Deep Analysis on the Color Language in Film and Television Animation Works via Semantic Segmentation Technique

Abstract

Color functions as a distinct type of ideographic symbol in animation for film and television, playing a crucial role in enhancing visual narratives and conveying emotions. In different types of animation, such as fantasy, horror, or children’s genres, color language influences audience perception and can convey meanings beyond the capabilities of image language alone. For instance, bright colors may symbolize innocence in children’s animation, while darker shades may evoke tension or fear in horror. However, current approaches to representing color in animation often fail to capture its full semantic richness and ideographic potential. Existing methods primarily focus on image-based analysis, overlooking the deeper layers of meaning encoded in color language. In this paper, we address these gaps by utilizing semantic segmentation techniques to combine the three modalities of color, content, and text to establish a consistent representation of color language in animation. We propose a method for semantic segmentation of color-depth (RGB-D) images using two-stream weighted Gabor convolutional network fusion. A weighted Gabor orientation filter builds a deep convolutional network (DCN) capable of extracting feature information adaptive to changes in orientation and scale, allowing for orientation- and scale-invariant features. Dual-stream picture features - color and depth - are extracted using a broad residual-weighted Gabor convolutional network and then combined into a lightweight feature extraction network. To evaluate the ideographic functions of color language quantitatively, we conducted extensive experiments using open databases. Our proposed method outperforms existing RGB-D picture semantic segmentation algorithms, demonstrating its effectiveness in representing color language in animation.