In this paper, machine learning and computer vision image segmentation methods are used to achieve classification of images and recognition and extraction of pattern elements. It also extracts colors from the imagery using intelligent image processing techniques, such as preprocessing and color clustering of Tangshan shadow images, and displays the results of color resolution. The accuracy of this paper’s method is reflected in the recognition and extraction of cultural heritage food images, as shown in the results. Tangshan shadow image elements are clustered into five categories, the proportion of colours used in Xiaodan are all between 0~12.5%~25%, and the proportion of high saturation in all five categories is more than 60%. The highest percentage of high brightness in Xiaodan colors is 36.2%. Applying Tangshan shadow image elements and color features to the brand logo and packaging design, a significant portion of respondents said they were very satisfied with the visual effect. The image elements and applications of cultural heritage also provide new perspectives and theoretical basis for modern visual design.