Spatial Context - Images
| Domain | Outdoor images |
| Media | Image |
| Size | 127 MB |
| Instances | 923 images |
| File Format | JPEG |
| Creation Date | 2009 |
| Task | Concept detection |
| Copyright | The dataset is publicly available for non-commercial use. |
| URL | http://mklab.iti.gr/project/scef |
Description
The dataset includes: a) the utilized image set, b) the image segmentation masks, c) region-level manual image annotation, d) the extracted low-level features, e) the computed fuzzy directional spatial relations, and f) the region classification results based solely on visual information.
Quality
JPG compressed images.
Source
The image set was provided by Alinari 24 ORE (http://www.alinari.com, info: euproject-2@alinari.it).
Ground Truth Annotation
Concept annotations
Includes the region-level manual image annotations, which were used for training and evaluation. Supported concepts: Building, Foliage, Mountain, Person, Road, Sailing-boat, Sand, Sea, Sky, Snow.
Directional Relations
Includes the fuzzy directional spatial relations computed for every image. Supported spatial relations: Above (A), Above-Right (AR), Above-Left (AL), Right (R), Left (L), Below (B), Below-Right (BR), Below-Left (BL). Each spatial relation is estimated for every possible image region pair and receives values in the interval []1,0. ‘Ground’ region denotes the image region used as reference, while ‘Figure’ region is the region whose relative position is estimated with respect to the ‘ground’ region.
Features
Segmentation masks
Includes the segmentation mask for every image. This is stored as grey scaled image (binary pgm format), where the intensity value at each pixel denotes the image region to which the particular pixel belongs. Pixel intensity receives values in the interval [0, N-1] where N is the number of regions to which every image is segmented. In order to better visualize the segmentation masks use a .pgm image viewer and perform histogram equalization
Visual features
The low-level features extracted for every image segment. Two sets of features are available: a) MPEG-7 descriptors: Homogeneous Texture, Scalable Color, Edge Histogram and Region Shape, b) wavelet-based features.
Classification results
The region-level classification results for the following feature-classifier combinations: MPEG-7-SVM, MPEG-7-ML, Wavelet-ML. A degree of confidence is estimated for every possible region-concept pair.
Copyright Remarks
The dataset is publicly available for non-commercial use.
Citation
Please refer to [Papadopoulos et al., Proc. WIAMIS'09, London, UK] if this dataset is used for experimentation in publications.
