English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Convolutional Neural Networks (ConvNets) at present achieve remarkableperformance in image classification tasks. However, current ConvNets cannotguarantee the capabilities of the mammalian visual systems such as invarianceto contrast and illumination changes. Some ideas to overcome the illuminationand contrast variations usually have to be tuned manually and tend to fail whentested with other types of data degradation. In this context, we present a newbio-inspired {entry} layer, M6, which detects low-level geometric features(lines, edges, and orientations) which are similar to patterns detected by theV1 visual cortex. This new trainable layer is capable of coping with imageclassification even with large contrast variations. The explanation for thisbehavior is the monogenic signal geometry, which represents each pixel value ina 3D space using quaternions, a fact that confers a degree of explainability tothe networks. We compare M6 with a conventional convolutional layer (C) and adeterministic quaternion local phase layer (Q9). The experimental setup {isdesigned to evaluate the robustness} of our M6 enriched ConvNet model andincludes three architectures, four datasets, three types of contrastdegradation (including non-uniform haze degradations). The numerical resultsreveal that the models with M6 are the most robust in front of any kind ofcontrast variations. This amounts to a significant enhancement of the C models,which usually have reasonably good performance only when the same training andtest degradation are used, except for the case of maximum degradation.Moreover, the Structural Similarity Index Measure (SSIM) is used to analyze andexplain the robustness effect of the M6 feature maps under any kind of contrastdegradations.

A trainable monogenic ConvNet layer robust in front of large contrast changes in image classification