respect to the threshold value T _FsizeDiff .
T _FsizeDiff = F _sizeDiffAve + N (6)
F _sizeDiff <T _FsizeDiff (7)

  The organ determination unit 4a determines that the organ imaged in the processing target image is the large intestine when Expression (7) is satisfied, and determines that it is the small intestine when Expression (7) is not satisfied. Then, this determination result is associated with the processing target image. Further, the organ determination unit 4a sequentially switches the processing target images in a series of observation images that have been determined that the small intestine or the large intestine has been previously captured, and performs the same determination for each switched processing target image. It is determined whether the organ imaged in each observation image is the small intestine or the large intestine. Thereby, the organ determination unit 4a determines that the organ imaged in each observation image in the series of all observation images is one of the esophagus or stomach, the small intestine, and the large intestine, and the determination result is displayed in each observation image. Can be associated.

Note that, as shown in the equation (4), the use of the overall average F _sizeAve of the file size for the calculation of the threshold value T _Fsize as the size determination criterion is because there are individual differences in organ characteristics for each subject, and the individual This is to reduce the influence of the difference. Similarly, as shown in Expression (6), the overall average F _sizeDiffAve of the change amount of the file size is used for calculating the threshold value T _FsizeDiff as the change amount determination criterion in order to reduce the influence of individual differences. . Further, the variables M and N are set by the observer inputting from the input unit 2 and can be appropriately changed.

  In the organ determination process described above, the organ determination unit 4a collectively determines the organs captured in each observation image in step S415. However, the determination based on Expression (5) and the determination based on Expression (7) are performed. Can also be performed individually. For example, by performing the determination according to the equation (5) immediately after step S412, step S413 can be performed only on the observation image determined that the small intestine or the large intestine has been imaged. Judgment processing can be performed.

Further, in the organ determination process described above, the organ determination unit 4a has been described as performing the determination based on the equation (5) and the determination based on the equation (7) sequentially in step S415. . For example, a feature vector (F _size , F _sizeDiff ) represented by a size average F _size and a variation average F _sizeDiff is obtained for each processing target image, and an organ is determined according to the region to which the feature vector belongs in the feature space. can do. Specifically, if the feature vector (F _size , F _sizeDiff ) is within the region satisfying the equation (5), it is determined that the esophagus or stomach is imaged, and the region other than that is represented by the equation (7 ), It can be determined that the large intestine has been imaged. Furthermore, when it exists in the area | region other than that, it can determine with the small intestine being imaged.

  In the organ determination process described above, the organ determination unit 4a determines the organ based on the average file size and the average change amount in the plurality of observation images. However, it is not always necessary to use the average. Alternatively, for example, organ determination can be performed based on the individual file size and the amount of change in the individual file size. As a result, the organ determination process can be performed more quickly when the request for determination accuracy is relatively loose.

(Modification 4)
Next, a fourth modification of the image processing apparatus according to this embodiment is described. In the fourth modification, the organ determination unit 4a performs organ determination based on the DCT coefficient calculated when decompressing the compressed image data and the amount of change.

  Usually, the esophagus and stomach are flat with less irregularities on the mucosal surface than the small intestine. Conversely, the small intestine has many irregularities on the surface due to villi and the like. For this reason, the low frequency component is dominant in the observation image in which the stomach is imaged, and the high frequency component is dominant in the observation image in which the small intestine is imaged. In the fourth modification, the organ determination unit 4a uses this property to determine whether the organ imaged in the observation image is the esophagus or stomach, the small intestine, or the large intestine. Specifically, when the series of observation images is stored as compressed image data compressed by the DCT compression encoding method such as JPEG, the organ determination unit 4a obtains the inverse DCT conversion performed when decompressing the compressed image data. The determination is made based on a plurality of DCT coefficients.

