Neural network device for retrieving image and operating method thereof
AssigneesSamsung Electronics Co., Ltd., Seoul National University R&Db Foundation
InventorsYoungkyun Jang, Namik Cho
Life of this patent
Estimating the term…
Abstract
A neural network device comprises a processor that performs an operation for training a neural network, a feature extraction module that extracts unlabeled feature vectors that correspond to unlabeled images and labeled feature vectors that correspond to labeled images, and a classifier that classifies classes of query images, wherein the processor performs first learning with respect to a plurality of codebooks by using the labeled feature vectors, and performs second learning with respect to the plurality of codebooks by optimizing an entropy based on all of the labeled feature vectors and the unlabeled feature vectors.
Claims19
A neural network device, comprising:
- a processor that performs an operation of training a neural network;
- a feature extraction module that extracts unlabeled feature vectors that corresponds to unlabeled images and labeled feature vectors that correspond to labeled images;
- and
- wherein the processor performs first learning with respect to a plurality of codebooks by using the labeled feature vectors but not the unlabeled feature vectors, and performs second learning with respect to the plurality of codebooks by optimizing an entropy based on all of the labeled feature vectors and the unlabeled feature vectors.
A method of operating a neural network, the method comprising:
- extracting unlabeled feature vectors that correspond to unlabeled images and labeled feature vectors that correspond to labeled images;
- performing first learning with respect to a plurality of codebooks by using the labeled feature vectors but not the unlabeled feature vectors;
- and
- performing second learning with respect to the plurality of codebooks by optimizing an entropy based on all of the labeled feature vectors and the unlabeled feature vectors.
A method of operating a neural network, the method comprising:
- receiving a plurality of biometric images that include at least one labeled image and at least one unlabeled image;
- extracting feature vectors that correspond to the plurality of biometric images;
- training a plurality of codebooks by using the feature vectors;
- receiving a query image and calculating distance values between the query image and the plurality of biometric images;
- estimating, based on the calculated distance values, expected values for classes that classify the plurality of biometric images;
- passing a biometric authentication when a greatest value of the expected values exceeds a preset threshold value, and
- retraining the plurality of codebooks by using the query image, in response to passing the biometric authentication.
Dependencies are inferred from the claim text.