T. Minka and R. Picard. Interactive learning with a `society of models. Pattern Recognition, 30(4):565--81, April 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the assumed hypothesis space p(M i ) p(M i ) shows how plausible we thought the alternative models were before the data arrived. Inference of probability distributions from observation of sensory data aims at finding the best stochastic models able to consistently characterize classes of images [3,10]. The Bayesian approach for data modelling is used. The information contained in a data set (provided by a unique sensor) is extracted in different assumptions. The assumptions are represented by different prior models (Fig. 1) In the case of a multispectral sensor the assumed prior models can ....

Minka T.P.,.Picard R.W (1997) Interactive learning with a society of models. Pattern Recognition, vol. 30, pp.565--581.

....to obtain characteristic signal classes, and finally present these to be utilized by a user in defining his specific interests. With this hierarchical procedure consisting of a slow, unsupervised clustering step, and a fast, userinteractive learning stage, the setup resembles the FourEyes system [12]. However, whereas in the original idea a self organizing map is used for modeling a weight space, we use di#erent Bayesian networks [13] to directly link user interests and signal classes. This provides the user of the image archive with a probabilistic classification of the content of each ....

....Hierarchical Modeling of Image Content We arrange the information on five levels of di#erent semantic abstraction as depicted in Fig. 1. The application of a family of signal models and the separation into unsupervised clustering and supervised learning resemble very much the FourEyes system [12]. However, note that from each level to the next we perform a step of Bayesian inference. This allows us to calculate on each level the most probable content description given the descriptions on the levels below. On the lowest level, the image data (Level 0) we apply stochastic models M in ....

T. P. Minka and R. W. Picard, "Interactive learning with a 'society of models'," Pattern Recognition, vol. 30, no. 4, pp. 565--581, 1997.

....is performed either globally, using techniques such as histogram matching and color layout indexing, or locally, based on decomposed regions (objects) of the images. There is a rich resource of prior work on this subject [2] 3] 4] 5] 7] 9] 10] 13] 14] 15] 16] 17] 19] [20], 21] 22] 24] 25] 27] 28] 29] 31] 32] 34] Due to limited space, we only review work most related to ours, which by no means represents the complete set. As a relatively mature method, histogram matching has been applied to many general purpose image retrieval Y. Chen is with ....

....color variations over the spatial extent of an image by Daubechies wavelet coefficients (in the lowest few frequency bands) and their variances. Schmid and Mohr [24] propose a method of indexing images based on local features of automatically detected interest points of images. Minka and Picard [20] describe a learning algorithm for selecting and grouping features. The user guides the learning process by providing positive and negative examples. The approach presented in [29] uses what is called the Most Discriminating Features for image retrieval. These features are extracted from a set of ....

T.P. Minka and R.W. Picard, "interactive Learning with a 'Society of Models'," Pattern Recognition, vol. 30, no. 4, pp. 565-581, 1997.

....similar images are retrieved based on low level image features. Therefore, the retrieval performance (usually in terms of precision and recall) of a CBIR system is severely limited when the sample image does not describe the user s need precisely. To overcome this limitation, many CBIR systems [1,3,4,5,6,7] have applied the relevance feedback technique, which improves the retrieval performance by adjusting the original query based on the relevant and irrelevant image examples designated by users. Although the current feedback technique has been proved effective in boosting the retrieval ....

....metric that best describes the desired images in the feature space. Therefore, if the desired images cannot be sufficiently described by low level features, they fail to return many relevant results even with a large number of feedbacks. On the other hand, most feedback approaches, except a few [3,4,5], do not have a learning mechanism to memorize the feedbacks conducted previously and reuse them in favor of future queries. If we define a retrieval session as a user query and its subsequent feedback process, most approaches can only improve the retrieval results within a single session ....

[Article contains additional citation context not shown here]

T.P. Minka and R.W. Picard, Interactive learning with a society of models , Proc. of IEEE CVPR, pp. 447-452, 1996.

....the image contents. 31] explores a heterogeneous clustering methodology that overcomes the drawback of single feature matching when dealing with images that are considered similar by computation but actually having different semantics. Approaches that depend on some form of user interaction are [7, 21, 29]. Mediated by user interaction, the system discussed in [7] defines a set of queries that correspond to a user concept. 21] is a system that learns how to combine various features in the overall retrieval process through user feedback. 29] introduces an exploration paradigm based on an advanced ....

....dealing with images that are considered similar by computation but actually having different semantics. Approaches that depend on some form of user interaction are [7, 21, 29] Mediated by user interaction, the system discussed in [7] defines a set of queries that correspond to a user concept. [21] is a system that learns how to combine various features in the overall retrieval process through user feedback. 29] introduces an exploration paradigm based on an advanced user interface simulating 3 D space. In this space, thumbnail images having the same user semantics are displayed close to ....

T. P. Minka and R. W. Picard, Interactive Learning with a Society of Models, Pattern Recognition, Volume 30, Number 4, 1997, pp. 565-581.

No context found.

T. Minka and R. Picard. Interactive learning with a `society of models. Pattern Recognition, 30(4):565--81, April 1997.

No context found.

T.P. Minka, R.W. Picard, Interactive learning with a &Society of Models, Pattern Recognition 30 (4) (1997).

No context found.

T.P. Minka and R.W. Picard, "Interactive Learning with a `Society of Models'," Pattern Recognition, vol. 30, no. 4, pp. 565--581, 1997.

No context found.

Minka TP, Picard RW (1997) Interactive learning with a society of models. Patt Recog 30:565--581

No context found.

T.P. Minka and R.W. Picard, \Interactive Learning with a \Society of Models"", Pattern Recognition, vol.30, (no.4):565-81, April 1997.

No context found.

T. Minka and R. Picard. Interactive learning with a `society of models. Pattern Recognition, 30(4):565--81, April 1997.

No context found.

T. P. Minka and R. W. Picard, \Interactive learning with a society of models," Pattern Recognition 30(4), pp. 565-581, 1997.

No context found.

T. P. Minka and R. W. Picard, \Interactive Learning with a Society of Models," Proceedings of CVPR-96, pp. 447-452, (1996).

No context found.

T.P. Minka and R.W. Picard. \Interactive Learning With A 'Society of Models'," Pattern Recognition, 3  0, pp 565-581, 1997.

No context found.

T. P. Minka and R. W. Picard, "Interactive Learning with a Society of Models," Proceedings of CVPR-96, pp. 447-452, (1996).

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