Endoscope insertion direction detecting device and endoscope insertion direction detecting method

1. An endoscope insertion direction detecting device comprising:

a scene feature value computing section for computing a specific feature value on a scene of an endoscope image of a moving image picked up by an endoscope inserted into a body cavity;

a classification section for computing a degree of similarity between the scene of the endoscope image and a plurality of training data as references for a plurality of classes, which are respectively defined for different structures in the body cavity, from the specific feature value and the plurality of training data, and classifying the scene of the endoscope image into one or more of the plurality of classes based on the degree of similarity computed; and

an insertion direction computing section for computing an insertion direction of the endoscope corresponding to the scene of the endoscope image, with respect to the one or more of the classes into which the scene of the endoscope image is classified,

wherein:

the classification section performs classification into classes of lumen dark sections, folds, and others as a plurality of specific feature values, and

if the scene of the endoscope image is classified into the class of lumen dark sections, the insertion direction computing section computes a region of a lumen dark section based on red image components in the endoscope image, if the scene of the endoscope image is classified into the class of folds, the insertion direction computing section computes a barycentric position of a fold based on the red image components in the endoscope image, and if the scene of the endoscope image is classified into the class of others, the insertion direction computing section computes a density gradient in the endoscope image.

2. The endoscope insertion direction detecting device according to claim 1 , wherein if the scene of the endoscope image is classified into more than one of the classes, the insertion direction computing section computes an insertion direction of the endoscope only for an essential class among the more than one of the classes into which the scene of the endoscope image is classified.

3. The endoscope insertion direction detecting device according to claim 1 , wherein the scene feature value computing section computes an autocorrelation coefficient of the scene of the endoscope image using a plurality of reference patterns.

4. The endoscope insertion direction detecting device according to claim 1 , wherein the scene feature value computing section computes a histogram in the scene of the endoscope image.

5. The endoscope insertion direction detecting device according to claim 1 , wherein the scene feature value computing section computes the specific feature value based on images of small regions into which the scene of the endoscope image is divided.

6. The endoscope insertion direction detecting device according to claim 1 , wherein the scene feature value computing section computes the specific feature value based on a converted image obtained by converting a resolution of the scene of the endoscope image.

7. An endoscope insertion direction detecting method comprising:

a scene feature value computing step of computing a specific feature value on a scene of an endoscope image of a moving image picked up by an endoscope inserted into a body cavity;

a classification step of computing a degree of similarity between the scene of the endoscope image and a plurality of training data as references for a plurality of classes, which are respectively defined for different structures in the body cavity, from the specific feature value and the plurality of training data, and classifying the scene of the endoscope image into one or more of the plurality of classes based on the degree of similarity computed; and

an insertion direction computing step of computing an insertion direction of the endoscope corresponding to the scene of the endoscope image, with respect to the one or more of the classes into which the scene of the endoscope is classified,

wherein the classification step performs classification into classes of lumen dark sections, folds, and others as a plurality of feature values, and if the scene of the endoscope image is classified into the class of lumen dark sections, the insertion direction computing step computes a region of a lumen dark section based on red image components in the endoscope image, if the scene of the endoscope image is classified into the class of folds, the insertion direction computing step computes a barycentric position of a fold based on the red image components in the endoscope image, and if the scene of the endoscope image is classified into the class of others, the insertion direction computing step computes a density gradient in the endoscope image.

8. The endoscope insertion direction detecting method according to claim 7 , further comprising an insertion direction display step of making an output for displaying the insertion direction of the endoscope in accordance with the degree of similarity to a display step of displaying the insertion direction of the endoscope in the scene of the endoscope image.

9. The endoscope insertion direction detecting device according to claim 1 , wherein if the scene of the endoscope image is classified into more than one of the classes, the insertion direction computing section computes a plurality of insertion directions of the endoscope for more than one of the classes into which the scene of the endoscope is classified.

10. The endoscope insertion direction detecting device according to claim 1 , further comprising an insertion direction display output section for making an output for displaying the insertion direction of the endoscope in accordance with the degree of similarity to a display section for displaying the insertion direction of the endo scope in the scene of the endoscope image.

11. An endoscope insertion direction detecting device comprising:

a scene feature value computing section for computing a specific feature value on a scene of an endoscope image of a moving image picked up by an endoscope inserted into a body cavity;

a classification section for computing a degree of similarity between the scene of the endoscope image and a plurality of training data as references for a plurality of classes, which are respectively defined for different structures in the body cavity, from the specific feature value and the plurality of training data, and classifying the scene of the endoscope image into one or more of the plurality of classes based on the degree of similarity computed;

an insertion direction computing section for computing an insertion direction of the endoscope corresponding to the scene of the endoscope image, with respect to the one or more of the classes into which the scene of the endoscope image is classified, and

an insertion direction display output section for making an output for displaying the insertion direction of the endoscope in accordance with the degree of similarity to a display section for displaying the insertion direction of the endoscope in the scene of the endoscope image,

wherein:

if the scene of the endoscope image is classified into more than one of the classes, the insertion direction computing section computes a plurality of insertion directions of the endoscope for more than one of the classes into which the scene of the endoscope is classified,

the classification section performs classification into classes of lumen dark sections, folds, and others as the plurality of specific feature values, and

if the scene of the endoscope image is classified into the class of lumen dark sections, the insertion direction computing section computes density gradients with respect to small regions into which the scene of the endoscope image is divided based on gradients or differences of gray values in adjacent small regions, and designates a small region toward which density gradients of all adjacent small regions point intensively as a candidate for a lumen dark section.

12. The endoscope insertion direction detecting method according to claim 7 , wherein if the scene of the endoscope image is classified into more than one of the classes, the insertion direction computing step computes a plurality of insertion directions of the endoscope for more than one of the classes into which the scene of the endoscope is classified.

13. The endoscope insertion direction detecting method according to claim 7 , wherein if the scene of the endoscope image is classified into more than one of the classes, the insertion direction computing step computes an insertion direction of the endoscope with respect to only an essential class among more than one of the classes into which the scene of the endoscope image is classified.

14. The endoscope insertion direction detecting method according to claim 7 wherein if the scene of the endoscope image is classified into the class of lumen dark sections, the insertion direction computing step computes density gradients with respect to small regions into which the scene of the endoscope image is divided based on gradients or differences of gray values in adjacent small regions, and designates a small region toward which density gradients of all adjacent small regions point intensively, as a candidate for a lumen dark section.