Model-based learning of local image features for unsupervised texture segmentation

• Verfasst von: Kiechle, Martin [VerfasserIn]
• Verfasst von: Storath, Martin [VerfasserIn]
• Verfasst von: Weinmann, Andreas [VerfasserIn]
• Verfasst von: Kleinsteuber, Martin [VerfasserIn]
• Titel: Model-based learning of local image features for unsupervised texture segmentation
• Verf.angabe: Martin Kiechle, Martin Storath, Andreas Weinmann, and Martin Kleinsteuber
• E-Jahr: 2018
• Jahr: January 26, 2018
• Umfang: 14 S.
• Fussnoten: Gesehen am 21.10.2019
• Titel Quelle: Enthalten in: Institute of Electrical and Electronics EngineersIEEE transactions on image processing
• Ort Quelle: New York, NY : IEEE, 1992
• Jahr Quelle: 2018
• Band/Heft Quelle: 27(2018), 4, Seite 1994-2007
• ISSN Quelle: 1941-0042
• DOI: doi:10.1109/TIP.2018.2792904
• Datenträger: Online-Ressource
• Sprache: eng
• Sach-SW: Computational modeling
• Sach-SW: Cost function
• Sach-SW: Data models
• Sach-SW: feature extraction
• Sach-SW: feature vector
• Sach-SW: geometric optimization
• Sach-SW: ground truth segmentation
• Sach-SW: histological images
• Sach-SW: image patches
• Sach-SW: image segmentation
• Sach-SW: Image segmentation
• Sach-SW: image texture
• Sach-SW: learning (artificial intelligence)
• Sach-SW: learning process
• Sach-SW: local image features
• Sach-SW: Mathematical model
• Sach-SW: model-based learning
• Sach-SW: Mumford-Shah model
• Sach-SW: piecewise constant feature image
• Sach-SW: piecewise constant Mumford-Shah model
• Sach-SW: Prague texture segmentation benchmark
• Sach-SW: suitable convolutional features
• Sach-SW: textural patterns
• Sach-SW: Texture segmentation
• Sach-SW: texture segmentation methods
• Sach-SW: Training data
• Sach-SW: unsupervised learning
• Sach-SW: unsupervised texture segmentation
• K10plus-PPN: 1679247190

Abstract

Features that capture well the textural patterns of a certain class of images are crucial for the performance of texture segmentation methods. The manual selection of features or designing new ones can be a tedious task. Therefore, it is desirable to automatically adapt the features to a certain image or class of images. Typically, this requires a large set of training images with similar textures and ground truth segmentation. In this paper, we propose a framework to learn features for texture segmentation when no such training data is available. The cost function for our learning process is constructed to match a commonly used segmentation model, the piecewise constant Mumford-Shah model. This means that the features are learned such that they provide an approximately piecewise constant feature image with a small jump set. Based on this idea, we develop a two-stage algorithm which first learns suitable convolutional features and then performs segmentation. We note that the features can be learned from a small set of images, from a single image, or even from image patches. The proposed method achieves a competitive rank in the Prague texture segmentation benchmark, and it is effective for segmenting histological images.