  In addition, since the large intestine is clogged with stool or the like, for example, when obtaining an observation image with a capsule endoscope, the movement of the capsule endoscope is stagnant, and the frequency between observation images that are continuous in time series There is almost no change in ingredients. On the other hand, in the small intestine, since the capsule endoscope can move more smoothly than in the large intestine, there is a significant change in the frequency component between observation images that are continuous in time series. The organ determination unit 4a uses this characteristic to determine whether the organ imaged in the observation image is the small intestine based on the amount of change in the frequency component between the observation images continuous in time series. It is determined whether it is. Specifically, the organ determination unit 4a, when a series of observation images is stored as compressed image data compressed by the DCT compression encoding method, changes in the DCT coefficient between observation images continuous in time series. Judgment is made based on the above.

  In general, as a method for obtaining frequency component information in an image, a method for obtaining a power spectrum by Fourier transform is well known. However, since the Fourier transform has a lot of calculation processing, a huge amount of processing time is usually required. On the other hand, when the frequency component is determined based on the DCT coefficient as described above, the DCT coefficient can be calculated during the decompression process of the compressed image data, and a special calculation process is required to determine the frequency component. do not do. In addition, since the DCT coefficient calculation process itself is simple and can be performed in a short time, the frequency component in the observation image can be quickly determined and compared with the observation image compared to the case where a power spectrum by Fourier transform is used. The imaged organ can be determined.

  FIG. 13 is a flowchart showing the organ determination processing procedure. As shown in this figure, the organ determination unit 4a first calculates a representative DCT coefficient as a weighted average of DCT coefficients for each observation image (step S510). Based on the representative DCT coefficients in the series of observation images, a moving average of the representative DCT coefficients is calculated (step S511), and an overall average of the representative DCT coefficients is calculated (step S512). Furthermore, the organ determination unit 4a calculates the amount of change in the representative DCT coefficient between successive observation images in the series of observation images, calculates the moving average of the amount of change in the representative DCT coefficient (step S513), and represents the representative. An overall average of the amount of change in the DCT coefficient is calculated (step S514). Thereafter, the organ determination unit 4a determines the organ imaged in each observation image based on the calculation results in steps S511 to S513 (step S515), ends the organ determination process, and returns to step S102.

  In step S510, the organ determination unit 4a first determines, based on a plurality of predetermined DCT coefficients ranging from a low frequency component to a high frequency component, for each 8 × 8 pixel block that is a processing unit when decompressing compressed image data for each observation image. The block average is calculated. Specifically, based on the DCT coefficients “DCT1” to “DCT64” of the 8 × 8 pixel block obtained as shown in FIG. 14, except for “DCT1” corresponding to the DC component, “DCT2” to “DCT64” "A weighted average for each frequency or a weighted average for each frequency of one or more DCT coefficients selected in advance from" DCT2 "to" DCT64 "is used as a block average. In addition, in the weighting for every frequency, it is good to make weighting heavy with a high frequency rather than a low frequency. Furthermore, the organ determination unit 4a calculates an overall average obtained by further averaging the block average of each 8 × 8 pixel block for each observation image as a representative DCT coefficient.

  In step S511, the organ determination unit 4a calculates a DCT coefficient average that is an average of the representative DCT coefficients in a plurality of observation images close to the time series including the processing target image with respect to the processing target image among the series of observation images. To do. Then, the calculated DCT coefficient average is associated with the processing target image. In the fourth modification, the DCT coefficient average is calculated using, for example, 100 observation images close to the time series in the series of observation images. However, the number of observation images for calculating the DCT coefficient average may be set to an appropriate number depending on the imaging interval when a series of observation images are captured. The organ determination unit 4a sequentially switches the processing target images in the series of observation images and calculates a DCT coefficient average for each of the switched processing target images, thereby obtaining a moving average of the representative DCT coefficients over the entire series of observation images.

