Object identification image device, method, and computer program product

1. An image analysis device, comprising one or more processors configured to:

receive an image;

calculate feature amount information indicating a feature of a region of the image;

recognize a known object from the image based at least in part on the feature amount information, an identity of the known object included in learning data used as part of an image recognition training process;

recognize an unknown object having no class or no category as a generalization object from the image based at least in part on the feature amount information and a plurality of features of a plurality of known objects included in the learning data, the plurality of features being combined to generate the generalization object;

integrate known-object data and generalization-object data into integrated data that is a single image data piece in which the known-object data and the generalization-object data are superimposed, the known-object data comprising the known object recognized by the known-object recognizer, the generalization-object data comprising the generalization object recognized by the generalization-object recognizer;

determine the known object when a position of the known object and a position of the generalization object in the integrated data match each other, and determine the unknown object as the generalization object when the position of the known object and the position of the generalization object in the integrated data do not match each other; and

output object information of an object identified from the image as the known object or the unknown object.

2. The device according to claim 1 , wherein the one or more processors:

integrate the known-object data and the generalization-object data of the integrated data into regions of interest, and

determine whether the position of the known object and the position of the generalization object match each other in each region of interest of the regions of interest.

3. The device according to claim 1 , wherein the one or more processors recognize the generalization object by using a learning model, the learning model is configured to be used for recognizing the generalization object as a single object category.

4. The device according to claim 1 , wherein the one or more processors identify the known object or the generalization object by a bounding box surrounding the known object or the generalization object.

5. The device according to claim 1 , wherein the one or more processors identify the known object or the generalization object by a mask on an area of the known object or the generalization object.

6. The device according to claim 1 , wherein

the object information comprises at least one of a number of known objects, a number of generalization objects, a position of the known object, a position of the generalization object, a bounding box surrounding the known object, a bounding box surrounding the generalization object, a mask on an area of the known object, or a mask on an area of the generalization object.

7. An image analysis method, comprising:

receiving an input image;

calculating feature amount information indicating a feature of a region of the image;

recognizing a known object from the image based at least in part on the feature amount information, an identity of the known object included in learning data used as part of an image recognition training process;

recognizing an unknown object having no class or no category as a generalization object from the image based at least in part on the feature amount information and a plurality of features of a plurality of known objects included in the learning data;

combining the plurality of features to generate the generalization object;

integrating known-object data and generalization-object data into integrated data that is a single image data piece in which the known-object data and the generalization-object data are superimposed, the known-object data comprising the known object recognized by the known-object recognizer, the generalization-object data comprising the generalization object recognized by the generalization-object recognizer;

determining the known object when a position of the known object and a position of the generalization object in the integrated data match each other, and determine the unknown object as the generalization object when the position of the known object and the position of the generalization object in the integrated data do not match each other; and

outputting object information of an object identified from the image as the known object or the unknown object.

8. A computer program product comprising a non-transitory computer readable medium comprising programmed instructions, the instructions causing the computer to execute:

receiving an input image;

calculating feature amount information indicating a feature of a region of the image;

recognizing a known object from the image based at least in part on the feature amount information, an identity of the known object included in learning data used as part of an image recognition training process;

recognizing an unknown object having no class or no category as a generalization object from the image based at least in part on feature amount information and a plurality of features of a plurality of known objects included in the learning data, the plurality of features being combined to generate the generalization object;

integrating known-object data and generalization-object data into integrated data that is a single image data piece in which the known-object data and the generalization-object data are superimposed, the known-object data comprising the known object recognized by the known-object recognizer, the generalization-object data comprising the generalization object recognized by the generalization-object recognizer;

determining the known object when a position of the known object and a position of the generalization object in the integrated data match each other, and determine the unknown object as the generalization object when the position of the known object and the position of the generalization object in the integrated data do not match each other; and

outputting object information of an object identified from the image as the known object or the unknown object.