Complex Texture Segmentation of Digital Painting Image Based on Widow-HoffLMS Adaptive Algorithm

Abstract

When digital painting images are constructed to ensure a better visual effect of an image, it is necessary to carry out targeted processing on different areas of an image. So, image segmentation is a basis to ensure this processing effect. Therefore, this study proposes a method, called Widow-HoffLMS, for conducting complex texture segmentation for digital painting images. Firstly, the proposed algorithm is used to filter and process digital painting images, and the processing capability of the Hammerstein spline optimization algorithm is introduced to avoid damaging texture details of images and ensure detailed integrity of images. Then, filtered digital painting images are input into the adaptive network based on ghost convolution. A Ghost Module functions as an image block for convolutional neural networks that aims to generate more attributes by utilizing fewer parameters. After texture features in images are extracted through the ghost convolution layer of the network, the feature map is fused and extracted through the U-Net segmentation layer. Finally, the decoder is implemented to restore the spatial resolution of images and produce the result of complex texture segmentations of digital painting images. The test results show that the filtering effect of the proposed method is good, and the contrast of the filtered digital painting image is above 0.932. Moreover, the proposed method can complete the texture feature extraction of images with different sizes and different pixel scores, and the results of the inverse variance are all below 0.018. Furthermore, the complex texture segmentation of animated digital images can be completed according to the preset needs. The integrity state of the complex texture segmentation is also high.