Human image retrieval system

• US 8,306,281 B2
• Filed: 10/29/2008
• Issued: 11/06/2012
• Est. Priority Date: 11/01/2007
• Status: Active Grant

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Hiroike, Atsushi, Morimoto, Yasutsugu, Akira, Naoto, Matsubara, Daisuke
• Primary Examiner(s): Desire, Gregory M
• Application Number:
  US12/260,489
• Publication Number:
  US 20090060294A1
• Time in Patent Office: 1,469 Days
• Field of Search: 345/16, 345/427, 382/118, 382/169, 382/195, 382/305, 382/103, 382/190, 382/218, 707/3
• US Class Current: 382/118
• CPC Class Codes: G06F 16/532 Query formulation, e.g. graphical querying
• CPC Class Codes: G06F 2218/00 Aspects of pattern recognition specially adapted for signal processing

Abstract

In a similar face retrieval system for retrieving an image photographing a face similar to a face detected from a retrieval query image from a retrieval target image group by using an image photographing a human face as the retrieval query image, whole image features as features representative of background information are extracted from each whole area of an each image of a retrieval target image group, to calculate a degree of similarity through comparison with each set of whole image features, and an image having a degree of similarity not lower than a certain value and having a lower retrieval result order from retrieval results. It is possible to efficiently retrieve the same person playing in different scenes by utilizing different features for a retrieval process and a filtering process.

Claims

1. A human image retrieval system for retrieving an image having a human similar to a human detected from a retrieval query image from a retrieval target image group, comprising:

means for inputting said retrieval query image;

means for detecting a face area from each image of said retrieval target image group;

means for extracting face image features from each face area detected from said retrieval target image group;

means for extracting whole image features from a whole area of each image of said retrieval target image group;

means for detecting a face area from said retrieval query image;

means for extracting face image features from said retrieval query image;

means for retrieving a face image similar to a face image in said retrieval query image from said retrieval target image group by using said face image features, and storing n retrieval results in storage means in an order of a higher degree of similarity;

means for comparing whole image features of a first image, in a query image, having a specific i-th order in said n retrieval results with whole image features of a second image having a lower degree of similarity than a degree of similarity of said first image, to calculate a degree of similarity between said first and second images;

means for excluding said second image from a set of said retrieval results if a value of said degree of similarity is not lower than a predetermined threshold value;

means for changing said specific order i←

i+1, i+2, . . . , n−

1, and comparing whole image features of an image of a changed specific order with whole image features of an image having a lower order than the changed specific order, to calculate a degree of similarity;

filtering means for excluding said image having a lower order from a set of said retrieval results if a value of said degree of similarity is not lower than a predetermined threshold value; and

means for outputting and displaying images left after filtering by said filtering mean.

2. The human image retrieval system according to claim 1, wherein a strength of a filtering process is changed by changing said threshold value.

3. A human image retrieval system for retrieving an image photographing a human similar to a human photographed in a retrieval query image from a retrieval target image group, comprising:

means for inputting said retrieval query image;

means for detecting a human area from each image of said retrieval target image group;

means for extracting color features from each human area detected from said retrieval target image group;

means for detecting a human area from said retrieval query image;

means for extracting color features from said retrieval query image;

means for extracting location features of a location at which each image of said retrieval target image group was photographed;

means for retrieving a human image similar to a human image in said retrieval query image from said retrieval target image group by using said color features, and storing n retrieval results in storage means in an order of a higher degree of similarity;

means for comparing location features of a first image, in a query image, having a specific i-th order in said n retrieval results with location features of a second image having a lower degree of similarity than a degree of similarity of said first image, to calculate a distance between a photographed location of said first image and a photographed location of said second image;

means for excluding said second image from a set of said retrieval results if said degree of similarity is not shorter than a predetermined threshold value;

means for changing said specific order i←

i+1, i+2, . . . , n−

1, and comparing location features of an image of a changed specific order with location features of an image having a lower order than the changed specific order, to calculate a distance between respective photographed locations;

filtering means for excluding said image having a lower order from a set of said retrieval results if said degree of similarity is not lower than a predetermined threshold value; and

means for outputting and displaying images left after filtering by said filtering mean.

4. The human image retrieval system according to claim 3, wherein a strength of a filtering process is changed by changing said threshold value.

5. The human image retrieval system according to claim 3, wherein said distance is one of a Manhattan distance, a Mahalanobis and a distance between vector data.

6. A human image retrieval system for retrieving an image photographing a human similar to a human photographed in a retrieval query image from a retrieval target image group, comprising:

means for inputting said retrieval query image;

means for detecting a human area from each image of said retrieval target image group;

means for extracting color features from each human area detected from said retrieval target image group;

means for extracting whole image features of a whole image of each image of said retrieval target image group;

means for detecting a human area from said retrieval query image;

means for extracting color features from said retrieval query image;

means for retrieving a human image similar to a human image in said retrieval target image from said retrieval target image group by using said color features, and storing n retrieval results in storage means in an order of a higher degree of similarity;

means for comparing whole image features of a first image, in a query image, having a specific i-th order in said n retrieval results with whole image features of a second image having a lower degree of similarity than a degree of similarity of said first image, to calculate a degree of similarity between said first and second images;

means for excluding said second image from a set of said retrieval results if said degree of similarity is not lower than a predetermined threshold value;

means for changing said specific order i←

i+1, i+2, . . . , n−

1, and comparing whole image features of an image of a changed specific order with whole image features of an image having a lower order than the changed specific order, to calculate a degree of similarity;

filtering means for excluding said image having a lower order from a set of said retrieval results if said degree of similarity is not lower than a predetermined threshold value; and

means for outputting and displaying images left after filtering by said filtering mean.

7. The human image retrieval system according to claim 6, wherein a strength of a filtering process is changed by changing said threshold value.