Table 3 Specific applications of intelligent techniques and optimisation algorithms in the field of colour recognition and colour segmentation

FCM algorithm

This research implemented an FCM algorithm to detect the density of single-system fabrics

The maximum error in the detection results was 0.24%

Pan et al. (2013)

This research proposed a novel FCM-based stepwise classification method for automatic recognition of colour yarn layout of single-system colour woven fabrics

The error of the preliminary automatic recognition was only 8.57% and 11.27%

Zhang, Pan, et al. (2015a)

This research aimed to automatically detect the colour effects of fabric of dyed yarn, using a minimum repetition-based unit recognition algorithm

The FCM algorithm categorized the colour of all floats with an error rate of 18.75%

Zhang et al. (2015b)

This research proposed an automatic analysis procedure for embroidery fabric images that includes colour analysis, pattern shape analysis, and texture analysis. In colour analysis, RGB images are pre-processed using wavelet transform and median filter, and then binary region segmentation is performed using the FCM clustering method, which can yield colour statistical values and colour features in terms of the number of colours

The proposed colour segmentation method is relatively more stable and can reduce the calculation time by 75% compared to the FCM and clustering validity index-based colour segmentation methods

Shih et al. (2016)

ANN

This research investigated the use of an ANN in a vision system to classify the colour fastness of woven fabrics

The vision system using ANN was able to identify the colour fastness of woven fabrics with 100% accuracy for red, purple, and yellow, 83.33% accuracy for green, and 91.67% accuracy for blue

Samsi et al. (2013)

Automatic Feedback Error Correction (AFEC algorithm)

This research developed an automatic feedback error correction colour-weave pattern recognition algorithm (AFEC algorithm) for automatic colour recognition of yarn-dyed fabrics

The X-means clustering algorithm achieved a false positive rate of about 1.33% in the classification of colours in the original image

Zhou et al. (2013)

A new unsupervised method

These researchers developed a new unsupervised method to detect colour regions that are prominent in images of yarn-dyed fabrics. The colour of the dominant colour region is correlated with the colour of the yarn through a probabilistic model. And it estimates the colour histogram of the main colour region based on the colour of the yarn in the image of the yarn-dyed fabric. A hierarchical segmentation structure is then designed to identify the dominant colour region within the image

The model processes a dominant colour region in images taken with macro and telephoto lenses in just 0.045 and 0.075 s, indicating suitability for industrial applications

Luo et al. (2013)

A grading algorithm based on rotationally invariant statistical features

This research focused on using image analysis techniques from the CIELAB colour model and a grading algorithm to describe the colour and automatically grade silk threads according to colour

The colour-sorting solution achieved an accuracy rate of 91%

Pal et al. (2013)

An improved model based on multi-stage fuzzy region competition, and fuzzy region-based segmentation (FRBS) approach

This research proposed an improved model based on multi-stage fuzzy region competition for colour segmentation of colour images, and recoloured textile images with different colour themes to achieve automatic colour theming of textile images

Not specified

Han et al. (2013)

This research proposes a new method that accurately divides colour regions for several types of fabrics using the CIELAB colour system and a fuzzy region-based segmentation (FRBS) approach

The FRBS approach outperformed traditional methods such as FCM clustering methods, Gustafson-Kessel clustering methods, expectation maximisation segmentation methods, traditional FRBS approaches as well as PCA-based fuzzy region competition methods

Zheng (2015)

This research proposed an intelligent colour pattern recognition algorithm for yarn-dyed fabrics that uses a fuzzy region competition strategy in the classification of yarn colours

The extracted yarn colours are highly similar to those recognised by human vision, with a computation time of fewer than 15 s for the whole detection process

Zheng et al. (2019)

A colour clustering method based on a node classification correction method using adjacency information

This research aimed to classify the colours of nodes in the digital intelligent recognition of fabric weave patterns using a node classification correction method based on adjacency information

The study improved the accuracy of node classification by using the type and colour information of neighbouring nodes

Zhong et al. (2013)

K-means clustering method, X-mean clustering algorithm, improved watershed algorithm and improved K-means clustering method

This research focused on developing a new colour clustering algorithm based on the K-means clustering method to classify dyed yarns according to their colours in interlaced multicoloured dyed woven fabrics

The new method was found to be superior to the traditional approach in terms of accuracy and robustness

Zhang et al., (2015a, 2015b, 2015c)

This research proposed an intelligent detection method based on image analysis to identify the colour and weaving patterns of yarn-dyed fabrics automatically

The experimental results showed that the proposed method was able to identify the colour of yarn-dyed fabrics very well

Li et al. (2019)

This research proposes a new approach for colour segmentation and extraction of multicoloured yarn woven fabrics via a hyperspectral imaging system. The approach includes an improved watershed algorithm and an improved K-means clustering method to address the over-segmentation issue of the watershed algorithm

The proposed approach outperforms ordinary K-means, FCM, and DPC methods in the evaluation metrics of compactness and separateness, with an execution efficiency improvement of a minimum of 55%. The colour difference between the presented approach and the spectrophotometrically measured CMC (2:1) was between 0.60 and 0.88 units, indicating high accuracy of colour measurement

Zhang et al. (2020)

NGSOM-based clustering algorithm

This research proposed an NGSOM-based clustering algorithm for intelligent colour classification of colour images and accurate delineation of regions of different colours for colour measurement

The proposed algorithm was found to have a better peak signal-to-noise ratio and higher time efficiency than some commonly used colour clustering algorithms

Ouyang et al. (2019)

SOM algorithm, DPC algorithm and EDSC algorithm

This research used a hyperspectral imaging system to determine optimum amounts of colour and merge clusters after multicolour measurements of printed fabrics by an algorithm that combines a SOM algorithm to determine the main clusters and a DPC algorithm

The approach efficiently identified the optimum amounts of colour for accurate colour segmentation of the printed fabric and required less execution time

Zhang et al. (2019)

This research focused on developing an algorithm for colour segmentation of multicoloured porous printed fabrics using a combination of SOM Neural Network and EDSC algorithms. After pre-processing the fabric images, primary clustering was achieved by the SOM algorithm and then secondary clustering was performed by the EDSC algorithm. In the clustering process of the EDSC algorithm, the optimal silhouette coefficient is introduced to automatically determine the number of clustering centres. Finally, post-processing such as grey scale transformation, binarization and open operation eliminates the incorrect segmentation of edge colours, making the algorithm more suitable for industrial applications

The algorithm was found capable of accurately identifying colours of small regions and colour segment complex printed fabric images with an accuracy of 88.3%

Qian et al. (2022)

SVM

This research performed colour sorting of recycled waste textiles by the SVM classification method in a computer vision system

The classification accuracy for 466 textile samples was 96.57%

J. Zhou et al., (2021a, 2021b)

ANN and deep learning convolutional neural networks trained with the SVM method

This research focused on digital image analysis for colour matching of knitted cotton fabrics using ANN and deep learning convolutional neural networks

The ANN predicted three colours (blue, red and mauve) with correct classification rates of 82.37%, 83.16% and 89.25%, respectively. The deep learning convolutional network trained with the SVM method predicted 100%, 100%, and 100% of the three colours

Das and Wahi (2022)