We present a method for recognizing the existence of outer frames in binary trademark images and segmenting a trademark that contains an outer frame into the frame and its inner figure, even if both touch. This focuses on the development of content-based image retrieval (CBIR) for trademark registration. Using our proposed method, CBIR systems examine the similarity between images using only main image components. This includes a study for describing image components. We detail criteria of trademark image outer frames and propose an algorithm for recognizing and segmenting outer frames based on the criteria. Experimental results using 1843 registered trademark images and experimental evaluation by 13 participants showed that 98.4% of recognitions agreed with human perception.