Using filter banks in Convolutional Neural Networks for texture classification

Abstract

Deep learning has established many new state of the art solutions in the last decade in areas such as object, scene and speech recognition. In particular Convolutional Neural Network (CNN) is a category of deep learning which obtains excellent results in object detection and recognition tasks. Its architecture is indeed well suited to object analysis by learning and classifying complex (deep) features that represent parts of an object or the object itself. However, some of its features are very similar to texture analysis methods. CNN layers can be thought of as filter banks of complexity increasing with the depth. Filter banks are powerful tools to extract texture features and have been widely used in texture analysis. In this paper we develop a simple network architecture named Texture CNN (T-CNN) which explores this observation. It is built on the idea that the overall shape information extracted by the fully connected layers of a classic CNN is of minor importance in texture analysis. Therefore, we pool an energy measure from the last convolution layer which we connect to a fully connected layer. We show that our approach can improve the performance of a network while greatly reducing the memory usage and computation.

Meta-Learning Convolutional Neural Architectures for Multi-Target Concrete Defect Classification With the COncrete DEfect BRidge IMage Dataset

“…Our images were acquired at high-resolution, partially using an unmanned aerial vehicle (UAV) to gain close-range access, and feature varying scale and context. We evaluate a variety of best-practice CNN architectures [21,32,1,38,16] in the literature on the CODEBRIM's multi-target defect recognition task. We show that meta-learned neural architectures achieve equivalent or better accuracies, while being more parameter efficient, by investigating and comparing two reinforcement learning neural architecture search approaches: MetaQNN [2] and "efficient neural architecture search" (ENAS) [27].…”

Texture CNN for Thermoelectric Metal Pipe Image Classification

“…The architecture of the Texture CNN is based on the T-CNN proposed by Andrearczyk and Whelan [8] which includes an energy layer that pools the feature maps of the last convolutional layer by calculating the average of its rectified activation output. This results in one single value per feature map, similar to an energy response to a filter bank.…”

“…That means that the flattened output of the energy layer is redirected directly after the last pooling layer, in the concatenation layer. This concatenation generates a new flattened vector containing information from the shape of the image and the texture, which are then propagated through the full connected layers [8].…”

“…However, these pre-trained CNNs focus on obtaining information about the overall shape of the image which leads to sparse and complex features that are less appropriate in texture analysis as we mainly seek repeated patterns of lower complexity [8]. Nevertheless, we have also evaluated some pre-trained models in Section IV.…”

Diversity in Fashion Recommendation Using Semantic Parsing

“…Therefore we propose to match the recovered semantic regions on the basis of texture rather than standard features obtained from ResNet [17] or VGGNet [18]. Texture has been a well-studied problem in computer vision where both traditional hand tuned [19,20] as well as deep learned features [21,22] have been proposed. We use the texture encoding layer proposed by Zhang et.…”