http://hdl.handle.net/2117/3335 Sat, 25 May 2013 04:49:32 GMT 2013-05-25T04:49:32Z webmaster.bupc@upc.edu Universitat Politècnica de Catalunya. Servei de Biblioteques i Documentació no http://hdl.handle.net/2117/19261 Title: Depth estimation of frames in image sequences using motion occlusions Authors: Palou Visa, Guillem; Salembier Clairon, Philippe Jean Abstract: This paper proposes a system to depth order regions of a frame belonging to a monocular image sequence. For a given frame, re- gions are ordered according to their relative depth using the previous and following frames. The algorithm estimates occluded and disoccluded pixels belonging to the central frame. Afterwards, a Binary Partition Tree (BPT) is constructed to obtain a hierarchical, region based repre- sentation of the image. The nal depth partition is obtained by means of energy minimization on the BPT. To achieve a global depth ordering from local occlusion cues, a depth order graph is constructed and used to eliminate contradictory local cues. Results of the system are evaluated and compared with state of the art gure/ground labeling systems on several datasets, showing promising results. Wed, 15 May 2013 15:11:26 GMT http://hdl.handle.net/2117/19261 2013-05-15T15:11:26Z Palou Visa, Guillem; Salembier Clairon, Philippe Jean no Binary partition Tree, Data sets, Depth Estimation, Energy minimization, Image sequence, Monocular image sequence, Order graph, Region-based, State of the art This paper proposes a system to depth order regions of a frame belonging to a monocular image sequence. For a given frame, re- gions are ordered according to their relative depth using the previous and following frames. The algorithm estimates occluded and disoccluded pixels belonging to the central frame. Afterwards, a Binary Partition Tree (BPT) is constructed to obtain a hierarchical, region based repre- sentation of the image. The nal depth partition is obtained by means of energy minimization on the BPT. To achieve a global depth ordering from local occlusion cues, a depth order graph is constructed and used to eliminate contradictory local cues. Results of the system are evaluated and compared with state of the art gure/ground labeling systems on several datasets, showing promising results. http://hdl.handle.net/2117/18542 Title: Can our TV robustly understand human gestures? Real-Time Gesture Localization in Range Data Authors: López Méndez, Adolfo; Casas Pla, Josep Ramon Abstract: The 'old' remote falls short of requirements when confronted with digital convergence for living room displays. Enriched options to watch, manage and interact with content on large displays demand improved means of interaction. Concurrently, gesture recognition is increasingly present in human-computer interaction for gaming applications. In this paper we propose a gesture localization framework for interactive display of audio-visual content. The proposed framework works with range data captured from a single consumer depth camera. We focus on still gestures because they are generally user friendly (users do not have to make complex and tiring movements) and allow formulating the problem in terms of object localization. Our method is based on random forests, which have shown an excellent performance on classification and regression tasks. In this work, however, we aim at a specific class of localization problems involving highly unbalanced data: positive examples appear during a small fraction of space and time. We study the impact of this natural unbalance on the random forest learning and we propose a framework to robustly detect gestures on range images in real applications. Our experiments with offline data show the effectiveness of our approach. We also present a real-time application where users can control the TV display with a reduced set of still gestures. Tue, 02 Apr 2013 12:43:38 GMT http://hdl.handle.net/2117/18542 2013-04-02T12:43:38Z López Méndez, Adolfo; Casas Pla, Josep Ramon no The 'old' remote falls short of requirements when confronted with digital convergence for living room displays. Enriched options to watch, manage and interact with content on large displays demand improved means of interaction. Concurrently, gesture recognition is increasingly present in human-computer interaction for gaming applications. In this paper we propose a gesture localization framework for interactive display of audio-visual content. The proposed framework works with range data captured from a single consumer depth camera. We focus on still gestures because they are generally user friendly (users do not have to make complex and tiring