Y. Rui, T. S. Huang and S. Mehrotra, "Relevance Feedback Techniques in Interactive Content-based Image Retrieval," in Proc. IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, San Jose, CA, USA, Jan. 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....based on the quality of the features extracted from images and the ability of the user to provide a good query. Relevance feedback can be richer than this. In particular, the information the user provides need not be expressible in the query language, but may entail modifying feature weights [22] or constructing new features on the fly [23] PicHunter takes this idea further with a Bayesian approach, representing its uncertainty about the user s goal by a probability distribution over possible goals. This Bayesian approach to the problem was pioneered by Cox et al. 3] With an explicit ....

Y. Rui, T. S. Huang, and S. Mehrotra, "Relevance feedback techniques in interactive content-based image retrieval," in Proc. of ISFJT and SPIE Storage and Retrieval of Image and Video Databases VI, San Jose, CA, January 1998.

.... yet artistically inclined user this nexibility manifests itself in allowing the generation of complex queries reinacting what is important or unimportant (i.e. inlcusion or exclusion of features within a hybrid query) This property can be advantageous when the technique of Relevance Feedback [10] is introduced into the CBIR system. The use of this concept would allow for desirable or undesirable features to be interactively, and iteratively promoted or culled during retrieval. Successive feedback queries would thus strive to achieve two goals; conform to each user s concept of an ideal ....

Y. Rui, T. Huang, S. Mehrotra, "Relevance Feedback Techniques in Interactive Content-Based Image Retrieval", SPIE, Vol. 3312, 1997.

....for applications such asimag editing analysis andimag publications etc. In addition, recent e#orts incombining the low level features with some type ofhig##BfbT information proves to be useful towards information retrieval from the imag content. Representative examples include relevance feedback [2,3], user intention prediction via Bayesian models [4] and semantic information extraction from low level features [5] etc. On top of that, numerous semantic features can also be added to provide a wide rang ofchoices for manual control and selection before any content based retrieval is even ....

Y. Rui, T.S.Huang S. Metrotra, Relevance feedback techniques in interactive content-basedimag retrieval, Proceeding ofIS& T and SPIEStorag and Retrieval of Imag and Video Databases VI, San Juan, PR, June 1997, pp. 762--768.

....and index artefacts. A variety of classification and indexing methods are described and used by developers, e.g. facets [12] keywords, enumerated schemes [21] multimedia properties [16, 20] or media indexes [7] Incorporate relevance feedback to deal with individual perception and preferences [15]. Resulting Context High quality artefacts that are properly classified and indexed can be easily retrieved and then reused within an organisation. For example, in an educational institution where some teachers lack sufficient experience in multimedia design, teaching staff could benefit from ....

....development task, but the suitability cannot be properly judged without taking use context into consideration [8, 22] Subjectivity. Also formulating and satisfying requirements describing intended use context might be hard due to individual perception of and preferences for multimedia properties [15]. Solution Development with reuse focuses on constructing products with the utilisation of reusable components drawn from the repository. Multimedia developer should apply intended use context in order to improve selected and rejected artefacts classification: Provide relevance feedback on ....

Rui, Y., Huang, T. S., and Mehrotra, S. Relevance Feedback Techniques in Interactive Content-Based Image Retrieval. in IS&T and SPIE Conference on Storage and Retrieval for Image and Video Databases VI. 1998. San Jose, CA, USA: SPIE, http://www.research.microsoft.com/~yongrui/html/publication.html.

....data to modify the query feature vector and give higher weights to the common components of the correlogram found in the relevant images. A similar approach was used in Rui et. al [23] on image texture feature representation. Instead of modifying feature level descriptions directly, Rui et. al [24] examined how relevance information may be used to optimize the ratios of combining various features in a multi feature image retrieval system. The user s relevance judgments were used to update these ratios dynamically in order to obtain a better way of combining the various features. In another ....

Y. Rui, T.S. Huang, and S. Mehrotra. Relevance feedback techniques in interactive contentbased image retrieval. In Proceedings of IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, 1998.

