Data processing: artificial intelligence – Machine learning
Reexamination Certificate
2000-10-03
2003-09-30
Follansbee, John (Department: 2121)
Data processing: artificial intelligence
Machine learning
C706S020000, C706S045000, C706S046000
Reexamination Certificate
active
06629088
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to evaluating descriptive data and more specifically to measuring the quality of descriptors and description schemes.
BACKGROUND OF THE INVENTION
The Motion Picture Expert Group (MPEG) develops standards concerning audiovisual content. One component of the MPEG standard scheme includes MPEG-7 standards which are directed to providing fast and efficient identification of audiovisual content that may be of interest to the user. Specifically, the MPEG-7 standards are developed to regulate information describing the audiovisual content. Descriptions of audiovisual content may be used in various areas, including storage and retrieval of audiovisual items from databases, broadcast media selection, tele-shopping, multimedia presentations, personalized news service on the Internet, etc. According to the MPEG-7 standards, the descriptions are organized in the form of description schemes and descriptors. A descriptor is a representation of a feature of an audiovisual object. Typically, the descriptor defines the syntax and the semantics of the feature representation. A description scheme specifies the structure and semantics of the relationships between its components, which may be either descriptors or other description schemes. Currently, no unifying measure exists to compare audio descriptors with video descriptors, i.e., an audio descriptive method cannot be compared with a text or video descriptive method. In addition, no methods exist to compare description schemes. That is, no means exists to ensure that each component of a description scheme is functioning as well as the others.
Existing methods of comparing descriptions are typically based on creating distance measures when measuring the quality of descriptions. The distance measures use ratio or metric data which carries the most knowledge about the content it describes as compared to other types of descriptive data (e.g., interval data, rank order data, categorical data, or boolean data). However, a significant number of descriptions do not possess such specific information about the content being described as needed for creating distance measures.
Another disadvantage of using metric representations of similarity is that metric representations do not match human perception of quality measures. As Amos Tversky points out in “Features of Similarity,” Psychological Review, v. 84, n. 4, 1977, dimensional representations may be appropriate for certain stimuli (e.g., colors, tones) but not for others (e.g., faces, countries, or personalities). The assessment of similarity for such stimuli as faces, countries, or personalities may be better described as a comparison of features rather than as the computation of metric distance between points. When comparing sets of features, humans perceive similarity between the sets, and problems arise if a distance measure is substituted for this perception. These relationships may be described using “fuzzy set” relations. See Rising, H. K., “Creating a Biologically Plausible Model of Recognition which Generalizes Multidimensional Scaling,” Rogowitz, B. and T. Pappas, eds, Proc. SPIE 3644, 1999, pp. 411-420. Fuzzy set relations refer to relations that are not symmetric. For instance, statement “a is like b” may not be equivalent to the converse similarity statement “b is like a” (e.g., a Volvo does not have the same similarity to a car as a car does to a Volvo). Accordingly, for sets of features, metric representations of similarity may not be as accurate as fuzzy set relations, which discriminate based on mutual inclusion of the sets of features.
In contrast to metric representations of similarity, classification schemes are not subject to the above problems. First, classification schemes are based on categorical data. While many forms of numeric descriptions and descriptions which exist only as membership to a group do not carry such specific knowledge about the content as the knowledge required for ratio data, they do possess categorical information sufficient for classification schemes. Further, classification or categorization tasks respect the structure of similarities based on fuzzy set relations because they respect set inclusion.
Therefore, it would be advantageous to create a universal quality measure that can use categorical data, be compliant with human perceptual similarities and be capable of comparing various descriptors, as well as description schemes.
SUMMARY OF THE INVENTION
A method and apparatus for determining quality of a description are described. According to one embodiment, an exemplary method for determining quality of a description includes posing a classification task concerning at least one audiovisual object to a descriptive method that is used to create the description, generating a set of probabilities from a result of the classification task, and measuring an entropy of the result using the set of probabilities.
REFERENCES:
P. B., Kantor, et. al., “The Maximum Entrophy Principle In Information Retrieval,” Proceeding of the 1986 ACM Conference on Research and Development in Information Retrieval, Sep. 1986, pp. 269-274.
J.R. Ohm, et al., “A Multi-Feature Description Scheme For Image And Video Database Retrieval,” IEEE, 3rdWorkshop on Multimedia Signal Processing, Sep. 1999, pp. 123-128.
A. Puri, et al., “MPEG-4: An Object-Based Multimedia Coding Standard Supporting Mobile Applications,” Mobile Networks and Applications, vol. 3, Issue 1, 1998, pp. 5-32.
PCT Search Report dated May 31, 2001, 5 pages.
H. K., Rising, “Creating A Biologically Plausible Model of Recognition Which Generalizes Multidimensional Scaling,” SPIE vol. 3644, Jan. 1999, pp. 411-420.
H. K., Rising, “Axiomatization of Domain Knowledge For Modeling DS,” MPEG99/M5190, Melbourne, AU, Oct. 1999, 7 pages.
A. Tversky, et al., “Features of Similarity,” Psychology Review, vol. 84, No. 4, Jul. 1977, pp. 327-352.
Kyle Siegrist, Dept. of Mathematical Science, University of Alabama in Huntsville, “Probability Spaces, Virtual Laboratories In Probability And Statistics,” (Section 4), 12 pages.
MSI Insight II, Felix User Guide, Release Mar. 2000, Felix Command Language Reference, (Section A, Command Reference nd2), 83 pages.
International Organisation For Standardisation ISO/IEC JTC1 SC29/WG11 Coding of Moving Pictures And Audio, Geneva Switzerland, May 2000, 10 pages.
Ohm et al; A Multi-Feature Description Scheme for Image and Video Database Retrieval; Sep. 1999, IEEE; 0-7803-5610 1/99; 123-128.
Blakely , Sokoloff, Taylor & Zafman LLP
Follansbee John
Hirl Joseph P.
Sony Corporation
LandOfFree
Method and apparatus for measuring the quality of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for measuring the quality of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for measuring the quality of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3112514