Image analysis – Color image processing – Pattern recognition or classification using color
Reexamination Certificate
1999-12-13
2004-03-23
Mehta, Bhavesh M. (Department: 2625)
Image analysis
Color image processing
Pattern recognition or classification using color
C382S167000, C382S168000, C382S170000, C382S171000, C382S190000
Reexamination Certificate
active
06711287
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to an image-feature extraction method of extracting feature parameters representing features of an image from the image. More particularly, this invention relates to an image-feature extraction method of enabling extraction of image-feature parameters suited to human sensibility on similarity from the image and also of automating a work to extract feature parameters from the image when feature parameters representing features of an image are to be extracted from the image.
BACKGROUND OF THE INVENTION
Recently, in association with development of computer technology and image processing technology, accumulating enormous amount of electronic images to construct an image database have actively been attempted. When constructing such an image database, the most important thing is the availability of an image retrieval method so that a desired image can easily be acquired from a large number of accumulated images.
As one of the image retrieval methods as described above, there is one comprising the steps of allocating a search key with some term or terms to each image, verifying a keyword inputted in the same method as that for document retrieval with the search key allocated to the image, and returning the image with the search key matching the keyword as a search result.
However, an impression which a person from about an image varies from person to person. Therefore, a search key allocated to an image does not always cover every impression a person has from the image. Thus, so that there are many cases in which a satisfactory image can not be acquired in image retrieval using a search key based on terms inputted by an operator. The reason behind this is that, people can not find in many cases appropriate terms to express features concerning a shape of things such as a person, animal, or a building appearing in the image (which are collectively called as “object” hereafter) and a positional relation among the objects.
Therefore, an image should not be retrieved based on a result of verification of a keyword against a search key with terms, but it is desirable to construct a system so as to provide a particular image (described “query image”) as a criteria for searching and enable retrieval of an image which is similar to the query image from the image database. To realize the retrieval method described above, there has been proposed a method in which feature parameters representing features of each image are retrieved from the query image as well as from an image as a target for retrieval, similarity between the feature parameters of the query image and those of the image as a target for retrieval are determined, and the image having the feature parameters similar to those of the query image is returned as a search result. Namely, this image retrieval method is a method of retrieving a similar image based on the idea that the similarity between features is equivalent to similarity between images themselves.
There are two papers that disclose such a retrieval method. The first paper discloses a query method of analyzing images themselves, extracting colors (a color histogram), texture, and shapes of objects or so in the images as feature parameters of each image to construct a database, and acquiring an image with specified feature parameters by specifying the feature parameters of the image in the database when the image is queried. This paper is by Myron Flickner et. al., “Query by Image and Video Content: The QBIC System” Computer, September 1995, pp 23-32.
The second paper discloses a method of retrieving a similar image by generating a color histogram of each image using a cylindrically-segmented HSV color space and by using the generated color histogram as feature parameters of each image. This paper is by John R. Smith and Shih-Fu Chang, “Tools and Techniques for Color Image Retrieval” SPIE, Vol. 2670, pp426-437.
With the image query method disclosed in the first paper, it is possible to query an image by using feature parameters of various images. However, when feature parameters are extracted from images and similarity between the images is determined based on the extracted feature parameters, the feature parameters do not fully cover the scale in which a human determines the similarity. Therefore, precision of retrieval is rather low. For example, feature parameters extracted based on the shape of an object only expresses how much the object resembles a circle (circularity) and in which direction the object is compressed (central moment). Namely, the extracted features are so rough that it is impossible to perform image query based on appropriate and detail impressions which a human has when looking at an object in the image. Further, there is a great possibility that the retrieval result contains noise.
Further, in the first paper, it is possible to perform image search based on a shape of an object as feature parameters of the image. However, a user is required to specify a locus of an edge of the object existing in the image when the object is extracted from the image. This requires an extra effort. An object in the image is one of the most noticeable feature when a person looks at the image. Therefore, by extracting a shape of an object as image-feature parameters and using the parameters for image retrieval, it is possible to perform high precision image retrieval based on the impressions which a human has from the image. However, as extra efforts are required, an enormous amount of manpower and time are required to construct an image database with a large amount of images accumulated therein.
In the second paper, the HSV color space used for extracting a color histogram from an image as feature parameters of the image is simply segmented into cylindrical shapes, so that some wastefully segmented areas are generated and increase in the amount of data is disadvantageously caused. More specifically, when there are a small number of segments in the color space, even colors dissimilar to each other may sometimes be determined as similar ones, and precision in determination of similarity between images gets worse. On the other hand, when a number of segments in the color space is too large, the amount of data for the color histogram becomes too much, and a long time is required for computing the similarity between the images. In addition, when the number of segments in the color space is too large, the colors which a human feels the same as or similar to each other may be determined as dissimilar ones. The method disclosed in the second paper corresponds to the case where the number of segments in the color space is too large, and the method includes the problems that a long time is required for computing similarity, and that human sensibility on feel are not sufficiently satisfied.
Namely, it may be said that, in the conventional technology, impressions which a human acquires from an image are not sufficiently reflected to feature parameters extracted from the image and an operational environment suited for use is not provided. For instance, a heavy work load is required for extracting feature parameters from an image. Further it may be said that, in the conventional technology, impressions which a human acquires from an image are not sufficiently reflected to features which are extracted from the image, and that it is difficult to acquire a result of retrieval as desired by an operator. In other words, in the conventional technology, attention is not focused on a human as an operator, and a user-friendly environment for working is not provided. To improve convenience in working, feature parameters should be extracted from an image without requiring manpower, and the extracted feature parameters should be suited to human sensibility on similarity, and whether a retrieved image completely or substantially satisfies the conditions for searching set by a human should be determined based on human sensibility on similarity.
SUMMARY OF THE INVENTION
It is an object of the present invention to automate, when feature parameter
Bayat Ali
Dickstein , Shapiro, Morin & Oshinsky, LLP
Mehta Bhavesh M.
Ricoh & Company, Ltd.
LandOfFree
Image-feature extraction method and computer-readable record... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image-feature extraction method and computer-readable record..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image-feature extraction method and computer-readable record... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3279373