Character generation and visual quality enhancement in animated films using deep learning

With the application and development of technologies such as artificial intelligence and deep learning in the generation of animated films, improving the quality and accuracy of generated images to enhance the visual communication effects of animated films has become an important research direction. This work aims to optimize the first order motion model (FOMM) to enhance its performance in generating animated character images. To this end, the convolutional block attention module (CBAM) is introduced into FOMM. Based on this, the CBAM is redesigned to enhance the network's ability to focus on important features, especially in terms of accuracy in complex backgrounds. Meanwhile, to address the image distortion problem caused by severe pose changes, a repainting image repair module is proposed. Through multi-scale upsampling and occlusion map prediction mechanisms, it effectively improves the coherence and completeness of image reconstruction. Ultimately, the proposed enhanced FOOM (E-FOOM) model realizes the deep coupling of attention mechanisms and reconstruction modules, and a more robust end-to-end character image generation framework is constructed. Experimental results on the VoxCeleb1 and TaiChiHD datasets show that the E-FOOM model outperforms existing models in terms of generated image quality, keypoint detection accuracy, and pose reconstruction. Additionally, the model's generated images exhibit a minimum peak signal-to-noise ratio increase of 1.11 dB and a minimum structural similarity index improvement of 0.014, indicating superior pixel-level, structural, and perceptual quality. This work intends to enhance the quality of generated character images in animated films, providing a technical pathway for achieving high-quality visual effects.