With the development of the times, on the basis of inheriting traditional ceramic decoration, modern ceramic decoration is also changing, and it is moving towards personalisation and diversification. In this paper, firstly, we preprocess the images of the decoration features on the surface of ceramic vessels and use the LBP feature extraction method to extract the detailed features on the surface of ceramic vessels, transform them into HOG data, and represent the pattern information by histogram. Subsequently, an attention mechanism is added to achieve accurate segmentation of ornaments on the surface of ceramic vessels by learning the semantic features of the image. Then, based on the CGM model, the extracted ornament is deconstructed and restructured to design new ornament elements, which are applied to the surface of ceramic vessels. The investigation of the market demand for ceramic ware surface decoration was done to create a data base for ceramic decoration design. The research results show that 53.456% of users buy ceramic ware mainly because of daily use, and 49.563% pay more attention to the color pattern of ceramics. In the characteristic colours of ceramic patterns completed in the design, the characteristic colours 1→2 and 1→5, the correlation between the two groups of characteristic colours is strong, and the intensity is 3.554 and 4.265, respectively.