movements) and allow formulating the problem in terms of object localization. Our method is based on random forests, which have shown an excellent performance on classification and regression tasks. In this work, however, we aim at a specific class of localization problems involving highly unbalanced data: positive examples appear during a small fraction of space and time. We study the impact of this natural unbalance on the random forest learning and we propose a framework to robustly detect gestures on range images in real applications. Our experiments with offline data show the effectiveness of our approach. We also present a real-time application where users can control the TV display with a reduced set of still gestures. http://hdl.handle.net/2117/18496 Title: Metric learning from poses for temporal clustering of human motion Authors: López Méndez, Adolfo; Gall, Juergen; Casas Pla, Josep Ramon; van Gool, Luc Abstract: Temporal clustering of human motion into semantically meaningful behaviors is a challenging task. While unsupervised methods do well to some extent, the obtained clusters often lack a semantic interpretation. In this paper, we propose to learn what makes a sequence of human poses different from others such that it should be annotated as an action. To this end, we formulate the problem as weakly supervised temporal clustering for an unknown number of clusters. Weak supervision is attained by learning a metric from the implicit semantic distances derived from already annotated databases. Such a metric contains some low-level semantic information that can be used to effectively segment a human motion sequence into distinct actions or behaviors. The main advantage of our approach is that metrics can be successfully used across datasets, making our method a compelling alternative to unsupervised methods. Experiments on publicly available mocap datasets show the effectiveness of our approach. Fri, 22 Mar 2013 13:04:35 GMT http://hdl.handle.net/2117/18496 2013-03-22T13:04:35Z López Méndez, Adolfo; Gall, Juergen; Casas Pla, Josep Ramon; van Gool, Luc no Temporal clustering of human motion into semantically meaningful behaviors is a challenging task. While unsupervised methods do well to some extent, the obtained clusters often lack a semantic interpretation. In this paper, we propose to learn what makes a sequence of human poses different from others such that it should be annotated as an action. To this end, we formulate the problem as weakly supervised temporal clustering for an unknown number of clusters. Weak supervision is attained by learning a metric from the implicit semantic distances derived from already annotated databases. Such a metric contains some low-level semantic information that can be used to effectively segment a human motion sequence into distinct actions or behaviors. The main advantage of our approach is that metrics can be successfully used across datasets, making our method a compelling alternative to unsupervised methods. Experiments on publicly available mocap datasets show the effectiveness of our approach. http://hdl.handle.net/2117/18356 Title: Depth ordering on image sequences using motion occlusions Authors: Palou Visa, Guillem; Salembier Clairon, Philippe Jean Abstract: This paper proposes a system to obtain the depth order of frames in image sequences using motion occlusion cues. The system first computes the forward and backward flows with the previous and next frames and estimates the occluded points. To obtain a region representation of the image, a Binary Partition Tree (BPT) is created for each frame. To estimate occlusion relations in the image, projective flow models are fitted to all regions in the image. The depth order solution is obtained by minimizing over the tree structure a cost function based on occlusion relations and the number of regions. Results show that optical flow algorithms can be used directly to estimate occlusion points. Promising results are obtained combining motion occlusions and region information by means of a BPT. Evaluation is performed comparing current state-of-the-art algorithms on figure/ground assignments, showing that the performance of the proposed system is comparable to current algorithms. Fri, 15 Mar 2013 16:02:14 GMT http://hdl.handle.net/2117/18356 2013-03-15T16:02:14Z Palou Visa, Guillem; Salembier Clairon, Philippe Jean no This paper proposes a system to obtain the depth order of frames in image sequences using motion occlusion cues. The system first computes the forward and backward flows with the previous and next frames and estimates the occluded points. To obtain a region representation