J. Bins B. Draper and K. Baek. Adore: Adaptive object recognition. In Christensen [12], pages 522--537.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....touch on a few representative examples of perceptual learning systems. Previous work related to specific methods and problems will be discussed in later chapters. The goal of Draper et al. s ADORE system is to learn reactive strategies for recognition of man made objects in aerial images [29]. The task is formulated as a Markov decision process, a well founded probabilistic framework in which a policy maps perceptual states to actions. In ADORE, an action is one of a set of image processing and computer vision operators. The output data produced by taking an action characterize the ....

....are clustered and represented by parametric models that define discrete states of the grasping system during its interaction with a target object. A reinforcement learning procedure is used to select appropriate closed loop grasp controllers at each of these states. Draper s ADORE system [29] is a rare example of perceptual strategy learning. The other systems cited in the previous section perform perceptual learning by subdividing the feature space spatially [114, 27, 25, 67] and or temporally [67, 22, 23] but they do not learn the features themselves. It is often argued that ....

Draper, B. A., Bins, J., and Baek, K. ADORE: Adaptive object recognition. Videre 1, 4 (2000), 86--99.

....operators that are implementable in reconfigurable hardware (FPGAs) Their program representation involves very simple bitwise operations and control flow, as one might expect when using FPGAs. Not using genetic programming, but similar in spirit to the work proposed here, Draper, Bins and Baek [3] use reinforcement learning techniques to learn good sequences of image operators for detecting houses in aerial photographs. Other AI researchers have also applied planning techniques to induce good sequences of operators for image processing tasks [9] 4 Experiments In this section, we ....

Bruce A. Draper, Jose Bins, and Kyungim Baek. ADORE: Adaptive object recognition. Videre: Journal of Computer Vision Research, 1(4), 2000.

.... of our autonomous robot, but a common knowledge based system applied simultaneously in vision and speech has already been established, as demonstrated in [18, 2] The modules of our vision software system are to some extent portable to other software systems, as the experiments on the system ADORE [3] prove, when this system was used to work with our algorithms to generate hypotheses for object recognition in an office room (see the article B. Draper et al. on Adapting Object Recognition Across Domains: A Demonstration in this volume) Currently, there has to be little time critical ....

J. Bins B. Draper and K. Baek. Adore: Adaptive object recognition. In Christensen [9], pages 52237.

....a method for detecting the most salient viewpoints of an object and used it to build an active object recognition system which moves the camera to the most discriminant viewpoint of an object in order to verify the presence of the hypothesized object. The Adore system developed by Draper et al. [2] dynamically selects vision procedures based on a Markov decision process model. Borotschnig et al. 1] combined eigenspace object representations with probability distributions to obtain classi cations that can be used as a gauge to perform view planning. While active object recognition systems ....

Bruce A. Draper, Jose Bins, and Kyungim Baek. Adore: Adaptive object recognition. In International Conference on Vision Systems, Las Palmas de Gran Canaria, Spain, 1999.

....a method for detecting the most salient viewpoints of an object and used it to build an active object recognition system which moves the camera to the most discriminant viewpoint of an object in order to verify the presence of the hypothesized object. The Adore system developed by Draper et al. [2] dynamically selects vision procedures based on a Markov decision process model. Borotschnig et al. 1] combined eigenspace object representations with probability distributions to obtain classifications that can be used as a gauge to perform view planning. While active object recognition systems ....

Bruce A. Draper, Jose Bins, and Kyungim Baek. Adore: Adaptive object recognition. In International Conference on Vision Systems, Las Palmas de Gran Canaria, Spain, 1999.

....basic shapes such as the various noteheads, stems and tails. Recently, some OMR systems have been developed that can be trained to classify new shapes, using advanced artificial intelligence and machine learning techniques such as genetic algorithms [9] neural networks and Markov models [10], 11] These techniques are generally trained o# line using test data sets and only apply to the identification stage. For example, Fujimori [12] investigated adaptive systems, using a genetic algorithm to create a classification feature set o# line, and a weighted nearest neighbour algorithm at ....

K. B. Bruce A. Draper, Jose Bins, "Adore: Adaptive object recognition," in Proceedings of the International Conference on Vision Systems, (Las Palmas de Gran Canaria, Spain), pp. 522--537, Jan 1999.

