Browse Prior Art Database

Method and System for Categorizing Images

IP.com Disclosure Number: IPCOM000236541D
Publication Date: 2014-May-02
Document File: 2 page(s) / 30K

Publishing Venue

The IP.com Prior Art Database

Related People

Chang-Chih Lin: INVENTOR [+3]

Abstract

A method and system for categorizing images based on similar characteristics of the images is disclosed. Features of the images are detected by extracting several local patches or regions. Further, the method and system also uses Diverse Density Support Vector Machine (DD-SVM) to identify new incoming images. Vectors of new images are analyzed to define the category of the images based on codebooks.

This text was extracted from a Microsoft Word document.
This is the abbreviated version, containing approximately 53% of the total text.

Method and System for Categorizing Images

Abstract

A method and system for categorizing images based on similar characteristics of the images is disclosed.  Features of the images are detected by extracting several local patches or regions.  Further, the method and system also uses Diverse Density Support Vector Machine (DD-SVM) to identify new incoming images.  Vectors of new images are analyzed to define the category of the images based on codebooks.

Description

Disclosed is a method and system for categorizing images to segment images into different categories based on similar characteristics of the images.  Initially, several images are inserted manually in various categories.  For example, the categories can be bags, hats, and shirts.  Features of the images are detected by extracting several local patches or regions.  By using scale invariant features transform (SIFT), every patch is transformed into a number of vectors so that every image becomes a collection of vectors.  The vector represented patches are then converted to codewords that produces a codebook by performing K-means clustering over all vectors.  Every codebook represents one category.  In any given category, images are strongly related.  The method and system can be also applied to use cases driven by the category.

The method and system uses Diverse Density Support Vector Machine (DD-SVM) to identify new incoming images.  The vectors of new images are analyzed to define the category of the images based on codebooks.  The accuracy of DD-SVM depends on the category of images.  For example, the highest accuracy is 99.7% for dinosaurs at category 4 and the lowest accuracy is 67.7% for category...