Accurate recognition of food ingredients is necessary for the health and religion of customers. The task addressed in this paper is to detect the names of specific ingredients on low-quality images of food labels. This paper proposes a method for detecting specific ingredients on food labels with high accuracy using a document classification approach that learns co-occurrence of ingredient names in food labels. This paper conducts experiments with food label images taken using smartphones, to determine the presence of ingredients that are sometimes restricted by the Islamic religion. Compared to a simple method based on string matching, the proposed method is less accurate for large images, but shows high accuracy for small images. In addition, a combination of the proposed method and the simple method was able to achieve detection of specific ingredients with less oversight.