Learning apparatus, density measuring apparatus, learning method, computer program product, and density measuring system

1. A learning apparatus, comprising:

a processor; and

a memory that stores executable instructions that, when executed by the processor, facilitate performance of operations, comprising:

first extracting unit a plurality of first partial images from a learning image;

first calculating a feature amount of the first partial image;

retrieving objects included in the first partial image, and gives, as a label, a vector to the feature amount, the vector representing relative positions between a first position in the first partial image and each of all objects included in the first partial image;

calculating, for each of the plurality of first partial images, a histogram representing a distribution of the relative positions and votes the histogram into a parameter space so as to generate a voting histogram;

dividing the feature amount corresponding to each of the plurality of first partial images into a plurality of clusters to reduce variation of the corresponding voting histogram, so as to learn a regression model representing a relationship between the feature amount of the first partial image and the relative position of the object included in the first partial image; and

first predicting, for each of the plurality of clusters, a representative label from the label given to the feature amount that belongs to the cluster.

2. The apparatus according to claim 1 , wherein

the regression model is one or a plurality of random trees,

the dividing determines a division index of each node from a root node to a plurality of leaf nodes and determines the feature amount that belongs to each of the plurality of leaf nodes to reduce variation of the voting histogram, so as to learn the regression model, and

the first predicting further comprises predicting, for each of the plurality of clusters as the plurality of leaf nodes, the representative label from the label given to the feature amount that belongs to the cluster.

3. The apparatus according to claim 2 , wherein

the first predicting further comprises

calculating, for each of the plurality of clusters, an average histogram of the voting histograms corresponding to the labels given to the respective plurality of feature amounts that belong to the cluster, and

predicting, for each of the plurality of clusters, a label corresponding to a voting histogram close to the average histogram, as the representative label, among the labels given to the respective plurality of feature amounts that belong to the cluster.

4. The apparatus according to claim 1 , wherein

the regression model is a nearest neighbor classifier.

5. A density measuring apparatus, comprising:

a processor; and

a memory that stores executable instructions that, when executed by the processor, facilitate performance of operations, comprising:

second extracting a plurality of second partial images from an input image;

second calculating a feature amount of each of the plurality of second partial images;

acquiring the regression model and the representative label from the learning apparatus according to claim 1 ;

second predicting that assigns the feature amount of the second partial image to a variable of the regression model so as to predict the representative label corresponding to each of the plurality of second partial images; and

density calculating an average density of objects included in the input image based on the relative positions represented by the representative labels corresponding to the respective plurality of second partial images.

6. The apparatus according to claim 5 , wherein

the density calculating comprising:

third calculating a density distribution of objects for each of the plurality of second partial images based on the relative positions represented by the representative labels corresponding to the respective plurality of second partial images;

fourth calculating that arranges, at a position corresponding to each of the plurality of second partial images in the input image, the density distribution of the corresponding second partial image; and

fifth calculating, for each pixel constituting the input image, a first average value of densities of objects, according to a frequency of overlap of the density distributions in the input image, and calculating an average value of the first average values of the pixels constituting the input image as the average density of objects included in the input image.

7. A learning method, comprising:

extracting a plurality of first partial images from a learning image;

calculating a feature amount of the first partial image;

retrieving objects included in the first partial image to give, as a label, a vector to the feature amount, the vector representing relative positions between a first position in the first partial image and each of all objects included in the first partial image;

calculating, for each of the plurality of first partial images, a histogram representing a distribution of the relative positions to vote the histogram into a parameter space so as to generate a voting histogram;

dividing the feature amount corresponding to each of the plurality of first partial images into a plurality of clusters to reduce variation of the corresponding voting histogram, so as to learn a regression model representing a relationship between the feature amount of the first partial image and the relative position of the object included in the first partial image; and

predicting, for each of the plurality of clusters, a representative label from the label given to the feature amount that belongs to the cluster.

8. A computer-readable storage medium, comprising executable instructions that, when executed by a processor, facilitate performance of operations, comprising:

extracting a plurality of first partial images from a learning image;

calculating a feature amount of the first partial image;

retrieving objects included in the first partial image to give, as a label, a vector to the feature amount, the vector representing relative positions between a first position in the first partial image and each of all objects included in the first partial image;

calculating, for each of the plurality of first partial images, a histogram representing a distribution of the relative positions to vote the histogram into a parameter space so as to generate a voting histogram;

dividing the feature amount corresponding to each of the plurality of first partial images into a plurality of clusters to reduce variation of the corresponding voting histogram, so as to learn a regression model representing a relationship between the feature amount of the first partial image and the relative position of the object included in the first partial image; and

predicting, for each of the plurality of clusters, a representative label from the label given to the feature amount that belongs to the cluster.

9. A density measuring system, comprising a learning apparatus and a density measuring apparatus connected to the learning apparatus, wherein

the learning apparatus includes:

a first extracting unit that extracts a plurality of first partial images from a learning image;

a first calculator that calculates a feature amount of the first partial image;

a retrieving unit that retrieves objects included in the first partial image, and gives, as a label, a vector to the feature amount, the vector representing relative positions between a first position in the first partial image and each of all objects included in the first partial image;

a voting unit that calculates, for each of the plurality of first partial images, a histogram representing a distribution of the relative positions and votes the histogram into a parameter space so as to generate a voting histogram;

a learning unit that divides the feature amount corresponding to each of the plurality of first partial images into a plurality of clusters to reduce variation of the corresponding voting histogram, so as to learn a regression model representing a relationship between the feature amount of the first partial image and the relative position of the object included in the first partial image; and

a first predicting unit that predicts, for each of the plurality of clusters, a representative label from the label given to the feature amount that belongs to the cluster, and

the density measuring apparatus includes:

a second extracting unit that extracts a plurality of second partial images from an input image;

a second calculator that calculates a feature amount of each of the plurality of second partial images;

an acquiring unit that acquires the regression model and the representative label from the learning apparatus according to claim 1 ;

a second predicting unit that assigns the feature amount of the second partial image to a variable of the regression model so as to predict the representative label corresponding to each of the plurality of second partial images; and

a density calculator that calculates an average density of objects included in the input image based on the relative positions represented by the representative labels corresponding to the respective plurality of second partial images.