COMPUTER READABLE MEDIUM STORING PROGRAM, IMAGE IDENTIFICATION INFORMATION ADDING APPARATUS, AND IMAGE IDENTIFICATION INFORMATION ADDING METHOD

• US 20120039541A1
• Filed: 01/14/2011
• Published: 02/16/2012
• Est. Priority Date: 08/12/2010
• Status: Active Grant

• Current Assignee: Fuji Xerox Company Limited (Xerox Holdings Corp.)
• Original Assignee: Fuji Xerox Company Limited (Xerox Holdings Corp.)
• Inventors: FUKUI, Motofumi, Kato, Noriji, Qi, Wenyuan
• Granted Patent: US 8,538,173 B2
• Time in Patent Office: Days
• US Class Current: 382/224
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers

Abstract

A computer readable medium storing a program causing a computer to execute a process for adding image identification information is provided. The process includes: calculating first feature vectors for partial regions selected from a target image to be processed; and adding a piece of first identification information indicating content of the target image to the target image using a group of decision trees that are generated in advance on the basis of second feature vectors calculated for partial regions of a learning image and a piece of second identification information added to the entire learning image.

Claims

1. A computer readable medium storing a program causing a computer to execute a process for adding image identification information, the process comprising:

calculating first feature vectors for partial regions selected from a target image to be processed; and

adding a piece of first identification information indicating content of the target image to the target image using a group of decision trees that are generated in advance on the basis of second feature vectors calculated for partial regions of a learning image and a piece of second identification information added to the entire learning image,

wherein the adding includes deciding the piece of the first identification information for the target image in order to add the piece of the first identification information to the target image by multiplying a prior probability of a piece of the second identification information by a ratio between a first product and a second product, the first product being calculated by multiplying likelihood functions obtained from a proportion of the number of pieces of the second identification information that have reached individual leaves of the group of decision trees with respect to the total number of pieces of the second identification information, the second product being calculated by multiplying prior probabilities of the calculated first feature vectors, when a group of the second feature vectors and a group of the pieces of the second identification information are supplied to the group of decision trees to calculate the first and the second products.

2. The computer readable medium according to claim 1, wherein the adding includes deciding the piece of the first identification information for the target image and adding the piece of the first identification information to the target image by performing weighting in accordance with the pieces of the second identification information that have reached the individual leaves of the group of decision trees.

3. The computer readable medium according to claim 1, wherein the adding includes deciding pieces of the first identification information for the partial regions of the target image and adding the pieces of the first identification information to the partial regions on the basis of the calculated first feature vectors and the group of decision trees.

4. An image identification information adding apparatus comprising:

a feature vector calculating unit that calculates first feature vectors for partial regions selected from a target image to be processed; and

an image identification information adding unit that adds a piece of first identification information indicating content of the target image to the target image using a group of decision trees that are generated in advance on the basis of second feature vectors calculated for partial regions of a learning image and a piece of second identification information added to the entire learning image,

wherein the image identification information adding unit decides the piece of the first identification information for the target image in order to add the piece of the first identification information to the target image by multiplying a prior probability of a piece of the second identification information by a ratio between a first product and a second product, the first product being calculated by multiplying likelihood functions obtained from a proportion of the number of pieces of the second identification information that have reached individual leaves of the group of decision trees with respect to the total number of pieces of the second identification information, the second product being calculated by multiplying prior probabilities of the calculated first feature vectors, when a group of the second feature vectors and a group of the pieces of the second identification information are supplied to the group of decision trees to calculate the first and the second products.

5. An image identification information adding method comprising:

calculating first feature vectors for partial regions selected from a target image to be processed; and

adding a piece of first identification information indicating content of the target image to the target image using a group of decision trees that are generated in advance on the basis of second feature vectors calculated for partial regions of a learning image and a piece of second identification information added to the entire learning image,

wherein the adding includes deciding the piece of the first identification information for the target image in order to add the piece of the first identification information to the target image by multiplying a prior probability of a piece of the second identification information by a ratio between a first product and a second product, the first product being calculated by multiplying likelihood functions obtained from a proportion of the number of pieces of the second identification information that have reached individual leaves of the group of decision trees with respect to the total number of pieces of the second identification information, the second product being calculated by multiplying prior probabilities of the calculated first feature vectors, when a group of the second feature vectors and a group of the pieces of the second identification information are supplied to the group of decision trees to calculate the first and the second products.