Browse Prior Art Database

An Efficient Method to Fuse Many Features for Visual Classification

IP.com Disclosure Number: IPCOM000244838D
Publication Date: 2016-Jan-21
Document File: 1 page(s) / 43K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method of efficient nonlinear feature fusion for image recognition using many features.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 100% of the total text.

Page 01 of 1

An Efficient Method to Fuse Many Features for Visual Classification

Traditional feature fusion methods can be grouped into two categories . The first is linear early fusion , done by concatenating all the features into a linear model . The second category is nonlinear late fusion , which trains a nonlinear model for each feature, and then fuses the scores of these models . Linear early fusion is limited to linear models while nonlinear late fusion is computationally expensive .

Previous work disclosed an efficient method for nonlinear late fusion ; however, the method is limited to a small number of features . A need remains to provide a nonlinear fusion method for large number of features .

The novel contribution is a method of efficient nonlinear feature fusion for image recognition. The method includes analyzing general object recognition:

• How to fuse heterogeneous features
- Where the subjects are medical images
- Where the subjects are videos


• How to efficiently fuse features even for many features

To implement the solution:


1. Train a first layer nonlinear classifier for each feature using kernel approximation


2. Evaluate the classifiers on the whole training set, and get scores


3. Train a second layer classification model based on multiple scores

1