Human beings get a lot of information from a picture based on what we see and our background knowledge. However, many computer vision researches are heavily dependent on the use of image features and have paid little attention to background knowledge we use in texture processing. The present study explores the degree to which onomatopoeia evoked by visual images is affected by the multimodal experience-based knowledge such as tactile experience. In Experiment 1 participants saw original complete images of Flickr Material Database (FMD) and answered onomatopoeia for expressing their textures and in Experiment 2 participants saw cut out images and answered onomatopoeia for expressing their textures. We obtained 17487 onomatopoeic words (1827 types) from experiment 1 and 30138 onomatopoeic words (2442 types) from experiment 2. We counted the number of types of onomatopoeia evoked by each image. Result showed that original image evoked significantly more variety of onomatopoeia than cut-off image. This result suggests that human texture evaluations based on the original complete images of FMD are affected more easily by experience-based knowledge about the material. Furthermore, we showed that image whose material category is relatively easy to recognize evokes significantly frequently tactile onomatopoeia than image whose material category is hard to recognize.