Histogram Layers for Texture Analysis

An essential aspect of texture analysis is the extraction features that describe distribution values in local, spatial regions. We present a localized histogram layer for artificial neural networks. Instead computing global histograms as done previously, proposed directly computes and parameters are estimated during backpropagation. compare our method with state-of-the-art encoding methods such Deep Encoding Network Pooling, Texture Network, Fisher Vector convolutional network, Multi-level Representation on three material/texture datasets: (1) Describable Dataset; (2) an extension ground terrain outdoor scenes; (3) subset Materials Context dataset. Results indicate inclusion improves performance. The source code publicly available: https://github.com/GatorSense/Histogram_Layer.