Image analysis – Image transformation or preprocessing – Image storage or retrieval
Reexamination Certificate
1999-07-08
2003-06-03
Couso, Yon J. (Department: 2625)
Image analysis
Image transformation or preprocessing
Image storage or retrieval
C358S403000
Reexamination Certificate
active
06574378
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to the field of computer-based image analysis, and more particularly to systems of classifying and searching image data dependent upon the content of the image data.
BACKGROUND
Text retrieval based on keywords has been the main stream in the field of information retrieval. A number of existing visual retrieval systems extract and annotate the visual content of data objects manually, often with some assistance by means of user interfaces. An example of such a system is given by Rowe, L. A., Borexzky, J. S., and Eads, C. A., “Indices for User Access to Large Video Databases,”
Storage and Retrieval for Image and Video Databases II, Proc. SPIE
2185, 1994, pp. 150-161. Once keywords are associated with the visual content, text retrieval techniques can be used easily. Although text descriptions may reflect the (largely conceptual) semantics of multimedia data, they may also result in a combinatorial explosion of keywords in the attempted annotation due to the ambiguous and variational nature of multimedia data. Also, there is a limit to how much semantic information textual attributes can provide, as described by Bolle, R. M., Yeo, B. L., and Yeung, M. M., “Video Query: Research Directions”,
IBM Journal of Research and Development,
42(2), March 1998, pp. 233-252.
On the other hand, visual content-based retrieval systems have mainly focused on using primitive features such as colour, texture, shape, etc., for describing and comparing visual contents. Examples of such systems include: Bach, J. R., et al., “Virage Image Search Engine: An Open Framework for Image Management,”
Storage and Retrieval for Image and Video Databases IV, Proc. SPIE
2670, 1996, pp. 76-87; Niblack, W., et al., “The QBIC Project: Querying Images By Content Using Colour, Textures and Shapes,”
Storage and Retrieval for Image and Video Databases, Proc. SPIE
1908, 1993. pp. 13-25; and Pentland, A., Picard, R. W., and Sclaroff, S., “Photobook: Content-Based Manipulation of Image Databases,”
Intl. J of Computer Vision,
18(3), 1996, pp. 233-254. When these feature-based techniques are applied to individual objects, an object is often the focus of retrieval. Not much consideration has been given to the interrelationship among the objects.
Region-based query methods rely on colour and texture segmentation to transform raw pixel data into a small set of localised, coherent regions in colour and texture space (“blobworld”) and perform similarity-based retrieval using these regions. Such a system is described by Carson, C. et al., “Region-based image querying,”
Proc. IEEE Workshop on Content
-
Based Analysis of Images and Video Libraries,
1997, pp. 42-49. However, the regions are derived from each individual image without reference to any consistent attributes across images in a domain. Moreover, the segmentation of regions is in general not robust and may result in perceptually incoherent regions without meaningful semantics.
The VisualSEEK system has been described by Smith, J. R. and Chang, S.-F., “VisualSEEk: A Fully Automated, Content-Based Image Query System,”
Proc. ACM Multimedia
96, Boston, Mass., Nov. 20, 1996. This system and its descendants consider the spatial relationship among regions and combine it with primitive features of the regions for image retrieval. The matching algorithm merges lists of image candidates, resulting from region-based matching between a query and database images, with respect to some threshold and tends to be rather complex and ad hoc in realisation. The segmentation of regions is based on colour only, and no object or type information is extracted from the segmented regions.
In a different approach that advocates the use of global configuration, Ratan, A. L., and Grimson, W. E. L., “Training Templates for Scene Classification Using a Few Examples,”
Proc. IEEE Workshop on Content
-
Based Analysis of Images and Video Libraries,
1997, pp. 90-97, describe a method for extracting relational templates that capture the colour, luminance and spatial properties of classes of natural scene images from a small set of examples. The templates are then used for scene classification.
Although the method automates previous effort that handcrafted the templates, such as Lipson, P., Grimson, E., and Sinha, P., “Configuration Based Scene Classification and Image Indexing,”
Proc. of CVPR'
97, 1997, pp. 1007-1013, scene representation and similarity matching are computed through the relationships between adjacent small regular partitions, which are rather complex for comprehension.
In general, for text documents, the segmentation and extraction of keywords are relatively straight forward, since keywords are symbolic in nature. For visual data, which are perceptual and pattern-based, no equivalent visual keywords have been proposed.
U.S. Pat. No. 4,839,853 issued to Deerwester, et al. on Jun. 13, 1989 describes a methodology that exploits higher-order semantic structure implicit in the association of terms with text documents known as Latent Semantic Analysis (LSA). Using singular value decomposition (SVD) with truncation, LSA captures underlying structure in the association of terms and documents, while attempting to remove the noise or variability in word usage that plagues word-based retrieval methods. The derived coded description achieves a reduction in dimensionality while preserving structural similarity in term-document association for similarity matching. However, no similar method has been proposed for visual domains since there is no equivalent notion of visual keywords.
U.S. Pat. No. 4,944,023 issued to Imao, et al. on Jul. 24, 1990 describes a method for describing an image by recursively and equally dividing the image into 2
n
regions until each region includes two or less kinds of regions, and each of the 2
n
regions further into 2
n
sub-regions of the same kind. Thus, an image is represented as a binary tree of local homogeneous regions. Though it cuts an image into types of regions and sub-regions, it does not further extract regularities of these regions across images nor uses the tree representation for comparing image similarities.
U.S. Pat. No. 5,710,877 issued to Marimount, et al. on Jan. 20, 1998 describes an image structure map (ISM) to represent the geometry, topology and signal properties of regions in an image and to allow spatial indexing of the image and those regions. Another objective of ISM is to support interactive examination and manipulation of an image by a user. No attempt is given to define object or type information across collection of images or to use ISM as a means for image retrieval. U.S. Pat. No. 5,751,852 issued to Marimount, et al. on May 12, 1998 elaborates on the ISM.
U.S. Pat. No. 5,751,286 issued to Barber, et al. on May 12, 1998 describes an image query system and method underlying the QBIC system. Queries are constructed in an image query construction area using visual characteristics such as colours, textures, shapes, and sizes. Retrieval of images is performed based on the values of representations of the visual characteristics and the locations of the representations of the query in the image query construction area. Aggregate measures of low-level features are often used for similarity matching. No object or type information is compiled from the images in the database for comparing similarity between two image contents.
U.S. Pat. No. 5,781,899 issued to Hirata on Jul. 14, 1998 describes a method and system to produce image index for image storage and management system. Images similarity matching is based on zones, which consist of at least one pixel in the original images, divided from the images using hue, brightness, and grid information. The size of the zone-divided image is adjusted according to the similarity for integrating the zones against some threshold to determine the total number of zones for use as an index. The zones are specific to an image. No prototypical zones are compiled in advance for describing image contents.
U.S. Pat. No. 5,802,361 issued to Wang, et
Couso Yon J.
Kent Ridge Digital Labs
Ladas & Parry
LandOfFree
Method and apparatus for indexing and retrieving images... 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 indexing and retrieving images..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for indexing and retrieving images... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3095704