of the image, a Binary Partition Tree (BPT) is created for each frame. To estimate occlusion relations in the image, projective flow models are fitted to all regions in the image. The depth order solution is obtained by minimizing over the tree structure a cost function based on occlusion relations and the number of regions. Results show that optical flow algorithms can be used directly to estimate occlusion points. Promising results are obtained combining motion occlusions and region information by means of a BPT. Evaluation is performed comparing current state-of-the-art algorithms on figure/ground assignments, showing that the performance of the proposed system is comparable to current algorithms. http://hdl.handle.net/2117/18320 Title: Depth map coding based on a optimal hierarchical region representation Authors: Maceira Duch, Marc; Ruiz Hidalgo, Javier; Morros Rubió, Josep Ramon Abstract: Multiview color information used jointly with depth maps is a widespread technique for 3D video. Using this depth information, 3D functionalities such as free view point video can be provided by means of depth-image-based rendering techniques. In this pa- per, a new technique to encode depth maps is proposed. Based on the usually smooth structure and the sharp edges of depth map, our proposal segments the depth map into homogeneous regions of ar- bitrary shape and encodes the contents of these regions using dif- ferent texture coding strategies. An optimal lagrangian approach is applied to the hierarchical region representation provided by our segmentation technique. This approach automatically selects the best encoding strategy for each region and the optimal partition to encode the depth map. To avoid the high coding costs of coding the resulting partition, a prediction is made using the associated decoded color image Thu, 14 Mar 2013 17:59:25 GMT http://hdl.handle.net/2117/18320 2013-03-14T17:59:25Z Maceira Duch, Marc; Ruiz Hidalgo, Javier; Morros Rubió, Josep Ramon no Multiview color information used jointly with depth maps is a widespread technique for 3D video. Using this depth information, 3D functionalities such as free view point video can be provided by means of depth-image-based rendering techniques. In this pa- per, a new technique to encode depth maps is proposed. Based on the usually smooth structure and the sharp edges of depth map, our proposal segments the depth map into homogeneous regions of ar- bitrary shape and encodes the contents of these regions using dif- ferent texture coding strategies. An optimal lagrangian approach is applied to the hierarchical region representation provided by our segmentation technique. This approach automatically selects the best encoding strategy for each region and the optimal partition to encode the depth map. To avoid the high coding costs of coding the resulting partition, a prediction is made using the associated decoded color image http://hdl.handle.net/2117/18317 Title: From local occlusion cues to global monocular depth estimation Authors: Palou Visa, Guillem; Salembier Clairon, Philippe Jean Abstract: In this paper, we propose a system to obtain a depth ordered seg- mentation of a single image based on low level cues. The algorithm first constructs a hierarchical, region-based image representation of the image using a Binary Partition Tree (BPT). During the building process, T-junction depth cues are detected, along with high convex boundaries. When the BPT is built, a suitable segmentation is found and a global depth ordering is found using a probabilistic framework. Results are compared with state of the art depth ordering and figure/ground labeling systems. The advantage of the proposed ap- proach compared to systems based on a training procedure is the lack of assumptions about the scene content. Moreover, it is shown that the system outperforms previously low-level cue based systems, while offering similar results to a priori trained figure/ground label- ing algorithms Thu, 14 Mar 2013 17:28:13 GMT http://hdl.handle.net/2117/18317 2013-03-14T17:28:13Z Palou Visa, Guillem; Salembier Clairon, Philippe Jean no In this paper, we propose a system to obtain a depth ordered seg- mentation of a single image based on low level cues. The algorithm first constructs a hierarchical, region-based image representation of the image using a Binary Partition Tree (BPT). During the building process, T-junction depth cues are detected, along with high convex boundaries. When the BPT is built, a suitable segmentation is found and a global depth ordering is found using a probabilistic framework. Results are compared with state of the art