In recent years, a tremendous research effort has been made in the area of generic object recognition. However, the most important thing is not the names but functions for robots to comprehend objects. Object functions refer to “the purpose that something has or the job that someone or something does”. Various elements (e.g., the physical information, material, appearance and human interaction) independently or mutually form object functions. There are many researches on object functions using human-object interaction, while there are few using appearance.
However, it can be believed that object functions may be formed by appearance. In this paper, we propose a new method to estimate object functions from appearance on images. Our approach is to estimate object functions using DPM by dividing object appearance into parts. There are important parts and less important parts in the appearance for the functions. Therefore, we identify the important parts in the object appearance for the functions. Experimental results show that the important parts about specific functions can be extracted and object functions are related to the appearance.