摘要 |
According to an embodiment, a feature point detection device includes a generator to generate a K-class classifier and perform, for T times, an operation in which a first displacement vector is obtained that approximates D number of initial feature points of each training sample classified on a class-by-class basis to true feature points; a calculator to calculate, from the first displacement vectors, second displacement label vectors each unique to one second displacement vector, and a second displacement coordinate vector common to the second displacement vectors; a classifier to apply the K-class classifiers to the input image and obtain a second displacement label vector associated with a class identifier output from each K-class classifier; an adder to add up the second displacement label vectors; and a detector to detect D number of true feature points based on the initial feature points, the added label vector, and the second displacement coordinate vector. |
主权项 |
1. A feature point detection device comprising:
a processor to execute a program stored on one or more memories to implement a generator, a K-class classifier, a calculator, an input unit, a classifier, an adder, and a detector, wherein a training data memory stores therein training data, which represents a set of N (N≧2) number of training samples each of which holding an image pattern of a sample image, D (D≧1) number of true feature points captured in the sample image, and D number of initial feature points corresponding to the D number of true feature points in association with one another; the generator performs, for T (T≧1) number of times, a generating operation in which
using the training data, the K-class classifier is generated which is expected to classify training samples having a similar image pattern at the initial feature points of the sample image in same class from among K-classes (K≧2) and which outputs a class identifier of the class in which the training samples are classified, andusing the K-class classifier, the N number of training samples are classified in the K-classes and, for each class, a first displacement vector is obtained that approximates the D number of initial feature points of each training sample classified in the class to the D number of true feature points; a classifier memory stores therein the T number of K-class classifiers; the calculator calculates, from K×T number of the first displacement vectors and in order to express second displacement vectors to which the K×T number of first displacement vectors are approximated, second displacement label vectors, each being unique to one of K×T number of the second displacement vectors, and a second displacement coordinate vector common to the K×T number of second displacement vectors; a displacement label memory stores therein, in association with each of K×T number of the second displacement label vectors, a class identifier of a class from which is obtained a first displacement vector approximated to a second displacement vector of the second displacement label vector; a displacement coordinate memory stores therein the second displacement coordinate vector; the input unit receives an input image in which D number of initial feature points are set; the classifier applies the T number of K-class classifiers to the input image and, for each K-class classifier, obtain, from the displacement label memory, a second displacement label vector associated with a class identifier output from the K-class classifier; the adder performs addition of T number of the second displacement label vectors so as to obtain an added label vector; and the detector detects D number of true feature points of the input image based on the D number of initial feature points set in the input image, based on the added label vector, and based on the second displacement coordinate vector. |