depth ordering and figure/ground labeling systems. The advantage of the proposed ap- proach compared to systems based on a training procedure is the lack of assumptions about the scene content. Moreover, it is shown that the system outperforms previously low-level cue based systems, while offering similar results to a priori trained figure/ground label- ing algorithms http://hdl.handle.net/2117/18291 Title: Microarray classification with hierarchical data representation and novel feature selection criteria Authors: Bosio, Mattia; Bellot Pujalte, Pau; Salembier Clairon, Philippe Jean; Oliveras Vergés, Albert Abstract: Microarray data classification is a challenging prob- lem due to the high number of variables compared to the small number of available samples. An effective methodology to output a precise and reliable classifier is proposed in this work as an improvement of the algorithm in [1]. It considers the sample scarcity problem and the lack of data structure typical of microarrays. Both problem are assessed by a two-step approach applying hierarchical clustering to create new features called metagenes and introducing a novel feature ranking criterion, inside the wrapper feature selection task. The classification ability has been evaluated on 4 publicly available datasets from Micro Array Quality Control study phase II (MAQC) classified by 7 different endpoints. The global results have showed how the proposed approach obtains better prediction accuracy than a wide variety of state of the art alternatives Wed, 13 Mar 2013 19:48:02 GMT http://hdl.handle.net/2117/18291 2013-03-13T19:48:02Z Bosio, Mattia; Bellot Pujalte, Pau; Salembier Clairon, Philippe Jean; Oliveras Vergés, Albert no Microarray data classification is a challenging prob- lem due to the high number of variables compared to the small number of available samples. An effective methodology to output a precise and reliable classifier is proposed in this work as an improvement of the algorithm in [1]. It considers the sample scarcity problem and the lack of data structure typical of microarrays. Both problem are assessed by a two-step approach applying hierarchical clustering to create new features called metagenes and introducing a novel feature ranking criterion, inside the wrapper feature selection task. The classification ability has been evaluated on 4 publicly available datasets from Micro Array Quality Control study phase II (MAQC) classified by 7 different endpoints. The global results have showed how the proposed approach obtains better prediction accuracy than a wide variety of state of the art alternatives http://hdl.handle.net/2117/18278 Title: INTAIRACT: Joint hand gesture and fingertip classification for touchless interaction Authors: Suau Cuadros, Xavier; Alcoverro Vidal, Marcel; López Méndez, Adolfo; Ruiz Hidalgo, Javier; Casas Pla, Josep Ramon Abstract: In this demo we present intAIRact, an online hand-based touchless interaction system. Interactions are based on easy-to-learn hand gestures, that combined with translations and rotations render a user friendly and highly configurable system. The main advantage with respect to existing approaches is that we are able to robustly locate and identify fingertips. Hence, we are able to employ a simple but powerful alphabet of gestures not only by determining the number of visible fingers in a gesture, but also which fingers are being observed. To achieve such a system we propose a novel method that jointly infers hand gestures and fingertip locations using a single depth image from a consumer depth camera. Our approach is based on a novel descriptor for depth data, the Oriented Radial Distribution (ORD) [1]. On the one hand, we exploit the ORD for robust classification of hand gestures by means of efficient k-NN retrieval. On the other hand, maxima of the ORD are used to perform structured inference of fingertip locations. The proposed method outperforms other state-of-the-art approaches both in gesture recognition and fingertip localization. An implementation of the ORD extraction on a GPU yields a real-time demo running at approximately 17fps on a single laptop Wed, 13 Mar 2013 16:13:16 GMT http://hdl.handle.net/2117/18278 2013-03-13T16:13:16Z Suau Cuadros, Xavier; Alcoverro Vidal, Marcel; López Méndez, Adolfo; Ruiz Hidalgo, Javier; Casas Pla, Josep Ramon no In this demo we present intAIRact, an online hand-based touchless interaction system. Interactions are based on easy-to-learn hand gestures, that combined with translations and rotations render a user friendly and highly configurable system. The main advantage with respect to existing approaches is