Learning scale-variant and scale-invariant features for deep image classification. (January 2017)

• Record Type: Journal Article
• Authors: van Noord, Nanne; Postma, Eric
• Is Part Of: Pattern recognition. Volume 61(2017:Jan.)
• Page Start: 583
• Page End: 592
• Publication Date: 2017-01
• Subjects: Convolutional Neural Networks -- Multi-scale -- Artist Attribution -- Scale-variant Features; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.06.005
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Convolutional Neural Networks (CNNs) require large image corpora to be trained on classification tasks. The variation in image resolutions, sizes of objects and patterns depicted, and image scales, hampers CNN training and performance, because the task-relevant information varies over spatial scales. Previous work attempting to deal with such scale variations focused on encouraging scale-invariant CNN representations. However, scale-invariant representations are incomplete representations of images, because images contain scale-variant information as well. This paper addresses the combined development of scale-invariant and scale-variant representations. We propose a multi-scale CNN method to encourage the recognition of both types of features and evaluate it on a challenging image classification task involving task-relevant characteristics at multiple scales. The results show that our multi-scale CNN outperforms single-scale CNN. This leads to the conclusion that encouraging the combined development of a scale-invariant and scale-variant representation in CNNs is beneficial to image recognition performance. Abstract : Highlights: We propose a new approach for learning scale-variant and scale-invariant features. The multi-scale CNN is an ensemble of scale-specific CNN. Our multi-scale CNN is used to achieve state-of-the-art results for artist attribution. The learnt deep multi-scale representation encodes both fine and coarse characteristics.