Deep integration of conditional gan, attention mechanism, and image clustering for automated color separation and correction in textile screen printing

Manual color separation and image processing are needed for pretreatment in traditional textile screen printing but the shortage of engineers has created a significant bottleneck in improving printing quality. Therefore, this research aimed to introduce an innovative method by proposing the squeeze-excitation attention and Pix2PixHD (SEAPix) incorporating KMeans clustering algorithm, forming KMeans+SEAPix model. The proposed SEAPix integrates a squeeze and excitation attention (SEA) mechanism into the conditional generative adversarial network (cGAN), Pix2PixHD, architecture. KMeans algorithm facilitates the separation of similar colors into distinct branches. Furthermore, SEAPix method improves the accuracy of high-resolution image generation by focusing on important regions and reducing irrelevant noise. To verify the sequence of color separation and image processing, this research developed two different frameworks, namely the Image Generation-Separation Framework (PF1) and the Image Generation-Separation Framework (PF2). PF1 starts with image generation, followed by Kmeans color separation, while PF2 adopted a different method by using color separation first, followed by generating images with cGAN. The experimental results of this research showed that PF2 was superior to PF1. Specifically, PF2 had better structural similarity index measure (SSIM), peak signal-to-noise ratio (PSNR), intersection over union (IoU), and pixel accuracy increased by 15.5643%, 0.5249%, 11.3757%, and 1.6508%, respectively. PF2 also has lower learned perceptual image patch similarity (LPIPS) and mean squared error (MSE) decreased by 24.3518% and 2.5963% compared with PF1. In conclusion, the proposed KMeans+SEAPix could revolutionize traditional textile screen printing methods by providing automated color separation and image correction.