that we are able to robustly locate and identify fingertips. Hence, we are able to employ a simple but powerful alphabet of gestures not only by determining the number of visible fingers in a gesture, but also which fingers are being observed. To achieve such a system we propose a novel method that jointly infers hand gestures and fingertip locations using a single depth image from a consumer depth camera. Our approach is based on a novel descriptor for depth data, the Oriented Radial Distribution (ORD) [1]. On the one hand, we exploit the ORD for robust classification of hand gestures by means of efficient k-NN retrieval. On the other hand, maxima of the ORD are used to perform structured inference of fingertip locations. The proposed method outperforms other state-of-the-art approaches both in gesture recognition and fingertip localization. An implementation of the ORD extraction on a GPU yields a real-time demo running at approximately 17fps on a single laptop http://hdl.handle.net/2117/18063 Title: Rich internet application for semi-automatic annotation of semantic shots on keyframes Authors: Carcel, Elisabet; Martos Asensio, Manel; Giró Nieto, Xavier; Marqués Acosta, Fernando Abstract: This paper describes a system developed for the semi-automatic annotation of keyframes in a broadcasting company. The tool aims at assisting archivists who traditionally label every keyframe manually by suggesting them an automatic annotation that they can intuitively edit and validate. The system is valid for any domain as it uses generic MPEG-7 visual descriptors and binary SVM classifiers. The classification engine has been tested on the multiclass problem of semantic shot detection, a type of metadata used in the company to index new content ingested in the system. The detection performance has been tested in two different domains: soccer and parliament. The core engine is accessed by a Rich Internet Application via a web service. The graphical user interface allows the edition of the suggested labels with an intuitive drag and drop mechanism between rows of thumbnails, each row representing a different semantic shot class. The system has been described as complete and easy to use by the professional archivists at the company Tue, 05 Mar 2013 11:49:33 GMT http://hdl.handle.net/2117/18063 2013-03-05T11:49:33Z Carcel, Elisabet; Martos Asensio, Manel; Giró Nieto, Xavier; Marqués Acosta, Fernando no annotation, classification, MPEG-7 visual descriptors, RIA, semantic shot This paper describes a system developed for the semi-automatic annotation of keyframes in a broadcasting company. The tool aims at assisting archivists who traditionally label every keyframe manually by suggesting them an automatic annotation that they can intuitively edit and validate. The system is valid for any domain as it uses generic MPEG-7 visual descriptors and binary SVM classifiers. The classification engine has been tested on the multiclass problem of semantic shot detection, a type of metadata used in the company to index new content ingested in the system. The detection performance has been tested in two different domains: soccer and parliament. The core engine is accessed by a Rich Internet Application via a web service. The graphical user interface allows the edition of the suggested labels with an intuitive drag and drop mechanism between rows of thumbnails, each row representing a different semantic shot class. The system has been described as complete and easy to use by the professional archivists at the company http://hdl.handle.net/2117/17826 Title: Temporal PolSAR image series exploitation with binary partition trees Authors: Alonso González, Alberto; López Martínez, Carlos; Salembier Clairon, Philippe Jean Abstract: In this paper, the processing of temporal PolSAR image series is addressed through a region-based and multi-scale data representation, the Binary Partition Tree (BPT). This structure contains useful information related to the data structure at different detail levels that may be employed for different applications. The construction of this structure ans its exploitation is addressed in this work in the context of the speckle filtering and data segmentation applications. A new region model and processing strategy are defined to tackle with the temporal dimension of the data. Finally, to illustrate the capabilities of the proposed technique, results are shown with a real RADARSAT-2 dataset. Mon, 18 Feb 2013 12:44:39 GMT http://hdl.handle.net/2117/17826 2013-02-18T12:44:39Z Alonso González, Alberto; López Martínez, Carlos; Salembier Clairon, Philippe Jean no In this paper, the processing