....to modify the query feature vector and give higher weights to the common components of the correlogram found in the relevant images. A similar approach was used in Rui et al. [20] on image texture feature representations. Instead of modifying feature level descriptions directly, Rui et al. [21] examined how relevance information may be used to optimise the ratios of combining various features in an integrated content based image retrieval system. The user s relevance judgements are used to update these ratios dynamically to obtain a better way of combining the various features. In ....

Rui Y., Huang T.S. and Mehrotra S. Relevance Feedback Techniques in Interactive Content-Based Image Retrieval. Proceedings of IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, San Jose, California, Jan. 1998.

....of the retrieval. From a user s perspective, searching for images in similarity search systems typically involves several steps. In the first few steps, a user refines his or her query with the help of relevance feedback until the query matches the information need sufficiently good (cf. MARS [RHM98] CHARIOT [The00] In the final step of the search process, the archive is extensively searched for (all) relevant images. Obviously, retrieval effectiveness in the first few steps is not so important as retrieval efficiency. In the final step, on the other hand, result quality plays the key ....

Y. Rui, T. Huang, and S. Mehrotra. Relevance Feedback Techniques in Interactive Content-Based Image Retrieval. In Storage and Retrieval for Image and Video Databases (SPIE), pages 25--36, San Jose, California, USA, Jan. 1998.

.... it must have 3 digit, e.g. 002, 100, 030, the relevanceI value may be a if all relevant images are used for the comparation process, or may be m if the mean of the query image is used, readWeights is 0 when the weights are not used, or 1 otherwise, weightsType is s (Mars type [4, 7, 6]) or m (rover type [8] the useWeightRange default value is 0 (false) otherwise is 1, the weightRange default value is 0.10, the exponent default value is 0 (this means that all the descriptors have the same weight; value 1 means that weight = 1=sqrt( value 2 means that weight = 1= ....

Y. Rui, T.S. Huang, and S. Mehrotra. Relevance Feedback Techniques in Interactive Content-Based Image Retrieval. In Proc. of IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, San Jose, CA, January 24-30 1998.

....based on the quality of the features extracted from images and the ability of the user to provide a good query. Relevance feedback can be richer than this. In particular, the information the user provides need not be expressible in the query language, but may entail modifying feature weights [22] or constructing new features on the y [23] PicHunter takes this idea further with a Bayesian approach, representing its uncertainty about the user s goal by a probability distribution over possible goals. This Bayesian approach to the problem was pioneered by Cox et al. 3] With an explicit ....

Y. Rui, T. S. Huang, and S. Mehrotra, \Relevance feedback techniques in interactive content-based image retrieval," in Proc. of IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, San Jose, CA, January 1998.

.... a k nearest neighbor (k NN) algorithm on each individual feature and then merging them to get the overall answers [17] 14] 6] 7] Due to the subjective nature of top k queries, the answers returned by the system to a user query usually do not satisfy the user s need right away [18] 4] [23], 11] This can happen due to several reasons: The starting examples may not be the best ones to capture the information need (IN) of the user or the starting weights may not accurately capture the users perception or both. In this case, the user would like to refine the query and resubmit it in ....

....new query is called the refined query. In a QBE environment (e.g. multimedia databases) the user typically refines the query by finding, among the answers returned to the starting query, one or more objects that are closest to what she wants and requesting for more objects like those [22] [23], 18] 4] 11] Based on the user feedback, the system will compute the new query objects and the new weights and execute the refined query. Another way to refine the query is that the user explicitly modifies the perception model, i.e. she explicitly changes the weights of the features so as ....

[Article contains additional citation context not shown here]

Y. Rui, T. Huang, and S. Mehrotra, "Relevance Feedback Techniques in Interactive Content-Based Image Retrieval," Proc. IS & T and SPIE Storage and Retrieval of Image and Video Databases, 1998.

