Slub extraction and measurement in denim fabric images based on conditional image-to-image translation

The dark color and irregular length, thickness, and inter-slub separation distance of the slub make it challenging to analyze the samples and extract the slub parameters in slub denim. A method for extracting and measuring slubs of slub denim based on conditional image-to-image translation was proposed to solve this problem. The generator commonly used in Pix2Pix model was modified by adding additional three layers of down-sampling and up-sampling, and expending the receptive field of the convolution kernel to extract deeper features from source images. The new network structure improved the ability to extract deep image features. The image generated by the model was processed by image morphology, and the connected components of a small area were eliminated to get the final result. The peak signal-to-noise ratio (PSNR), mean square error (MSE), structural similarity (SSIM), and Fréchet Inception Distance (FID) were used as evaluation metrics. The slub length measurements in the generated image and ground truth were also compared. The evaluation metrics value of 17.44, 0.0176, 0.95, and 10.75 respectively, indicating that the proposed method can generate the distribution of slubs on denim fabric surface under constraints. Moreover, the method can measure the length of slubs in the fabric even without corresponding yarns, with most measured slub lengths having errors less than 8 mm. The proposed method is effective and superior in extracting and measuring slubs on denim fabric surfaces, offering valuable reference for enterprises in analyzing slub denims, saving labor and material resources in the proofing process.