Method of extracting candidate human region within image, system for extracting candidate human region, program for extracting candidate human region, method of discerning top and bottom of human image, system for discerning top and bottom, and program for discerning top and bottom

• US 20050196044A1
• Filed: 01/26/2005
• Published: 09/08/2005
• Est. Priority Date: 01/27/2004
• Status: Abandoned Application

• Current Assignee: Seiko Epson Corporation (Seiko Group)
• Original Assignee: Seiko Epson Corporation (Seiko Group)
• Inventors: Nagahashi, Toshinori, Hyuga, Takashi
• Application Number:
  US11/043,908
• Publication Number:
  US 20050196044A1
• Time in Patent Office: Days
• US Class Current: 382/190
• CPC Class Codes: G06T 11/60 Editing figures and text; Combining figures or text
• CPC Class Codes: G06T 2207/30196 Human being; Person
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/194 involving foreground-background segmentation
• CPC Class Codes: G06V 10/242 by image rotation, e.g. by 90 degrees
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

A method is provided for extracting a candidate human image region from within a picture image. The top-bottom direction of a picture image containing a human image is discerned. Based on the result of the discernment, the human image is modified to the actual top-bottom direction. Then, the variance values of image features are found about vertical and horizontal lines comprising vertical and horizontal rows of pixels constituting the picture image. Regions having variance values of image features in excess of their threshold values are selected about the vertical and horizontal line directions. An area within the picture image in which the selected regions in the line directions overlap with each other is extracted. Thus, an area within the picture image in which the human image exists can be robustly and quickly extracted.

Claims

1. A method of extracting a candidate human image region from within a picture image containing a human image, said method comprising the steps of:

finding variance values of image features about vertical and horizontal lines respectively including vertical and horizontal rows of pixels constituting said picture image;

selecting regions having variance values of image features in excess of threshold values about vertical and horizontal line directions; and

extracting an area within the picture image in which the selected regions in the line directions overlap each other as the region in which the human image exists.

2. A method of extracting a candidate human image region from within a picture image containing a human image, said method comprising the steps of:

discerning a top-bottom direction of the human image within said picture image;

rotating said picture image based on results of the discernment to modify said human image to an actual top-bottom direction;

thereafter finding variance values of image features about vertical and horizontal lines respectively including vertical and horizontal rows of pixels constituting said picture image;

selecting regions having variance values of image features in excess of threshold values about vertical and horizontal line directions; and

extracting an area within the picture image in which the selected regions in the line directions overlap each other as the region in which the human image exists.

3. A method of extracting a candidate human image region from within a picture image as set forth in claim 2, wherein:

when said picture image is surrounded by a polygonal frame having at least three angles and the human image contained in said picture image is a part principally including the upper half or head of a person, variance values of the image-characterizing values are found about rows of pixels constituting at least one of;

sides of the image frame; and

lines close to the sides; and

the side providing the highest variance value or the side close to the line providing the highest variance value is taken as a ground side in discerning the top-bottom direction of the human image within said picture image.

4. A method of extracting a candidate human image region from within a picture image as set forth in claim 3, wherein plural lines in said picture image which are close to each other are selected, and wherein an average value of variance values of image-characterizing values about the lines is used.

5. A method of extracting a candidate human image region from within a picture image as set forth in claim 2, wherein every given number of lines in lateral and height directions of said picture image are used as said vertical and horizontal lines.

6. A method of extracting a candidate human image region from within a picture image as set forth in claim 2, wherein said image features are calculated based on at least one of edge intensities and hue angles of the pixels of the rows of pixels constituting said lines.

7. A method of extracting a candidate human image region from within a picture image as set forth in claim 6, wherein said edge intensities use a Sobel edge detection operator.

8. A system for extracting a candidate human image region from within a picture image containing a human image, said system comprising:

an image reading portion for reading said picture image;

an image feature calculation portion for calculating image features about vertical and horizontal lines respectively including rows of pixels constituting the picture image read by said image reading portion;

a variance value calculation portion for finding variance values of the image features about vertical and horizontal line directions obtained by said image feature calculation portion; and

a candidate human image region detection portion for detecting a region in which the human image exists from regions in the line directions having the variance values of the image features in excess of threshold values in the vertical and horizontal line directions, the variance values being obtained by said variance value calculation portion.

