Normalised cuts implementation in image segmentation
has a drawback in its computation. An image that has a size
of m × n pixels will require the normalised cuts algorithm to
form the W matrix with a size of (m × n) × (m × n). For
example, an image with a size of 120 × 90 will end up
computing a 10800 × 10800 adjacency matrix, which is
considered a very large matrix for a computer to solve for
eigenvalues. This indicates the size of W is proportionally
to the square of the image size. With the current digital
camera that can produce an image with image size more
than 10 mega pixels, doing image segmentation using
normalised cuts algorithm in this image size would be
impractical. An image segmentation done in hierarchical
manner is proposed here to counter the issue.