METHOD, APPARATUS, AND COMPUTER PROGRAM PRODUCT FOR GENERATING AN IMAGE INDEX AND FOR INTERNET SEARCHING AND QUERYING BY IMAGE COLORS
 Inventor: Helen Roxlo Delp, Rochester, Minn.
 Assignee: International Business Machines Corporation, Armonk, N.Y.
 Appl. No.: 08/965,621  Filed: Nov. 6, 1997
 Int. CI.7 G06F 17/30
 U.S. CI 707/104; 707/3; 382/162;
 Field of Search 707/1, 3, 101,
707/104; 382/162, 169
 References Cited
U.S. PATENT DOCUMENTS
4,743,959 5/1988 Fredericksen 358/il
5,091^718 2/1992 Beatty 340/703
5,155,589 10/1992 Gere 358/81
5,204,665 4/1993 Bollman et al 340/703
5,289,297 2/1994 Bollman et al 358/537
5,502,458 3/1996 Braudaway et al 345/153
5,509,111 4/1996 Hong et al 395/131
5,517,334 5/1996 Morag et al 358/518
5,586,197 12/1996 Tsujimura et al 382/612
5,652,881 7/1997 Takahashi et al 707/104
5,670,986 9/1997 Leak 345/154
5,889,523 3/1999 Wilcox et al 345/357
Draft International Standard ISO/IEC DIS 10918-1, "Information Technology—Digital Compression and coding ol continuous-Tone Still Images", International Organization for Standardization, 1991 International Electrotechnical Commission, 1991.
U.S. patent application Serial No. 08/749,150 filed Nov. 14, 1996 by Helen R. Delp and entitled "Method and Apparatus for Creating a Color Name Dictionary and for Querying an Image by Color Name".
Corridoni et al, Pyramid Retrieval by Color Perceptive
Regions, Visual Languages, 1997, Proceedings. 1997 IEEE
Symposium, pp. 205-211, Sep. 1997.
Primary Examiner—-Thomas Black
Assistant Examiner—Frantz Coby
Attorney, Agent, or Firm—-Joan Pennington
A method, apparatus and computer program product are provided for building an image index and for searching and querying by image colors from a plurality of images including images from the internet or other network environment. First an index is built for each of the plurality of images. A selected number of levels is identified for image processing, and an upper resolution and a lower resolution are identified for each of the selected number of levels for image processing. An image is filtered at each selected level. A sample of pixels is acquired and stored at each selected level in the index. At query time, an image color is selected. A entry of the index is acquired. The acquired entry is compared with the selected image color. A match is identified responsive to the acquired entry being near the selected image color.
28 Claims, 12 Drawing Sheets