of temporal PolSAR image series is addressed through a region-based and multi-scale data representation, the Binary Partition Tree (BPT). This structure contains useful information related to the data structure at different detail levels that may be employed for different applications. The construction of this structure ans its exploitation is addressed in this work in the context of the speckle filtering and data segmentation applications. A new region model and processing strategy are defined to tackle with the temporal dimension of the data. Finally, to illustrate the capabilities of the proposed technique, results are shown with a real RADARSAT-2 dataset. http://hdl.handle.net/2117/17818 Title: Variable local weight filtering for PolSAR data speckle noise reduction Authors: Alonso González, Alberto; López Martínez, Carlos; Salembier Clairon, Philippe Jean Abstract: This paper presents a Polarimetric SAR data speckle filtering technique, based on a combined filtering in the spatial and polarimetric domains. It is based on a bilateral filtering employing distance measures over these domains. These measures concentrate all the information related to the domain structure that is needed for an adaptation to the scene morphology. A weighted average is performed over a given window favoring closer and similar pixels. As a consequence, an adaptive filtering is achieved, attaining higher filtering over homogeneous areas whereas point scatters remain almost unchanged. Results will be shown over a real RADARSAT-2 data. Mon, 18 Feb 2013 10:53:28 GMT http://hdl.handle.net/2117/17818 2013-02-18T10:53:28Z Alonso González, Alberto; López Martínez, Carlos; Salembier Clairon, Philippe Jean no This paper presents a Polarimetric SAR data speckle filtering technique, based on a combined filtering in the spatial and polarimetric domains. It is based on a bilateral filtering employing distance measures over these domains. These measures concentrate all the information related to the domain structure that is needed for an adaptation to the scene morphology. A weighted average is performed over a given window favoring closer and similar pixels. As a consequence, an adaptive filtering is achieved, attaining higher filtering over homogeneous areas whereas point scatters remain almost unchanged. Results will be shown over a real RADARSAT-2 data. http://hdl.handle.net/2117/17456 Title: Multi-view body tracking with a detector-driven hierarchical particle filter Authors: Navarro, S.; López Méndez, Adolfo; Alcoverro Vidal, Marcel; Casas Pla, Josep Ramon Abstract: In this paper we present a novel approach to markerless human motion capture that robustly integrates body part detections in multiple views. The proposed method fuses cues from multiple views to enhance the propagation and observation model of particle filtering methods aiming at human motion capture. We particularize our method to improve arm tracking in the publicly available IXMAS dataset. Our experiments show that the proposed method outperforms other state-ofthe- art approaches. Mon, 21 Jan 2013 17:47:42 GMT http://hdl.handle.net/2117/17456 2013-01-21T17:47:42Z Navarro, S.; López Méndez, Adolfo; Alcoverro Vidal, Marcel; Casas Pla, Josep Ramon no 3D reconstruction, Body part detection, Human motion capture, Inverse kinematics, Multi-view In this paper we present a novel approach to markerless human motion capture that robustly integrates body part detections in multiple views. The proposed method fuses cues from multiple views to enhance the propagation and observation model of particle filtering methods aiming at human motion capture. We particularize our method to improve arm tracking in the publicly available IXMAS dataset. Our experiments show that the proposed method outperforms other state-ofthe- art approaches. http://hdl.handle.net/2117/17275 Title: Promeds: an adaptive robust fundamental matrix estimation approach Authors: Irurueta Carro, Alberto; Morros Rubió, Josep Ramon Abstract: Accurate fundamental matrix estimation from computed correspondences is hard to achieve depending on the constraints on computational time and available data (i.e. correspondences and quality scores). Several algorithms exist for this task, like the 8-points, the 7-points algorithm [1] or robust methods such as RANSAC [2], MSAC [3] or LMedS [4]. Robust methods are capable of discriminating correspondence outliers, thus, obtaining better results. Additionally, some variations of the previous methods have been proposed. For instance PROSAC [5] is an improvement of RANSAC which takes into account additional information of the quality of the matches to