.... executing a k nearest neighbor (k NN) algorithm on each individual feature 2 and then merging them to get the overall answers [24, 21, 9, 10] Due to the subjective nature of top k queries, the answers returned by the system to a user query usually does not satisfy the user s need right away [25, 5, 32, 15]. This can happen due to several reasons: the starting examples may not be the best ones to capture the information need (IN) of the user or the starting weights may not accurately capture the users perception or both. In this case, the user would like to refine the query and resubmit it in order ....

....and the new query is called the refined query. In a QBE environment (e.g. multimedia databases) the user typically refines the query by finding, among the answers returned to the starting query, one or more objects that are closest to what she wants and requesting for more objects like those [30, 31, 32, 25, 5, 15]. Based on the user feedback, the system will compute the new query objects and the new weights and execute the refined query. Another way to refine the query is that the user explicitly modifies the perception model i.e. she explicitly changes the weights of the features so as to better capture ....

[Article contains additional citation context not shown here]

Y. Rui, T. Huang, and S. Mehrotra. Relevance feedback techniques in interactive content-based image retrieval. Proc. of IS&T and SPIE Storage and Retrieval of Image and Video Databases, 1998.

.... by first executing a nearest neighbor (k NN) algorithm on each individual feature 2 and then merging them to get the overall answers [17, 14, 7, 8] Due to the subjective nature of top queries, the answers returned by the system to a user query usually do not satisfy the user s need right away [18, 3, 23, 11]. This can happen due to several reasons: the starting examples may not be the best ones to capture the information need (IN) of the user or the starting weights may not accurately capture the users perception or both. In this case, the user would like to refine the query and resubmit it in order ....

....and the new query is called the refined query. In a QBE environment (e.g. multimedia databases) the user typically refines the query by finding, among the answers returned to the starting query, one or more objects that are closest to what she wants and requesting for more objects like those [22, 23, 18, 3, 11]. Based on the user feedback, the system will compute the new query objects and the new weights and execute the refined query. Another way to refine the query is that the user explicitly modifies the perception model i.e. she explicitly changes the weights of the features so as to better capture ....

[Article contains additional citation context not shown here]

Y. Rui, T. Huang, and S. Mehrotra. Relevance feedback techniques in interactive content-based image retrieval. Proc. of IS&T and SPIE Storage and Retrieval of Image and Video Databases, 1998.

.... first executing a k nearest neighbor (k NN) algorithm on each individual feature 2 and then merging them to get the overall answers [17, 14, 7, 8] Due to the subjective nature of top k queries, the answers returned by the system to a user query usually do not satisfy the user s need right away [18, 3, 23, 11]. This can happen due to several reasons: the starting examples may not be the best ones to capture the information need (IN) of the user or the starting weights may not accurately capture the users perception or both. In this case, the user would like to refine the query and resubmit it in order ....

....and the new query is called the refined query. In a QBE environment (e.g. multimedia databases) the user typically refines the query by finding, among the answers returned to the starting query, one or more objects that are closest to what she wants and requesting for more objects like those [22, 23, 18, 3, 11]. Based on the user feedback, the system will compute the new query objects and the new weights and execute the refined query. Another way to refine the query is that the user explicitly modifies the perception model i.e. she explicitly changes the weights of the features so as to better capture ....

[Article contains additional citation context not shown here]

Y. Rui, T. Huang, and S. Mehrotra. Relevance feedback techniques in interactive content-based image retrieval. Proc. of IS&T and SPIE Storage and Retrieval of Image and Video Databases, 1998.

....mechanisms for users to specify relative importance (weights) among features [10, 1] Overall similarity (distance) between an object and the query image is then computed as a weighted summation of similarities (distances) over the feature set. There are several shortcomings to such an approach [23, 21]. It places too much of a burden on users to formulate their exact information need. Users may find it difficult to express their query appropriately in terms of the provided features since they may not initially have a clear idea of their information need. Furthermore, there may be a mismatch ....

....query point movement. In the query point movement approach, a query is represented by a single point in a feature space and refinement process attempts to move that point toward the direction where relevant points were located. A query point movement approach has been explored in our previous work [21] as well as in MindReader [15] While the framework described in this paper can support query point movement, we instead focus on the query expansion technique. Unlike query point movement, query expansion does not assume that a query is represented as a point in a multidimensional space. Instead, ....

