Image analysis – Applications – Personnel identification
Reexamination Certificate
1994-08-09
2001-01-30
Johns, Andrew W. (Department: 2721)
Image analysis
Applications
Personnel identification
Reexamination Certificate
active
06181805
ABSTRACT:
BACKGROUND OF THE INVENTION
This invention relates to an object image detecting method in which an image of an object having features, such as a human face, is taken by a camera from an arbitrary direction and the image thus taken is matched to dictionary or library images previously generated and stored whereby a region of the object in the image is detected, and relates to a method of determining which category the object in the detected object region belongs to among predetermined categories such as human faces classified into corresponding categories and previously registered as well as detecting whether the object in the detected object region belongs to a predetermined category or not and also relates to a system using that method.
There may be a case where it is needed, for example, to automatically identify persons who wish to enter or exit from a particular room, buiding or area so that only predetermined particular persons such as employees or customers are to be permitted to enter or exit therefrom. Also, there may be a case where it is required to discriminate a doubtful or suspicious character or characters from among a great number of unspecified people who enter or exit from a bank or intend to do so. In such cases, it has been proposed to take pictures of people who enter or exit from a particular place or area or intend to do so by a camera and to identify or determine the faces of the people from the images thereof taken.
A prior art apparatus for extracting the object region in the image thereof is based on a thresholding of intensity and/or color information. For example, when a facial region of a person is extracted from the image thereof, intensity and/or saturation and/or hue corresponding to the skin are obtained to set a threshold for detecting the skin region thereby extracting the skin region from the whole image. Alternatively, there is a method of extracting the object region by modeling the shape of the object and fitting it with the edge of the image.
Another prior art object identifying apparatus is provided with a dictionary or library of facial images in a particular direction of a person or persons to be identified such as the direct frontal faces thereof, the side faces thereof or the like and matches a test image inputted thereinto to the facial images of the dictionary whereby the person is identified (see, for example, Ashoc Samal, Prasana A. Iyengar: “Automatic recognition and analysis of human faces and facial expressions” Pattern recognition, Vol. 25, No. 1, pp65-77, 1992).
The prior art object extracting apparatus using thresholding mentioned above is needed to set the optimum threshold. This optimum threshold may often vary with the direction or orientation of the object. Therefore, if the direction or orientation of the object is unknown, it is difficult to set the optimum threshold for extracting the object region. Also, the prior art method of fitting a modeled shape of the object with the edge of the image cannot extract the object region with high accuracy if the shape of the object in its image varies with the direction or orientation of the object, or the edge representing the contour or outline of the object cannot be extracted correctly from the image due to occlusion of a portion of the object or shadow of the object.
Further, the prior art object identifying apparatus is required to input the image (e.g., the direct frontal face) taken from the same direction as that from which the image prepared and stored in the dictionary is taken, and therefore has a limited scope to which the appratus is applied since it is necessary to limit the direction from which the object is taken by any procedure or means.
SUMMARY OF THE INVENTION
An object of the present invention is to provide an object image detecting method and system which are capable of detecting the object region of the image by use of high speed processing and stably without depending on the direction from which the object is taken, and to provide a method and system which can detect which category the object belongs to among various predetermined categories (for example, registered faces of people, children's faces, adult female faces, adult male faces or the like), and to provide a method and system which are able to detect whether the object detected belongs to a predetermined category or not.
In accordance with the present invention, images of an object or objects belonging to either one of a predetermined number of categories (L kinds) to be recognized are previously taken by a camera or cameras from a plurality of predetermined M directions, and a region (matching region) of the object to be recognized is extracted from each image, and N dictionary images (N=M×L) representative of the directions and categories are generated from the images, and the dictionary images are stored in dictionary image generating means together with the directions or orientations of the object (the directions from which the object is taken, namely, the directions of the camera). When a test image of an object to be analyzed is given, the degree of similarity or resemblance r(n, X, Y) to the n-th dictionary image (n=1, 2, . . . , N) at a matching position (X, Y) in the test image is computed by matching means, and this matching process is performed with (X, Y) and n varied, and the position (X
max
, Y
max
) where the degree of similarity r(n, X, Y) thereof becomes the highest is outputted as a region corresponding to the object region.
According to the object region detecting method and system of the present invention, if the number M of the directions of the dictionary images representative of the objects belonging to the categories (L kinds) to be recognized is properly set in the dictionary image generating means, even though test images including objects to be recognized which look in different directions are provided, the position of the matching region of the test image where the degree of similarity between the matching region of the test image and the dictionary image becomes the highest accords with the position of the object to be recognized so that the dictionary image at that time can be determined to represent the category to which the object to be recognized belongs. Thus, it is possible to extract the matching region so as to include the object to be recognized without depending on the direction or orientation of the object to be recognized in the test image.
REFERENCES:
patent: 3805238 (1974-04-01), Rothfjell
patent: 4021778 (1977-05-01), Ueda et al.
patent: 4486775 (1984-12-01), Catlow
patent: 4754487 (1988-06-01), Newmuis
patent: 5008946 (1991-04-01), Ando
patent: 5012522 (1991-04-01), Lambert
patent: 5046109 (1991-09-01), Fujimori et al.
patent: 5164992 (1992-11-01), Turk et al.
patent: 5410609 (1995-04-01), Kado et al.
Ishii Kenichiro
Iso Toshiki
Koike Hideki
Shimada Satoshi
Tomono Akira
Johns Andrew W.
Nippon Telegraph & Telephone Corporation
Pollock, Vande Sande & Amernick, R.L.L.P.
LandOfFree
Object image detecting method and system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Object image detecting method and system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Object image detecting method and system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2557557