Manual analysis of the bulk data generated by computed tomography angiography (CTA) is time consuming, and interpretation of such data requires previous knowledge and expertise of the radiologist. Therefore, an automatic method that can isolate the coronary arteries from a given CTA dataset is required. We present an automatic yet effective segmentation method to delineate the coronary arteries from a three-dimensional CTA data cloud. Instead of a region growing process, which is usually time consuming and prone to leakages, the method is based on the optimal thresholding, which is applied globally on the Hessian-based vesselness measure in a localized way (slice by slice) to track the coronaries carefully to their distal ends. Moreover, to make the process automatic, we detect the aorta using the Hough transform technique. The proposed segmentation method is independent of the starting point to initiate its process and is fast in the sense that coronary arteries are obtained without any preprocessing or postprocessing steps. We used 12 real clinical datasets to show the efficiency and accuracy of the presented method. Experimental results reveal that the proposed method achieves 95% average accuracy.
We propose an approach to automatically extracting foreground regions. This is a novel method for segmenting salient objects from still images by background elimination. To extract foreground regions, a new method of background elimination based on multiscale segmentation is proposed to detect candidate object regions. To this end, we use a trimap consisting of foreground, background, and undefined regions and a region adjacency graph. A graph-cut technique is finally used to extract exact foreground regions from the candidates. Experimental results have shown that the proposed method yields a better foreground extraction than Kim's method under various environments containing multiple objects and clutter backgrounds in natural images.
We propose an effective method of extracting targets from a region-of-interest (ROI) in infrared images by using a 2-D histogram, considering intensity values and distance values from a center of the ROI. Existing approaches for extracting targets have utilized only intensity values of pixels or an analysis of a 1-D histogram of intensity values. Because the 1-D histogram has mixed bins containing false-negative bins from the target region as well as false-positive bins from the background region, it is difficult to extract target regions effectively due to the mixed bins. In order to solve the problem of the mixed bins, we propose a novel 2-D histogram-based approach for extracting targets. Experimental results have shown that the proposed method achieves better performance of extracting targets than existing methods under various environments, such as target regions with irregular intensities, dim targets, and cluttered backgrounds.