Yong Rui, Thomas S. Huang, and Sharad Mehrotra. Relevance feedback techniques in interactive content-based image retrieval. In Storage and Retrieval of Images/Video Databases VI, EI'98, 1998.

....few objects that are most similar to the submitted example (e.g. top 10 images that match the query image) The task of the DBMS is to incrementally return the top matches to the user as efficiently as possible. An important aspect of multimedia similarity retrieval is that of query refinement [31, 13, 24]. Due to the subjective nature of multimedia retrieval, it is unlikely that the answers returned to the starting query will satisfy the user s need right away. Rather, among the answers returned, the user may find one or more objects that are closer to what she had mind than the original example ....

....either of the 2 following ways: ffl Query Point Movement (QPM) In this model, the objects marked relevant by the user during feedback is represented by a single point in each feature space F : the weighted centroid. Effectively, for each F , the query point is moved towards the relevant objects [30, 32, 31, 13]. ffl Query Expansion (QEX) In this model, the relevant objects are represented by multiple points (called representatives [24] in each feature space giving rise to multipoint queries [24, 25, 26] The weight (wF i s) of any representative R in the multipoint query is proportional to the ....

[Article contains additional citation context not shown here]

Y. Rui, T. Huang, and S. Mehrotra. Relevance feedback techniques in interactive content-based image retrieval. Proc. of IS&T and SPIE Storage and Retrieval of Image and Video Databases, 1998.

No context found.

Y. Rui, T. S. Huang and S. Mehrotra, "Relevance Feedback Techniques in Interactive Content-based Image Retrieval," in Proc. IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, San Jose, CA, USA, Jan. 1998.

No context found.

Rui, Y., Huang, T., and Mehrotra, S. Relevance feedback techniques in interactivecontent based image retrieval. In Proceedings of IS & TandSPIE Storage and Retrieval of Image and Video Databases (San Jose, CA, 1998), pp. 25--36.

No context found.

Y. Rui, T. S. Huang and S. Mehrotra, "Relevance Feedback Techniques in Interactive Content-based Image Retrieval," in Proc. IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, San Jose, CA, USA, Jan. 1998.

No context found.

Rui, Y., Huang, T., and Mehrotra, S. Relevance feedback techniques in interactivecontent based image retrieval. In Proceedings of IS & TandSPIE Storage and Retrieval of Image and Video Databases (San Jose, CA, 1998), pp. 25--36.

No context found.

Y. Rui, T. S. Huang, and S. Mehrotra, "Relevance feedback techniques in interactive content-based image retrieval," in Storage and Retrieval for Image and Video Databases (SPIE), pp. 25--36, 1998.

No context found.

Y. Rui, T. S. Huang, S. Mehrotra, Relevance feedback techniques in interactive content-- based image retrieval, in: I. K. Sethi, R. C. Jain (Eds.), Storage and Retrieval for Image and Video Databases VI, Vol. 3312 of SPIE Proceedings, 1997, pp. 25--36.

No context found.

Y. Rui, T.S. Huang & S. Mehrotra. Relevance feedback techniques in interactive contentbased image retrieval. In Proceedings of IS&T and SPIE Storage and Retrieval of Image and Video Databases VI, 1998.

No context found.

Y Rui, T S Huang, and S Mehrotra. Relevance feedback techniques in interactive content-based image retrieval. In Storage and Retrieval for Image and Video Databases (SPIE), pages 25--36, 1998.

No context found.

Y. Rui, T. Huang, S. Mehrotra, "Relevance Feedback Techniques in Interactive Content-Based Image Retrieval

No context found.

Y. Rui, T. S. Huang, and S. Mehrotra, "Relevance Feedback Techniques in Interactive Content-Based Image Retrieval", Proceedings of the Conference on Storage and Retrieval of Image and Video Databases VI, (1S&T/SPIE-1998), San Jose, California, Jan. 1998.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC