3D image spectral segmentation based on the region mean histogram

Spectral segmentation algorithms can extract the global impression of an image and are widely used in many areas and applications related to image segmentation. Traditional spectral algorithms need to construct affinity matrix based on all the voxels in the 3D image and compute the eigenvectors of the matrix to find the global optimum segmentation. With the growing size of the image, the spatial and computational complexities increase egregiously. We propose a novel approach for solving the problem. Rather than focusing on voxel of the 3D image, our approach aims at the gray histogram of the mean of the sub regions generated by region growing method. Our method decreases the affinity matrix dimension to 256×256 at most for gray images, and thus decreases the complexity sharply. The spatial and computational complexities are affected slightly by the image size. We have applied this approach to segmenting 3D Printed Circuit Board (PCB) CT images and found the segmentation results to be very encouraging.