Image classification method, device, and program
In classifying images by machine learning, provided are an image classification method, device, and program for classifying the image from which the feature difference is hardly detected, in particular, classifying the interference fringe image of tear fluid layer by the dry eye types. The method includes a step of acquiring a feature value from an interference fringe image of tear fluid layer for learning, a step of constructing a model for classifying an image from the feature value acquired from the interference fringe image of tear fluid layer for learning, a step of acquiring the feature value from an interference fringe image of tear fluid layer for testing, and a step of performing classification processing for classifying the interference fringe image of tear fluid layer for testing by types of dry eye using the model and the feature value acquired from the interference fringe image of tear fluid layer.
1. An image classification method comprising:
a step of acquiring a feature value from an interference fringe image of tear fluid layer for learning;
a step of constructing a model for classifying an image from the feature value acquired from the interference fringe image of tear fluid layer for learning;
a step of acquiring the feature value from an interference fringe image of tear fluid layer for testing; and
a step of performing classification processing for classifying the interference fringe image of tear fluid layer for testing by types of dry eye using the model and the feature value acquired from the interference fringe image of tear fluid layer for testing.
2. The image classification method according to claim 1 , wherein the feature value is information relating to a color of the image and includes at least one piece of information relating to a color, color diversity, color brightness, and local color variation.
3. The image classification method according to claim 2 , wherein the information relating to the color is information based on a luminance average of at least one of red and blue luminance images.
4. The image classification method according to claim 2 , wherein the information relating to the color diversity is information based on a standard deviation of a luminance of at least one of red and blue luminance images.
5. The image classification method according to claim 2 , wherein the information relating to the color brightness is information based on a difference value of luminances of pixels.
6. The image classification method according to claim 2 , wherein the local color variation is information based on a standard deviation calculated from a luminance of a pixel of interest and a luminance of a pixel adjacent thereto.
7. The image classification method according to claim 1 , wherein in a case where the interference fringe image of tear fluid layer for testing is a moving image, a plurality of predetermined frames are subjected to the classification processing and a most frequent result is regarded as a classification result of the image for testing.
8. The image classification method according to claim 1 , wherein the feature value is acquired from a predetermined region of the interference fringe image of tear fluid layer for learning and the interference fringe image of tear fluid layer for testing.
9. The image classification method according to claim 1 , wherein the interference fringe image of tear fluid layer for testing is a moving image within a predetermined period of time subsequent to eyelid-opening detection or one or more still images acquired at an optional timing within a predetermined period of time subsequent to eyelid-opening detection.
10. The image classification method according to claim 9 , wherein the predetermined period of time is 5 seconds.
11. An image classification device configured to execute the method according to claim 1 .
12. An image classification program stored in a non-transitory computer readable medium for making a computer to execute the method according to claim 1 .