9. A system for extracting a candidate human image region from within a picture image containing a human image, said system comprising:

an image reading portion for reading said picture image;

a top-bottom discerning portion for discerning a top-bottom direction of the human image within the picture image read by said image reading portion;

a top-bottom modification portion for rotating said picture image based on results of the discernment performed by said top-bottom discerning portion to modify said human image to an actual top-bottom direction;

an image feature calculation portion for calculating image features about vertical and horizontal lines respectively including rows of pixels constituting the picture image whose top-bottom direction has been modified by said top-bottom modification portion;

a variance value calculation portion for finding variance values of the image features about vertical and horizontal line directions obtained by said image feature calculation portion; and

a candidate human image region detection portion for detecting a region in which the human image exists from regions in the line directions having the variance values of the image features in excess of threshold values in the vertical and horizontal line directions, the variance values being obtained by said variance value calculation portion.

10. A system for extracting a candidate human image region from within a picture image as set forth in claim 9, wherein:

when the picture image is surrounded by a polygonal frame having at least three angles and the human image contained in the picture image is a part principally including the upper half or head of a person, the top-bottom discerning portion for discerning a top-bottom direction of the human image within said picture image finds variance values of image features about at least one of;

sides of the picture image; and

lines close to the sides; and

takes the side providing the highest variance value or the side close to the line providing the highest variance value as a ground side.

11. A program for extracting a candidate human image region from within a picture image containing a human image by causing a computer to perform functions of:

a top-bottom discerning portion for discerning a top-bottom direction of the human image within said picture image;

a top-bottom modification portion for rotating said picture image based on results of the discernment performed by said top-bottom discerning portion to modify said human image to an actual top-bottom direction;

an image feature calculation portion for calculating image features about vertical and horizontal lines respectively including rows of pixels constituting the picture image whose top-bottom direction has been modified by said top-bottom modification portion;

a variance value calculation portion for finding variance values of the image features about vertical and horizontal line directions obtained by said image feature calculation portion; and

a candidate human image region detection portion for detecting a region in which the human image exists from regions in the line directions having the variance values of the image features in excess of threshold values in the vertical and horizontal line directions, the variance values being obtained by said variance value calculation portion.

12. A program for extracting a candidate human image region from within a picture image as set forth in claim 11, wherein:

when said picture image is surrounded by a polygonal frame having at least three angles and the human image contained in said picture image is a part principally including the upper half or head of a person, the top-bottom discerning portion for discerning a top-bottom direction of the human image within said picture image finds variance values of image features about at least one of;

sides of the picture image; and

lines close to the sides; and

takes the side providing the highest variance value or the side close to the line providing the highest variance value as a ground side.

13. A method of discerning a top-bottom direction of a human image within a picture image in which the human image exists within a polygonal frame having at least three angles, the human image principally including the upper half or head of a person, said method comprising the steps of:

finding variance values of image features of rows of pixels constituting at least one of;

sides of said picture image; and

lines close to the sides; and

identifying the side providing the highest variance value or the side close to the line providing the highest variance value as a ground side.

14. A system for discerning top-bottom direction of a human image within a picture image containing the human image within a polygonal frame having at least three angles, the human image principally including the upper half or head of a person, said system comprising:

a variance value calculation portion for calculating variance values of image features of rows of pixels forming at least one of;

sides of said picture image; and

lines close to the sides; and

a top-and-bottom discerning portion for identifying the side providing the highest one of the variance values obtained by said variance value calculation portion or the side close to the line providing the highest one of the variance values obtained by said variance value calculation portion as a ground side.

15. A program for discerning top-bottom direction of a human image within a picture image containing the human image within a polygonal frame having at least three angles, the human image principally including the upper half or head of a person, said program comprising causing a computer to perform functions of:

a variance value calculation portion for calculating variance values of image features of rows of pixels forming at least one of;

sides of said picture image; and

lines close to the sides; and

a top-and-bottom discerning portion for identifying the side providing the highest one of the variance values obtained by said variance value calculation portion or the side close to the line providing the highest one of the variance values obtained by said variance value calculation portion as a ground side.