Limits...
Perceptual influence of elementary three-dimensional geometry: (2) fundamental object parts.

Tamosiunaite M, Sutterlütti RM, Stein SC, Wörgötter F - Front Psychol (2015)

Bottom Line: Additionally we control against segmentation reliability and we find a clear trend that reliable convex segments have a high degree of name-ability.In addition, we observed that using other image-segmentation methods will not yield nameable entities.This indicates that convex-concave surface transition may indeed form the basis for dividing objects into meaningful entities.

View Article: PubMed Central - PubMed

Affiliation: Faculty of Physics - Biophysics and Bernstein Center for Computational Neuroscience, University of Göttingen Göttingen, Germany ; Department of Informatics, Vytautas Magnus University Kaunas, Lithuania.

ABSTRACT
Objects usually consist of parts and the question arises whether there are perceptual features which allow breaking down an object into its fundamental parts without any additional (e.g., functional) information. As in the first paper of this sequence, we focus on the division of our world along convex to concave surface transitions. Here we are using machine vision to produce convex segments from 3D-scenes. We assume that a fundamental part is one, which we can easily name while at the same time there is no natural subdivision possible into smaller parts. Hence in this experiment we presented the computer vision generated segments to our participants and asked whether they can identify and name them. Additionally we control against segmentation reliability and we find a clear trend that reliable convex segments have a high degree of name-ability. In addition, we observed that using other image-segmentation methods will not yield nameable entities. This indicates that convex-concave surface transition may indeed form the basis for dividing objects into meaningful entities. It appears that other or further subdivisions do not carry such a strong semantical link to our everyday language as there are no names for them.

No MeSH data available.


Related in: MedlinePlus

Overview of the computer vision method use for convex 3D-scene segmentation (for details see Stein et al., 2014a,b). (A) Two test objects (B1) Initial point clouds are reduced to supervoxels (Papon et al., 2013) with graph edges showing how voxels are neighbors. (B2) Conventional definition of convex and concave configurations. (B3) Singular locations like the one shown are not treated as concave, which massively improves algorithmic performance (C) Resulting convex (black) and concave (red) connectivity graph. (D1) Segmentation. (D2) Noise reduction mechanisms avoiding over-smoothing inside corners. (E) Final segmentation. Figure modified from Stein et al. (2014b).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4585234&req=5

Figure 1: Overview of the computer vision method use for convex 3D-scene segmentation (for details see Stein et al., 2014a,b). (A) Two test objects (B1) Initial point clouds are reduced to supervoxels (Papon et al., 2013) with graph edges showing how voxels are neighbors. (B2) Conventional definition of convex and concave configurations. (B3) Singular locations like the one shown are not treated as concave, which massively improves algorithmic performance (C) Resulting convex (black) and concave (red) connectivity graph. (D1) Segmentation. (D2) Noise reduction mechanisms avoiding over-smoothing inside corners. (E) Final segmentation. Figure modified from Stein et al. (2014b).

Mentions: We segmented the scenes along convex-concave transitions in the 3D-data by a machine vision algorithm. Figure 1 provides an overview of this method shown by ways of two simple test objects (Figure 1A). Point clouds are first reduced to few so-called supervoxels (Papon et al., 2013) which capture the scene geometry by their neighborhood relations (graph-edges in Figure 1B1). Convex and concave edge configuration are found using a conventional criterium (Figure 1B2) employed at the surface normals of each point but corrected against singularities as shown in Figure 1B3. (Some surface normals are shown graphically in Figure 1D2 by ways of arrows.) This results is convex (black) and concave (red) connections (Figure 1C), which are used to break up the scene (Figure 1D1). Corners such as the one shown in Figure D2 lead to an over-smoothing of the normals (see red arrow) and the algorithm at the end corrects for this leading to the final segmentation as shown in Figure 1E. Details of the algorithm are described elsewhere (Stein et al., 2014a,b). Note, this is a model-free, purely data-driven segmentation algorithm, as required for the purpose of this study, which does not use any additional features for segmentation. Due to the limited spatial resolution of the RGB-D sensors, small objects cannot be consistently labeled. Thus, segments smaller than 0.3% of the image size were manually blackened out by us as they most often represent sensor noise, and the same was done with reflecting surfaces, which the Kinect sensor cannot measure. After this we received a total of 247 segments (i.e., about 20–30 per image). Segments are labeled on the 2D RGB image with different colors to make them distinguishable for the observer.


