Apparatus, method and computer program for analyzing image

1. An image analysis method comprising:

receiving a query image;

extracting one or more regions of interest from the query image;

calculating one or more first features for each of the regions of interest extracted from the query image by respectively applying each region of interest to a corresponding ROI (region of interest) feature extraction model independently learned for each region of interest in order to extract the one or more first features for each of the regions of interest;

calculating one or more second features of an entire query image by applying the query image to an entire region feature extraction model learned with a plurality of entire learning images, wherein the one or more second features are calculated independently from the first features; and

calculating analysis values of the query image for identifying the query image by applying the one or more first features of each of the regions of interest and the one or more second features of the entire query image to a pre-learned integration analysis model.

2. The image analysis method of claim 1 , wherein the integration analysis model is a neural network or a regression model learned with features calculated by applying the learning images to the entire image feature extraction model and one or more features calculated by respectively applying one or more regions of interest extracted from the learning images to the one or more ROI feature extraction model.

3. The image analysis method of claim 1 , wherein the ROI feature extraction model is a deep neural network composed of one or more layers and the one or more first features are values that are output from any one of the layers or values that are input to the any one layer.

4. The image analysis method of claim 1 , wherein the ROI feature extraction model and the entire region feature extraction model are deep neural networks each composed of one or more layers and the one or more first features and the one or more second features are values that are output from any one of the layers or values that are input to the any one layer.

5. The image analysis method of claim 1 , wherein the ROI feature extraction model is learned with using one or more regions of interest classified in accordance with predetermined identification values as training data, and the identification values are the same kind of the analysis values of the query image or different kind of values having correlation with the analysis values of the query image.

6. The image analysis method of claim 1 , comprising:

calculating analysis values of regions of interest by respectively applying one or more regions of interest to the one or more ROI feature extraction models; and

providing the analysis values of the query image and the analysis values of the regions of interest to a user.

7. The image analysis method of claim 1 , wherein the extracting of regions of interest includes extracting a region of the query image that corresponds to a specific region as a region of interest by applying a learned automatic detection model to the specific region.

8. An image analysis system comprising:

an input unit receiving a query image;

a detector extracting one or more regions of interest from the query image;

an interest image analyzer calculating one or more first features for each of the regions of interest extracted from the query image by respectively applying each region of interest to a corresponding ROI (region of interest) feature extraction model independently learned for each region of interest in order to extract the one or more first features for each of the regions of interest;

an entire image analyzer calculating second features of an entire query image by applying the query image to an entire region feature extraction model learned with a plurality of entire learning images, wherein the one or more second features are calculated independently from the first features; and

an integration analyzer calculating analysis values of the query image by applying the one or more first features of each of the regions of interest and the one or more second features of the entire query image to a pre-learned integration analysis model, and identifying the query image using the analysis values.

9. An image analysis system comprising:

at least one processor; and

at least one non-transitory computer readable medium in communication with the at least one processor, the at least one non-transitory computer readable medium having stored thereon computer software comprising a set of instructions that, when executed by the at least one processor, causes the image analysis system to:

receive a query image;

extract one or more regions of interest from the query image;

calculate one or more first features for each of the regions of interest extracted from the query image by respectively applying each region of interest to a corresponding ROI (region of interest) feature extraction model independently learned for each region of interest in order to extract the one or more first features for each of the regions of interest;

calculate one or more second features of an entire query image by applying the query image to an entire region feature extraction model learned with a plurality of entire learning images, wherein the one or more second features are calculated independently from the first features; and

calculating analysis values of the query image for identifying the query image by applying the one or more first features of each of the regions of interest and the one or more second features of the entire query image to a pre-learned integration analysis model.

10. The image analysis system of claim 9 , wherein the integration analysis model is a neural network or a regression model learned with features calculated by applying the learning images to the entire image feature extraction model and one or more features calculated by respectively applying one or more regions of interest extracted from the learning images to the one or more ROI feature extraction model.

11. The image analysis system of claim 9 , wherein the ROI feature extraction model is a deep neural network composed of one or more layers and the one or more first features are values that are output from any one of the layers or values that are input to the any one layer.

12. The image analysis system of claim 9 , wherein the ROI feature extraction model and the entire region feature extraction model are deep neural networks each composed of one or more layers and the one or more first features and the one or more second features are values that are output from any one of the layers or values that are input to the any one layer.

13. The image analysis system of claim 9 , wherein the ROI feature extraction model is learned with using one or more regions of interest classified in accordance with predetermined identification values as training data, and the identification values are the same kind of the analysis values of the query image or different kind of values having correlation with the analysis values of the query image.

14. The image analysis system of claim 9 , wherein the set of instructions that, when executed by the at least one processor, further causes the image analysis system to:

calculate analysis values of regions of interest by respectively applying one or more regions of interest to the one or more ROI feature extraction models; and

provide the analysis values of the query image and the analysis values of the regions of interest to a user.

15. The image analysis system of claim 9 , wherein the extracting of regions of interest includes extracting a region of the query image that corresponds to a specific region as a region of interest by applying a learned automatic detection model to the specific region.