A fast multilevel method for selective segmentation model of 3-D digital images

A. K. Jumaat, K. Chen

Selective segmentation is capable of extracting an object or region in a given image. Recently, a 3-D convex variational selective segmentation model has been proposed and solved using projection algorithm. For moderate size of image, the projection algorithm method is effective. However, as the image size is larger, a fast iterative solver need to be develop. This research first proposed a fast optimization based multilevel algorithm in 3-D formulation that has the optimal complexity to solve the recent 3-D selective segmentation model. To achieve faster convergence, we reformulated the recent 3-D model into a new localized model. The proposed multilevel algorithm was used to solve the new localized model. MATLAB coding was developed to implement the segmentation process. The accuracy of the segmented image was evaluated using the Jaccard similarity measure. The execution time was recorded to measure the efficiency of the models. Test results demonstrated that our model is capable of successfully segmenting a targeted object in optimal computational time. Advantages of our algorithm can be seen in processing large size of 3-D images, where magnitudes of speed-up are observed over competing algorithms.