  In step S513, the organ determination unit 4a changes the representative DCT coefficient between the observation images in a plurality of observation images close to the time series including the processing target image with respect to the processing target image in the series of observation images. The average of the DCT change amount that is the average of the above is calculated. Then, the calculated DCT variation average is associated with the processing target image. In the fourth modification, the DCT variation average is calculated using, for example, 100 observation images close to the time series in the series of observation images. However, the number of observation images for calculating the DCT variation average may be set to an appropriate number depending on the imaging interval when a series of observation images are captured. The organ determination unit 4a sequentially switches the processing target images within the series of observation images, and calculates the DCT variation average for each of the switched processing target images, thereby moving the variation of the representative DCT coefficient over the entire series of observation images. Get the average.

In step S515, the organ determination unit 4a applies the processing target image to the processing target image in accordance with the magnitude relationship between the DCT coefficient average calculated in step S511 and a predetermined DCT determination criterion. It is determined whether the imaged organ is the esophagus or stomach, small intestine or large intestine. Specifically, the organ determination unit 4a calculates a threshold T _dct as a DCT determination criterion by the following equation (8) based on the overall average F _dctAve calculated in step S512 and a preset variable K. Then, it is determined whether or not the DCT coefficient average F _dct satisfies the following equation (9) with respect to the threshold value T _dct .
T _dct = F _dctAve + K (8)
F _dct <T _dct (9)

  When the expression (9) is satisfied, the organ determination unit 4a determines that the organ imaged in the processing target image is the esophagus or the stomach. When the expression (9) is not satisfied, the organ determination unit 4a determines that the organ is the small intestine or the large intestine. To do. Then, this determination result is associated with the processing target image. Furthermore, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and performs the same determination for each of the switched processing target images, so that the organs captured in each observation image in the series of observation images are displayed. Determine if it is the esophagus or stomach, small intestine or large intestine.

  When it is clear that the stomach, the small intestine, and the large intestine are captured in this order in the series of observation images, the organ determination unit 4a satisfies the formula (9) by switching the processing target images in order from the top image. Otherwise, it is determined that all observation images after the first observation image determined are taken from the small intestine or the large intestine. As a result, an observation image obtained by imaging the esophagus or stomach and an observation image obtained by imaging the small intestine or the large intestine can be discriminated more quickly.

Next, the organ determination unit 4a determines the DCT variation average calculated in step S513 and a predetermined DCT variation determination criterion with respect to the processing target image among a series of observation images determined to have captured the small intestine or the large intestine. Is determined whether the organ imaged in the processing target image is the small intestine or the large intestine. Specifically, the organ determination unit 4a, based on the overall average F _dctDiffAve calculated in step S514 and the variable L set in advance, _uses the threshold T _dctDiff as a DCT variation determination criterion according to the following equation (10). _Is calculated, and it is determined whether or not the DCT variation average F _dctDiff satisfies the following equation (11) with respect to the threshold value T _dctDiff .
T _dctDiff = F _dctDiffAve + L (10)
F _dctDiff <T _dctDiff (11)

  The organ determination unit 4a determines that the organ imaged in the processing target image is the large intestine when Expression (11) is satisfied, and determines that it is the small intestine when Expression (11) is not satisfied. Then, this determination result is associated with the processing target image. Further, the organ determination unit 4a sequentially switches the processing target images in a series of observation images that have been determined that the small intestine or the large intestine has been previously captured, and performs the same determination for each switched processing target image. It is determined whether the organ imaged in each observation image is the small intestine or the large intestine. Thereby, the organ determination unit 4a determines that the organ imaged in each observation image in the series of all observation images is one of the esophagus or stomach, the small intestine, and the large intestine, and the determination result is displayed in each observation image. Can be associated.

Note that, as shown in the equation (8), the overall average F _dctAve of the representative DCT coefficient is used for calculating the threshold value T _dct as a DCT determination criterion because there is an individual difference in organ characteristics for each subject. This is to reduce the influence of individual differences. Similarly, as shown in the equation (10), the use of the overall average F _{dctDiffAve of the} amount of change in the representative DCT coefficient for the calculation of the threshold value T _dctDiff as the DCT change amount determination criterion also reduces the influence of individual differences. It is. The variables K and L are set by the observer inputting from the input unit 2 and can be appropriately changed.

