Deep analysis on the color language in film and television animation works via semantic segmentation technique

Abstract

In film and television animation works, colour is an ideographic symbol distinct from other elements. Color language and image language are separate concepts, with image language unable to fully capture the ideographic nature of color. Even when images appear visually similar, differences in brightness can create significant contrasts or opposing meanings. Thus, color language plays an ideographic role between color and concept. This paper uses semantic segmentation techniques to integrate the three modalities of color content and text, establishing a consistent representation relationship. We propose a color-depth (RGB-D) image semantic segmentation method based on two-stream weighted Gabor convolutional network fusion. To capture Orientation- and scale-invariant features, we design a weighted Gabor orientation filter within a deep convolutional network (DCN) to adapt to changes in Orientation and scale. A wide residual-weighted Gabor convolutional network extracts features from the dual-stream images of colour and depth. To quantitatively assess our method’s representational ability regarding the ideographic functions of color language, we conduct extensive experiments on public datasets. The results demonstrate that the proposed algorithm outperforms existing RGB-D image semantic segmentation methods. Specifically, the technique achieves superior accuracy across several performance metrics, with a notable improvement of 2.5%–6.6% compared to baseline models. Our approach enhances segmentation precision for objects of varying scales and directions. It exhibits robustness in complex lighting environments, thus confirming its potential in real-world applications of color language in animation.