The aesthetic orientation of character modeling design is closely related to the final presentation effect and connotation expression of film and television animation. This paper takes the visual expression of cultural symbols in modeling design as the entry point, and chooses two forms of hue information extraction and gray processing to realize the digital compilation of cultural symbols. Combining YOLOv8m and image enhancement and other means to improve the performance of image detection in the symbol recognition link, using DBNet and CRNN to extract the text, accessing TSRAG for recognition, and adopting a weighted fusion decision-making mechanism to synthesize the two links, to establish an image-based dual-link logo recognition algorithm. On this basis, a compressed-aware image reconstruction method based on multi-scale inter-level feature fusion is proposed with compressed-aware and deep unfolding network as the core technology. The method optimizes the acquisition of local features and contextual information of cultural symbols by constructing a feature fusion proximal mapping module (MIFFPMM), and adds a new multi-scale inter-level feature fusion module in the deep reconstruction network to reduce information loss. The image reconstruction method can comfortably cope with the fluctuation of image sampling frequency, and the root-mean-square error is always lower than 0.25, and the lowest is only 0.01145, which is an effective visual reconstruction path for cultural symbols.
