Entity recognition from an image
Aspects of the current disclosure include systems and methods for identifying an entity in a query image by comparing the query image with digital images in a database. In one or more embodiments, a query feature may be extracted from the query image and a set of candidate features may be extracted from a set of images in the database. In one or more embodiments, the distances between the query feature and the candidate features are calculated. A feature, which includes a set of shortest distances among the calculated distances and a distribution of the set of shortest distances, may be generated. In one or more embodiments, the feature is input to a trained model to determine whether the entity in the query image is the same entity associated with one of the set of shortest distances.
1. A method comprising:
generating a query feature vector based at least in part on a query image;
determining a set of distances that comprises a respective distance between the query feature vector and each respective candidate feature vector of a set of candidate feature vectors;
generating a decision tree vector that comprises a selected subset of the determined set of distances;
inputting the decision tree vector to a trained decision tree to obtain a probability value; and
determining, in response to the probability value satisfying a threshold value, that an entity associated with the query image is a same entity that is associated with the respective candidate feature vector that produced a shortest distance of the selected subset of the determined set of distances included in the decision tree vector.
2. The method of claim 1 , further comprising:
collecting a set of profile images;
generating a set of original feature vectors based on the set of profile images; and
training a principal component analysis (PCA) on the set of original feature vectors.
3. The method of claim 2 , further comprising:
training a model using a set of features that correspond to a set of candidate images associated with entities having known identities.
4. The method of claim 2 , further comprising:
selecting a value of a dimension count;
generating a set of test vector pairs by transforming the set of original feature vectors into corresponding PCA feature vectors comprising a set of top PCA dimensions; and
calculating relative errors of distance between each test vector pairs.
5. The method of claim 4 , further comprising:
determining that the relative errors are within an acceptable range;
setting a particular value as a smallest amount of PCA dimensions that produced an acceptable range; and
discretizing each dimension of the PCA feature vectors and storing each dimension as a byte.
6. The method of claim 1 , further comprising:
generating a query feature for a query image, the query feature including one or more features about the entity extracted from the query image; and
determining a set of similarity measures between the query feature and each of a set of candidate features that correspond to a set of candidate images.
7. The method of claim 6 , further comprising:
reducing, prior to determining the set of similarity measures, a search space comprising a plurality of features to obtain the set of candidate features.
8. The method of claim 7 , further comprising:
using an index parameter associated with the query image, selecting the set of candidate features among the plurality of features.
9. The method of claim 7 , further comprising:
using one or more k-d trees, selecting the set of candidate images among the plurality of features.
10. The method of claim 1 , further comprising:
generating an original feature that includes one or more features extracted from the query image; and
converting the original feature into a query feature by reducing dimensionality of the original feature; and
inputting the original feature to a principal component analysis map to obtain the query feature.
11. A system comprising:
one or more processors; and
at least one memory including instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
generating a query feature vector based at least in part on a query image;
determining a set of distances that comprises a respective distance between the query feature vector and each respective candidate feature vector of a set of candidate feature vectors;
generating a decision tree vector that comprises a selected subset of the determined set of distances and distribution of the set of short distances;
inputting the decision tree vector to a trained decision tree to obtain a probability value; and
determining, in response to the probability value satisfying a threshold value, that an entity associated with the query image is a same entity that is associated with the respective candidate feature vector that produced a shortest distance of the selected subset of the determined set of distances included in the decision tree vector.
12. The system of claim 11 , wherein the operations further comprise:
collecting a set of profile images;
generating a set of original feature vectors based on the set of profile images; and
training a principal component analysis (PCA) on the set of original feature vectors.
13. The system of claim 12 , wherein the operations further comprise:
training a model using a set of features that correspond to a set of candidate images associated with entities having known identities.
14. The system of claim 12 , wherein the operations further comprise:
selecting a value of a dimension count;
generating a set of test vector pairs by transforming the set of original feature vectors into corresponding PCA feature vectors comprising a set of top PCA dimensions; and
calculating relative errors of distance between each test vector pairs.
15. The system of claim 14 , wherein the operations further comprise:
determining that the relative errors are within an acceptable range;
setting a particular value as a smallest amount of PCA dimensions that produced an acceptable range; and
discretizing each dimension of the PCA feature vectors and storing each dimension as a byte.
16. The system of claim 11 , wherein the operations further comprise:
generating a query feature for a query image, the query feature including one or more features about the entity extracted from the query image; and
determining a set of similarity measures between the query feature and each of a set of candidate features that correspond to a set of candidate images.
17. The system of claim 16 , wherein the operations further comprise:
reducing, prior to determining the set of similarity measures, a search space comprising a plurality of features to obtain the set of candidate features.
18. The system of claim 17 , wherein the operations further comprise:
using an index parameter associated with the query image, selecting the set of candidate features among the plurality of features.
19. The system of claim 11 , wherein the operations further comprise:
generating an original feature that includes one or more features extracted from the query image; and
converting the original feature into a query feature by reducing dimensionality of the original feature; and
inputting the original feature to a principal component analysis map to obtain the query feature.
20. A non-transitory computer-readable medium or media comprising one or more sets of instructions, which causes one or more processors to perform operations comprising:
generating a query feature vector based at least in part on a query image;
determining a set of distances that comprises a respective distance between the query feature vector and each respective candidate feature vector of a set of candidate feature vectors;
generating a decision tree vector that comprises a selected subset of the determined set of distances;
inputting the decision tree vector to a trained decision tree to obtain a probability value; and
determining, in response to the probability value satisfying a threshold value, that an entity associated with the query image is a same entity that is associated with the respective candidate feature vector that produced a shortest distance of the selected subset of the determined set of distances included in the decision tree vector.