In this paper, we propose a novel method for detection and clustering of defects in ultrasound images based on rosette pattern and image processing techniques. The method employs rosette pattern to extract main features of the image and then according to rosette pattern properties, the extracted samples are mapped to a linear plane. In this step, by the use of image processing techniques, all of the samples are clustered and the number of defects is examined. Finally, the samples are remapped to the main plane. Experimental results show that the proposed method increases capability of defect detection effectively, and outperforms fuzzy c-means clustering.