  In the organ determination process described above, the organ determination unit 4a collectively determines the organs captured in each observation image in step S515. However, the determination based on Expression (9) and the determination based on Expression (11) are performed. Can also be performed individually. For example, by performing the determination according to the equation (9) immediately after step S512, step S513 can be performed only on the observation image determined that the small intestine or the large intestine has been imaged. Judgment processing can be performed.

Further, in the organ determination process described above, the organ determination unit 4a has been described as performing the determination based on the equation (9) and the determination based on the equation (11) sequentially in step S515. However, the determination can be performed collectively. . For example, a feature vector (F _dct , F _dctDiff ) represented by a DCT coefficient average F _dct and a DCT change amount average F _dctDiff is obtained for each processing target image, and an organ is determined according to the region to which the feature vector belongs in the feature space. Can be determined. Specifically, if the feature vector (F _dct , F _dctDiff ) is within a region that satisfies the equation (9), it is determined that the esophagus or stomach is imaged, and the region other than that is represented by the equation (11 ), It can be determined that the large intestine has been imaged. Furthermore, when it exists in the area | region other than that, it can determine with the small intestine being imaged.

  In the organ determination process described above, the organ determination unit 4a performs organ determination based on the DCT coefficient average and the DCT variation average in a plurality of observation images. However, it is not always necessary to use the average. For example, organ determination can also be performed based on individual representative DCT coefficients and the amount of change of each representative DCT coefficient. As a result, the organ determination process can be performed more quickly when the request for determination accuracy is relatively loose.

(Modification 5)
Next, a fifth modification of the image processing apparatus according to this embodiment is described. In the fifth modification, the organ determination unit 4a obtains a feature vector based on a plurality of DCT coefficients for each observation image, and performs organ determination based on the feature vector.

  FIG. 15 is a flowchart showing a processing procedure of organ determination processing by the organ determination unit 14a. As shown in this figure, the organ determination unit 4a first calculates a feature vector based on the DCT coefficient for each observation image (step S611), and reads the organ determination reference data from the organ determination reference data storage unit 3b (step S611). S612). Then, based on the calculated feature vector and the read organ determination reference data, the organ imaged in each observation image is determined (step S613), the organ determination process is terminated, and the process returns to step S102.

  In step S611, the organ determination unit 4a first sets a predetermined one or more DCT coefficients for each 8 × 8 pixel block, which is a processing unit when decompressing compressed image data, for a processing target image in a series of observation images. In addition, the block representative value of the low frequency component and the block representative value of the high frequency component are calculated. Specifically, based on the DCT coefficients “DCT1” to “DCT64” of the 8 × 8 pixel block obtained as shown in FIG. 14, for example, “DCT2” to “DCT10” as block representative values of the low frequency component. A weighted average for each frequency is calculated, and a weighted average for each frequency of “DCT55” to “DCT64” is calculated as a block representative value of the high frequency component. In addition, in the weighting for every frequency, it is good to make weighting heavy with a high frequency rather than a low frequency.

  Furthermore, the organ determination unit 4a sets the block representative value of the low frequency component of each 8 × 8 pixel block in the processing target image, the block representative value of the high frequency component, and “DCT1” indicating the DC component in the processing target image. Are averaged over all 8 × 8 pixel blocks and calculated as feature amounts D to F. Then, the vector on the feature space indicated by the feature amounts D to F is associated with the processing target image as a feature vector indicating the frequency distribution of the processing target image. Further, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and performs the same processing for each switched processing target image, thereby calculating the feature vector of each observation image.

