Local feature descriptors are the most frequently used feature representation in many Computer Vision problems. In particular, high level semantic information extraction from low-level features in classification and retrieval is also quite successful. Region based approaches to classification and retrieval have become very popular. In this study, popular segmentation methods in the literature are investigated for the clustering of local feature descriptors and recently intensively used methods are used to group these features with a new perspective. Instead of all image pixels, segmentatiom is done using the pixels around the extracted features. It was observed that the proposed method in the obtained results made the similar segmentation process faster.