Perceptual influence of elementary three-dimensional geometry: (2) fundamental object parts.

Tamosiunaite M, Sutterlütti RM, Stein SC, Wörgötter F - Front Psychol (2015)

Overview of the computer vision method use for convex 3D-scene segmentation (for details see Stein et al., 2014a,b). (A) Two test objects (B1) Initial point clouds are reduced to supervoxels (Papon et al., 2013) with graph edges showing how voxels are neighbors. (B2) Conventional definition of convex and concave configurations. (B3) Singular locations like the one shown are not treated as concave, which massively improves algorithmic performance (C) Resulting convex (black) and concave (red) connectivity graph. (D1) Segmentation. (D2) Noise reduction mechanisms avoiding over-smoothing inside corners. (E) Final segmentation. Figure modified from Stein et al. (2014b).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4585234&req=5

Figure 1: Overview of the computer vision method use for convex 3D-scene segmentation (for details see Stein et al., 2014a,b). (A) Two test objects (B1) Initial point clouds are reduced to supervoxels (Papon et al., 2013) with graph edges showing how voxels are neighbors. (B2) Conventional definition of convex and concave configurations. (B3) Singular locations like the one shown are not treated as concave, which massively improves algorithmic performance (C) Resulting convex (black) and concave (red) connectivity graph. (D1) Segmentation. (D2) Noise reduction mechanisms avoiding over-smoothing inside corners. (E) Final segmentation. Figure modified from Stein et al. (2014b).
Mentions: We segmented the scenes along convex-concave transitions in the 3D-data by a machine vision algorithm. Figure 1 provides an overview of this method shown by ways of two simple test objects (Figure 1A). Point clouds are first reduced to few so-called supervoxels (Papon et al., 2013) which capture the scene geometry by their neighborhood relations (graph-edges in Figure 1B1). Convex and concave edge configuration are found using a conventional criterium (Figure 1B2) employed at the surface normals of each point but corrected against singularities as shown in Figure 1B3. (Some surface normals are shown graphically in Figure 1D2 by ways of arrows.) This results is convex (black) and concave (red) connections (Figure 1C), which are used to break up the scene (Figure 1D1). Corners such as the one shown in Figure D2 lead to an over-smoothing of the normals (see red arrow) and the algorithm at the end corrects for this leading to the final segmentation as shown in Figure 1E. Details of the algorithm are described elsewhere (Stein et al., 2014a,b). Note, this is a model-free, purely data-driven segmentation algorithm, as required for the purpose of this study, which does not use any additional features for segmentation. Due to the limited spatial resolution of the RGB-D sensors, small objects cannot be consistently labeled. Thus, segments smaller than 0.3% of the image size were manually blackened out by us as they most often represent sensor noise, and the same was done with reflecting surfaces, which the Kinect sensor cannot measure. After this we received a total of 247 segments (i.e., about 20–30 per image). Segments are labeled on the 2D RGB image with different colors to make them distinguishable for the observer.

Bottom Line: Additionally we control against segmentation reliability and we find a clear trend that reliable convex segments have a high degree of name-ability.In addition, we observed that using other image-segmentation methods will not yield nameable entities.This indicates that convex-concave surface transition may indeed form the basis for dividing objects into meaningful entities.

View Article: PubMed Central - PubMed

Affiliation: Faculty of Physics - Biophysics and Bernstein Center for Computational Neuroscience, University of Göttingen Göttingen, Germany ; Department of Informatics, Vytautas Magnus University Kaunas, Lithuania.

ABSTRACT
Objects usually consist of parts and the question arises whether there are perceptual features which allow breaking down an object into its fundamental parts without any additional (e.g., functional) information. As in the first paper of this sequence, we focus on the division of our world along convex to concave surface transitions. Here we are using machine vision to produce convex segments from 3D-scenes. We assume that a fundamental part is one, which we can easily name while at the same time there is no natural subdivision possible into smaller parts. Hence in this experiment we presented the computer vision generated segments to our participants and asked whether they can identify and name them. Additionally we control against segmentation reliability and we find a clear trend that reliable convex segments have a high degree of name-ability. In addition, we observed that using other image-segmentation methods will not yield nameable entities. This indicates that convex-concave surface transition may indeed form the basis for dividing objects into meaningful entities. It appears that other or further subdivisions do not carry such a strong semantical link to our everyday language as there are no names for them.

No MeSH data available.


Related in: MedlinePlus