  In step S612, the organ determination unit 4a reads organ determination reference data as a class dictionary in which each organ is classified in advance in the feature space, for example, as shown in FIG. In step S613, the organ determination unit 4a uses a known determination method such as the kNN method (k-Nearest Neighbor Method) or the subspace method, and performs step determination based on the organ determination reference data read in step S612. The type of organ to which the feature vector of each observation image calculated in S611 belongs is determined. At that time, the organ determination unit 4a sequentially switches the processing target images in the series of observation images, and determines the type of organ to which the feature vector belongs for each switched processing target image. Thus, the organ determination unit 4a determines that the organ imaged in each observation image is one of the esophagus or stomach, the small intestine, and the large intestine, and associates the determination result with each observation image.

  In the organ determination process described above, the organ determination unit 4a calculates the feature vector based on the three feature amounts D to F and performs the organ determination. Alternatively, a feature vector can be calculated based on four or more feature amounts. For example, in step S611, the DCT coefficients “DCT1” to “DCT64” in each 8 × 8 pixel block are set as block representative values, and an average value over all 8 × 8 pixel blocks in the processing target image is calculated for each block representative value. By calculating the feature amount, it is possible to obtain a feature vector composed of a maximum 64 dimensional feature amount. Thereby, organ determination can be performed based on the feature vector reflecting all frequency components in the DCT coefficient, and organ determination can be performed with higher accuracy. However, since the processing time for derivation of the feature vector increases by increasing the number of dimensions, it is preferable to appropriately set the number of dimensions according to the required determination accuracy.

  So far, the best mode for carrying out the present invention has been described as the embodiment including the first and second modified examples. However, the present invention is not limited to the above-described embodiment, and does not depart from the spirit of the present invention. If so, various modifications are possible.

  For example, in the above-described embodiment, the image processing control unit 6a sets the parameter value of the abnormality detection parameter based on the organ determined by the organ determination unit 4a and the imaging distance estimated by the imaging distance estimation unit 4b. However, the parameter value can also be set based on one of the organ determined by the organ determination unit 4a and the imaging distance estimated by the imaging distance estimation unit 4b.

  In that case, the abnormality detection parameter storage unit 3d can be provided with a storage table in which parameter values are stored in advance according to either the organ or the imaging distance. Specifically, when the image processing control unit 6a sets a parameter value based on an organ, the abnormality detection parameter storage unit 3d stores a storage table that stores the parameter value for each organ, for example, as illustrated in FIG. Can be provided. Further, when the image processing control unit 6a sets the parameter value based on the imaging distance, for example, as shown in FIG. 17-2, a storage table that stores the parameter value according to the imaging distance can be provided.

  Further, when the image processing control unit 6a sets a parameter value based on either the organ or the imaging distance, the image processing unit 4 selects either the organ determination unit 4a or the imaging distance estimation unit 4b. The storage unit 3 can include only one of the organ determination reference data storage unit 3b and the distance estimation reference data storage unit 3c. In this case, in the image processing procedure shown in FIG. 2, either the organ determination process in step S102 or the imaging distance estimation process in step S103 can be omitted.

  In the organ determination process in the above-described embodiment, the organ determination unit 4a has been described as performing the organ determination for all observation images in a series of observation images. However, for example, up to a predetermined number of images or image numbers. It is also possible to perform organ determination only for the observed images. Alternatively, a desired organ can be designated and an observation image in which the organ is imaged can be processed. As a result, the organ determination process can be performed more rapidly only for the observation image obtained by imaging the desired organ.

  Similarly, in the imaging distance estimation processing, parameter setting processing, and abnormal region detection processing, not only all observation images but also predetermined observation images can be processed, thereby detecting abnormal regions. It is possible to further reduce the overall processing time required for the process.

  In the above-described embodiment, the series of observation images processed by the image processing apparatus 1 is obtained by sequentially imaging the inside of the esophagus, stomach, small intestine, and large intestine. However, the esophagus or stomach, small intestine, and large intestine are captured. The present invention can also be applied to an image group in which at least one of the interiors is individually captured.

