Image analyzing apparatus, image analyzing method, and recording medium

1. An image analyzing apparatus, comprising:

a memory that stores a plurality of instructions; and

processing circuitry that executes the plurality of instructions to:

reproject an input image in a plurality of different directions to divide the input image into a plurality of partial images;

extract a feature amount from each of the partial images;

calculate a degree of importance of each pixel of the input image based on the extracted feature amount and in accordance with a predetermined regression model;

calculate a likelihood distribution of an attention point from the calculated degree of importance of each pixel of the input image in accordance with the predetermined regression model, and

calculate the attention point from the likelihood distribution.

2. The image analyzing apparatus according to claim 1 , wherein the processing circuitry calculates, as the attention point, a position of a pixel corresponding to one of a maximum likelihood value, an average value, and a local maximum value of the likelihood distribution of the attention point.

3. The image analyzing apparatus according to claim 1 , wherein the processing circuitry further executes the plurality of instructions to add the degree of importance to calculate the likelihood distribution of the attention point.

4. The image analyzing apparatus according to claim 1 , wherein the processing circuitry is further configured to train a neural network to

calculate a likelihood distribution of an attention point from the calculated degree of importance of each pixel of the input image in accordance with the predetermined regression model, and

calculate the attention point from the likelihood distribution.

5. The image analyzing apparatus according to claim 1 , wherein the input image is of an equirectangular projection format.

6. The image analyzing apparatus according to claim 5 , wherein the input image includes pixel values plotted on a unit sphere.

7. The image analyzing apparatus according to claim 1 , wherein the input image is an omnidirectional image.

8. The image analyzing apparatus according to claim 1 , wherein the processing circuitry extracts the feature amount from each of the partial images to minimize distortion.

9. The image analyzing apparatus according to claim 1 , wherein the degree of importance of each pixel of the input image is training data for training a neural network.

10. A method for extracting a degree of importance of an input image by position, the method comprising:

reprojecting the input image in a plurality of different directions to divide the input image into a plurality of partial images;

extracting, by processing circuitry of an image analyzing apparatus, a feature amount from each of the partial images;

calculating, by the processing circuitry, a degree of importance of each pixel of the input image based on the extracted feature amount and in accordance with a predetermined regression model;

calculating a likelihood distribution of an attention point from the calculated degree of importance of each pixel in the input image in accordance with the predetermined regression model; and

calculating an attention point from the likelihood distribution.

11. The method according to claim 10 , wherein the calculating the attention point includes calculating a position of a pixel corresponding to one of a maximum likelihood value, an average value, and a local maximum value of the likelihood distribution of the attention point as the attention point.

12. The method according to claim 10 , wherein the calculating the likelihood distribution of the attention point includes adding the degree of importance.

13. An image analyzing apparatus, comprising:

processing circuitry configured to:

reproject an input image in a plurality of different directions to divide the input image into a plurality of partial images;

extract a feature amount from each of the partial images; and

calculate a degree of importance of each pixel of the input image based on the extracted feature amount and in accordance with a predetermined regression model; and

a neural network is configured to

calculate a likelihood distribution of an attention point from the calculated degree of importance of each pixel of the input image in accordance with the predetermined regression model; and

calculate the attention point from the likelihood distribution.

14. The image analyzing apparatus according to claim 13 , wherein the neural network calculates, as the attention point, a position of a pixel corresponding to one of a maximum likelihood value, an average value, and a local maximum value of the likelihood distribution of the attention point.

15. The image analyzing apparatus according to claim 13 , wherein the neural network is further configured to add the degree of importance to calculate the likelihood distribution of the attention point.