largely reduce the computational cost of the fundamental matrix estimation process. This work proposes a new robust method for fundamental matrix estimation that combines the benefits of PROSAC and LMedS algorithms, namely improved quality, reduced computational time and less parameters to adjust Fri, 11 Jan 2013 11:49:11 GMT http://hdl.handle.net/2117/17275 2013-01-11T11:49:11Z Irurueta Carro, Alberto; Morros Rubió, Josep Ramon no Accurate fundamental matrix estimation from computed correspondences is hard to achieve depending on the constraints on computational time and available data (i.e. correspondences and quality scores). Several algorithms exist for this task, like the 8-points, the 7-points algorithm [1] or robust methods such as RANSAC [2], MSAC [3] or LMedS [4]. Robust methods are capable of discriminating correspondence outliers, thus, obtaining better results. Additionally, some variations of the previous methods have been proposed. For instance PROSAC [5] is an improvement of RANSAC which takes into account additional information of the quality of the matches to largely reduce the computational cost of the fundamental matrix estimation process. This work proposes a new robust method for fundamental matrix estimation that combines the benefits of PROSAC and LMedS algorithms, namely improved quality, reduced computational time and less parameters to adjust http://hdl.handle.net/2117/16580 Title: Oriented radial distribution on depth data: application to the detection of end-effectors Authors: Suau Cuadros, Xavier; Ruiz Hidalgo, Javier; Casas Pla, Josep Ramon Description: End-effectors are considered to be the main topological extremities of a given 3D body. Even if the nature of such body is not restricted, this paper focuses on the human body case. Detection of human extremities is a key issue in the human motion capture domain, being needed to initialize and update the tracker. Therefore, the effectiveness of human motion capture systems usually depends on the reliability of the obtained end-effectors. The increasing accuracy, low cost and easy installation of depth cameras has opened the door to new strategies to overcome the body pose estimation problem. With the objective of detecting the head, hands and feet of a human body, we propose a new local feature computed from depth data, which gives an idea of its curvature and prominence. Such feature is weighted depending on recent detections, providing also a temporal dimension. Based on this feature, some end-effector candidate blobs are obtained and classified into head, hands and feet according to three probabilistic descriptors. Wed, 26 Sep 2012 10:23:04 GMT http://hdl.handle.net/2117/16580 2012-09-26T10:23:04Z Suau Cuadros, Xavier; Ruiz Hidalgo, Javier; Casas Pla, Josep Ramon no http://hdl.handle.net/2117/16487 Title: Upper-bound assessment of the spatial accuracy of hierarchical region-based image representations Authors: Pont Tuset, Jordi; Marqués Acosta, Fernando Abstract: Hierarchical region-based image representations are versatile tools for segmentation, filtering, object detection, etc. The evaluation of their spatial accuracy has been usually performed assessing the final result of an algorithm based on this representation. Given its wide applicability, however, a direct supervised assessment, independent of any application, would be desirable and fair. A brute-force assessment of all the partitions represented in the hierarchical structure would be a correct approach, but as we prove formally, it is computationally unfeasible. This paper presents an efficient algorithm to find the upper-bound performance of the representation and we show that the previous approximations in the literature can fail at finding this bound. Fri, 14 Sep 2012 09:22:51 GMT http://hdl.handle.net/2117/16487 2012-09-14T09:22:51Z Pont Tuset, Jordi; Marqués Acosta, Fernando no Hierarchical region-based image representations are versatile tools for segmentation, filtering, object detection, etc. The evaluation of their spatial accuracy has been usually performed assessing the final result of an algorithm based on this representation. Given its wide applicability, however, a direct supervised assessment, independent of any application, would be desirable and fair. A brute-force assessment of all the partitions represented in the hierarchical structure would be a correct approach, but as we prove formally, it is computationally unfeasible. This paper presents an efficient algorithm to find the upper-bound performance of the representation and we show that the previous approximations in the literature can fail at finding this bound.