....this hypothesis given knowledge gained about this area of the page from other sources. 2. 3 Classification Algorithms for feature extraction One of the more recent developments in the field of OMR is the use of machinelearning techniques to develop shape descriptions, given a set of training data [Ala95, BAD99, SD98]. These techniques could be investigated to design feature sets for classification of musical primitives for either the current Primela framework, or some new, replacement method for di#erentiating objects. 2.4 Illustration of the Concept There is currently an existing prototype which is ....

Kyungim Baek Bruce A. Draper, Jose Bins. Adore: Adaptive object recognition. In Proceedings of the International Conference on Vision Systems, pages 522--537, Las Palmas de Gran Canaria, Spain, Jan 1999.

....data set contains images of cats and dogs, and has been previously used as a testbed to compare appearance based recognition methods [25] 4.1. Experiment #1: aerial images The first data set consist of 891 features computed over regions of interest extracted from aerial images of Fort Hood [7]. Typical features here are statistical features (mean, st. dev. etc. computed over the raw image or first or second derivatives of the raw image, and repeated at different scales. Other features include eigenprojections, probes, histograms comparisons, etc. These features match the assumptions ....

B. A. Draper, J. Bins, K. Baek. "ADORE: Adaptive Object Recognition", Videre 1(4):86-99, 2000.

No context found.

B. A. Draper, J. Bins, and K. Baek. ADORE: Adaptive Object Recognition. International Conference on Vision Systems, Las Palmas de Gran Canaria, Spain, 1999.

....while still using object, scene and domain information to direct processing. These systems use machine learning techniques to acquire information from training images. This simplifies system construction, and makes it possible to port them from domain to domain. Examples of these systems include [19 22]. For the last several years, the first author has advocated the use of Markov models for high level computer vision, with reinforcement learning as the training method [18, 23] As a prototype, ADORE (for Adaptive Object Recognition) was built and trained to find buildings in aerial imagery ....

.... For the last several years, the first author has advocated the use of Markov models for high level computer vision, with reinforcement learning as the training method [18, 23] As a prototype, ADORE (for Adaptive Object Recognition) was built and trained to find buildings in aerial imagery [19]. In this paper we demonstrate that ADORE really can be ported from one domain to another, by training it to recognize objects in a new domain (office supplies) using images and routines developed at another university (Erlangen Nrnberg) The system was ported by two people in the span of one ....

[Article contains additional citation context not shown here]

Draper, B.A., J. Bins, and K. Baek, ADORE: Adaptive Object Recognition. Videre, 2000. 1(4): p. 86-99.

....viewed as updated versions of the same basic idea. This paper tries to synthesize Kosslyn s iconic recognition theory with the purposive approach. In particular, it builds on the author s previous work on using reinforcement learning to acquire purposive, multi stage object recognition strategies [4]. Unlike in previous work, however, this time we assume that memory is iconic and that object recognition is therefore an image matching task. We then use the match score between the stored image (memory) and the sensed image (input) as a reward signal for optimizing the recognition process. In ....

....that learns control strategies for object recognition from training samples. The system, called ADORE, formalized the object recognition control problem as a Markov decision problem, and used reinforcement learning to develop nearly optimal control policies for recognizing houses in aerial images [4]. Unfortunately, the use of this system in practice has been hindered by the need to provide large numbers of hand labeled training images. Kosslyn s theory suggests that hand labeled training images may not be necessary. By using the result of image matching as the training signal, we can move ....

B. A. Draper, J. Bins, and K. Baek. ADORE: Adaptive Object Recognition. International Conference on Vision Systems, Las Palmas de Gran Canaria, Spain, 1999.

No context found.

J. Bins B. Draper and K. Baek. Adore: Adaptive object recognition. In Christensen [12], pages 522--537.

No context found.

J. Bins B. Draper and K. Baek. Adore: Adaptive object recognition. In Videre, 1(4), pages 86--99, 2000.

No context found.

B. Draper, J. Bins, and K. Baek, "ADORE: Adaptive object recognition," in Proc. International Conference on Vision Systems, pp. 522--537, (Las Palmas de Gran Canaria, Spain), Jan. 1999.

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