Anime-style video game background generation adversarial neural networks

At present, anime-style video games have a growing market and audience. In the development of this type of game, there is a problem that high-quality background drawing requires huge resources. Traditional anime stylization algorithms are usually based on conventional convolution operations and are difficult to produce high-quality anime-style images. The animation stylization algorithm based on generative adversarial network uses unpaired data set training and can generate animation images with obvious animation style characteristics. However, the color of the generated image is prone to distortion, losing the content characteristics of the input image, and producing artifacts. In response to the above problems, a generative adversarial network algorithm based on an improved image segmentation algorithm was proposed. This study innovatively utilized image segmentation algorithms to construct a loss function that can accurately restore colors, and added residual units in the generator and discriminator stages. This study innovatively uses the Felzenszwalb image segmentation algorithm to extract the global color structure information of the image, and integrates it into the loss function construction of the generative adversarial network - by calculating the structural loss of the segmentation region, the color partition consistency between the generated image and the original image is constrained; Simultaneously incorporating residual units into both the generator and discriminator to alleviate network degradation issues. Experiments have shown that this algorithm can convert photos into animation images with lower color distortion and smoother texture. The research results provide technical support for the improvement of visual effects in animated images, as well as new ideas and methods for the development of image processing technology, which is more conducive to promoting research and technological progress in related fields.