Fabric surface characterization: assessment of deep learning-based texture representations using a challenging dataset

Hu, Yuting; Long, Zhiling; Sundaresan, Anirudha; Alfarraj, Motaz; AlRegib, Ghassan; Park, Sungmee; Jayaraman, Sundaresan

doi:10.1080/00405000.2020.1757296

Cited by 7 publications

(6 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some image sets were collected in lab settings from cropped stand-alone samples (e.g., CUReT [ 35 ] in 1999, KTH-TIPS [ 36 ] in 2005); meanwhile, others were collected in the wild (e.g., FMD [ 37 ] in 2009, OpenSurfaces [ 38 ] in 2013, MINC [ 39 ] in 2015, and LFMD [ 40 ] in 2016) with more diverse samples and real-world scene contexts. The number of classes and the number of samples in each class varies greatly from one dataset to another (e.g., 10 classes/810 images in total for KTH-TIPS, 61 classes/5612 images in total for CUReT); likewise, the diversity of input parameters also varies significantly (e.g., small viewpoint changes in KTH-TIPS, larger viewpoint changes in CUReT) [ 41 ]. The KTH-TIPS (Textures under varying Illumination, Pose and Scale) image database was created to extend the CUReT database by providing variations in scale [ 36 ].…”

Section: Related Workmentioning

confidence: 99%

Multi-View Learning for Material Classification

Sumon

Muselet

et al. 2022

J. Imaging

View full text Add to dashboard Cite

Material classification is similar to texture classification and consists in predicting the material class of a surface in a color image, such as wood, metal, water, wool, or ceramic. It is very challenging because of the intra-class variability. Indeed, the visual appearance of a material is very sensitive to the acquisition conditions such as viewpoint or lighting conditions. Recent studies show that deep convolutional neural networks (CNNs) clearly outperform hand-crafted features in this context but suffer from a lack of data for training the models. In this paper, we propose two contributions to cope with this problem. First, we provide a new material dataset with a large range of acquisition conditions so that CNNs trained on these data can provide features that can adapt to the diverse appearances of the material samples encountered in real-world. Second, we leverage recent advances in multi-view learning methods to propose an original architecture designed to extract and combine features from several views of a single sample. We show that such multi-view CNNs significantly improve the performance of the classical alternatives for material classification.

show abstract

Section: Related Workmentioning

confidence: 99%

Multi-View Learning for Material Classification

Sumon

Muselet

et al. 2022

J. Imaging

View full text Add to dashboard Cite

show abstract

“…Alternative practise has been required to reconfigure CNN's performance in order to improve limited pattern localization. Another key concern in deep learning models is that input from practical cases is not always accessed broadly among labels [7]. As a result, when designing the CNN for detecting FDs, 2 important tasks are considered: the restoration of limited structures and the managing an imbalanced dataset.…”

Section: Introductionmentioning

confidence: 99%

“…In Eqns. ( 6)- (7), 𝑔 is the gradient of the fitness value 𝐽wherein the cross-entropy is applied as the fitness value. Here, 𝛽 , modifies over interval 𝑡 as: 𝛽 , ( 7 ) So, it modifies the grade to estimate how much the previous gradients weigh in the assessment.…”

mentioning

confidence: 99%

“…( 6)- (7), 𝑔 is the gradient of the fitness value 𝐽wherein the cross-entropy is applied as the fitness value. Here, 𝛽 , modifies over interval 𝑡 as: 𝛽 , ( 7 ) So, it modifies the grade to estimate how much the previous gradients weigh in the assessment. The major dissimilarity of NWM-Adam from Adam is that it uses a modifying 𝛽 , to control the exponential decay percentage of moving mean of the squared gradient rather than fixed 𝛽 in Adam.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Optimizing Gradients Weight of Enhanced Pairwise-Potential Activation Layer in CNN for Fabric Defect Detection

Vinothini¹,

Sheeja²

2022

INDJCSE

View full text Add to dashboard Cite

Imperfection classification is the most involved task in the cotton sector for finding Fabric Defects (FDs) and improving fiber productivity. Several approaches have been suggested in ancient times to automatically classify FDs. Presently, an Enhanced Pairwise-Potential Activation Layer in Convolutional Neural Network (EPPAL-CNN) approach depends on improved external memory and Dynamic Conditional Random Fields (DCRFs) to solve the complex pattern correlation of FDs and detect the defective fabrics from the given images. On the contrary, the gradient-based optimization schemes for learning the weights of CNN tend to unusual convergence nature, resulting in inefficient classification. Hence in this paper, an EPPAL-Optimized CNN (EPPAL-OCNN) approach is proposed which introduces an individual weight optimization scheme depending on NWM-Adam for solving the unwanted convergence of CNN. In this approach, a novel first-order gradient descent optimization method is introduced, which applies an adaptive exponential decay percentage for second-moment approximation rather than a preconfigured and constant one. Also, it can simply modify the grade to which how much the previous gradients weigh in the approximation. This novel exponential moving mean deviation is designed based on the fact that assigning additional memory to the previous gradients compared to the current gradients. Thus, it guarantees effective convergence and increases detection accuracy. At last, the testing results reveal that the EPPAL-OCNN achieves 94.64% of accuracy to different state-of-the-art approaches on the TILDA database.

show abstract

“…Analysis of texture content in digital images plays an important role in the automated visual inspection of textile images. Such approaches are used mainly to assess the properties and quality of fabrics, including flaw detection [28] and surface structure [29]. The most popular method for quantitative assessment of texture is the use of Haralick features [30,31] .…”

Section: Introductionmentioning

confidence: 99%

A Method for the Assessment of Textile Pilling Tendency Using Optical Coherence Tomography

Sekulska‐Nalewajko

Gocławski

Korzeniewska

2020

Sensors

View full text Add to dashboard Cite

Pilling is caused by friction pulling and fuzzing the fibers of a material. Pilling is normally evaluated by visually counting the pills on a flat fabric surface. Here, we propose an objective method of pilling assessment, based on the textural characteristics of the fabric shown in optical coherence tomography (OCT) images. The pilling layer is first identified above the fabric surface. The percentage of protruding fiber pixels and Haralick’s textural features are then used as pilling descriptors. Principal component analysis (PCA) is employed to select strongly correlated features and then reduce the feature space dimensionality. The first principal component is used to quantify the intensity of fabric pilling. The results of experimental studies confirm that this method can determine the intensity of pilling. Unlike traditional methods of pilling assessment, it can also detect pilling in its early stages. The approach could help to prevent overestimation of the degree of pilling, thereby avoiding unnecessary procedures, such as mechanical removal of entangled fibers. However, the research covered a narrow group of fabrics and wider conclusions about the usefulness and limitations of this method can be drawn after examining fabrics of different thickness and chemical composition of fibers.

show abstract

Fabric surface characterization: assessment of deep learning-based texture representations using a challenging dataset

Cited by 7 publications

References 45 publications

Multi-View Learning for Material Classification

Multi-View Learning for Material Classification

Optimizing Gradients Weight of Enhanced Pairwise-Potential Activation Layer in CNN for Fabric Defect Detection

A Method for the Assessment of Textile Pilling Tendency Using Optical Coherence Tomography

Contact Info

Product

Resources

About