Digital Image Analysis Using Deep Learning Convolutional Neural Networks for Color Matching of Knitted Cotton Fabric

Abstract

The customers make use of Pantone color cards as quality control for their reference to know the color consistency of the dyed cloth. The three different color shades like blue, red and pastel violet were selected from the pantone color shades. The three-color hues of the manufactured knitted cotton fabric were captured by the high-resolution optical device and these images were considered for the training and test purpose. A simple backpropagation based artificial neural network and a deep learning convolution network was considered for the training and test purpose. The pantone color hues of three images were offered to the network as training samples. A backpropagation algorithm trained artificial neural network (ANN) and deep learning neural network trained with support vector machine were employed in training phase. The back propagation trained ANN predicted 82.37%, 83.16% and 89.25% correct classification on three color hues. A deep learning convolution network trained with support vector machine method forecasted 100%, 100% and 100% on three color hues. The better performance result was secured by the second method. The second network reduces the features from the color images in training and test phase.

Keywords

• knitted cotton fabric
• panton tcx
• hue
• image capture
• color matching
• convolutional neural network (cnn)