It is a block diagram which shows the principal part structure of the image processing apparatus concerning embodiment of this invention. It is a flowchart which shows the image processing procedure which an image processing apparatus performs. It is a flowchart which shows an organ determination processing procedure. It is a figure explaining organ criteria data. It is a figure which shows the rebound brightness | luminance characteristic with respect to imaging distance. It is a flowchart which shows an imaging distance estimation procedure. It is a figure which shows the parameter value which the abnormality detection parameter memory | storage part memorize | stored. 10 is a flowchart showing an organ determination processing procedure according to Modification 1; It is a figure which shows the organ determination reference | standard data concerning the modification 1. 10 is a flowchart showing an organ determination processing procedure according to Modification 2. 10 is a flowchart showing an organ determination processing procedure according to Modification 3. It is a figure which shows the file size of a series of observation images. It is a figure which shows the moving average of the file size of a series of observation images. It is a figure which shows the moving average of the file size change amount of a series of observation images. 14 is a flowchart illustrating an organ determination processing procedure according to Modification Example 4. It is a figure which shows the DCT coefficient in an 8x8 pixel block. 16 is a flowchart showing an organ determination processing procedure according to Modification 5. It is a figure explaining organ criteria data. It is a figure which shows the parameter value which the abnormality detection parameter memorize | stored. It is a figure which shows the parameter value which the abnormality detection parameter memorize | stored.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image processing apparatus 2 Input part 3 Memory | storage part 3a Observation image memory | storage part 3b Organ judgment reference | standard data memory | storage part 3c Distance estimation reference | standard data memory | storage part 3d Abnormality detection parameter memory | storage part 4 Image processing part 4a Organ judgment part 4b Imaging distance estimation part 4c Abnormality Area detection unit 5 Output unit 6 Control unit 6a Image processing control unit

Claims (8)

In an image processing apparatus for detecting a characteristic image region that is a unique region from a series of observation images captured at different imaging distances,
Distance estimation means for estimating an imaging distance when a processing target image of the series of observation images is captured;
A region detecting means for detecting the feature image region from within the processing target image using a processing parameter for detecting the feature image region;
A setting control unit that sets a parameter value corresponding to the estimation result of the distance estimation unit as the processing parameter, and causes the region detection unit to detect the feature image region using the processing parameter;
An image processing apparatus comprising: Parameter storage means for storing in advance a parameter value corresponding to the imaging distance;
The setting control means sets a parameter value corresponding to an imaging distance estimated by the distance estimation means among the parameter values stored in the parameter storage means as the predetermined processing parameter. The image processing apparatus described.   The image processing apparatus according to claim 1, wherein the parameter value indicates a value that defines a feature of the feature image region or a processing unit pixel block size of the processing target image by the region detection unit.   The image processing apparatus according to claim 3, wherein the value defining the feature is a region size or chromaticity.   The distance estimation unit estimates an imaging distance when the processing target image is captured based on at least one of luminance information and gain information of the processing target image. The image processing apparatus according to any one of the above.   The distance estimation unit sequentially switches the processing target images in the series of observation images, and estimates the imaging distance for each of the switched processing target images. An image processing apparatus according to 1. The series of observation images is an image group in which at least one organ of the esophagus, stomach, small intestine or large intestine is imaged,
The image processing apparatus according to claim 1, wherein the region detection unit detects an abnormal part in the organ. In an image processing device for detecting a characteristic image region that is a unique region from a series of observation images captured at different imaging distances,
A distance estimation procedure for estimating an imaging distance when a processing target image in the series of observation images is captured;
A region detection procedure for detecting the feature image region from within the processing target image using a processing parameter for detecting the feature image region;
A setting control procedure for setting a parameter value corresponding to the estimation result of the distance estimation unit as the processing parameter, and causing the region detection unit to detect the feature image region using the processing parameter;
An image processing program for executing

Images