DTI IMAGES SEGMENTATION BASED ON ADAPTIVE BANDWIDTH MEAN SHIFT ALGORITHM

Traditional mean shift image segmentation algorithms need selecting fixed bandwidth manually, which leads to under-segmentation and non-global optimum. To overcome these disadvantages, a bandwidth adaptive mean shift algorithm is proposed. In this algorithm, a new bandwidth window function is defined, with the bandwidth is determined automatically based on probability distribution characteristics of the pixels, namely, small bandwidths are applied in the large density data point areas, and big bandwidths are for the small density data point areas. Based on the new bandwidth window function, new mean shift vectors are acquired, and the new mean shift vectors and sample mean iterations are adopted to determine and implement cluster. The algorithm does not need artificially selecting bandwidth, and hence can improve the efficiency and decrease under-segmentation and inaccurate clustering. Based on the proposed algorithm, experiments are designed to segment the corpus callosum of different layers' human brain DTI images, with the results showing that the adaptive bandwidth segmentation algorithm has better segmentation effect.