Computer Vision News - March 2017

28 Computer Vision News Research Research As before, S is the set of pixels belonging to the segmentation, since is an indication of a pixel’s membership of a cluster. R is the ‘membership’ difference between the pixels inside the region selected for segmentation and the pixels outside the region. R receives values between -1 and 1. A negative value indicates a relatively homogenous segmentation region, and a positive value a heterogeneous region, which the iterative process will tend to shrink. The next component of the equation is G, defined as follows: is a matrix - with every element of the matrix representing a pixel in the image that is either pulling or pushing the contour of the object. is the local factor of the equation; unlike R it is not influenced by the pixels inside the region or outside the region, but by the local pixels only. γ is a parameter between 0 and 1. 0 is the balloon force. The last component is E: Where g is the normalized edge indicator, based on the image gradient = − ( ( , the purpose of which is to give weight in the equation to edges - based on the assumption that the segmentation will most likely follow edges in the image. The principal steps implementing this level set model for selective image segmentation can be summarized as follows: = [1 − (2 − 1 ] 0 1. Conduct FCM. 2. Choose the candidate objects of interest [ ]. 3. Compute the enhanced object indication function E and the signed balloon force g. 4. Initialize the dynamic interface . 5. Compute ∇ φ, 1φ, H(φ) and δ(φ). 6. Compute the force of fuzzy region competition R. 7. Evolve and regularize the dynamic interface φ. 8. If not converging, go back to step 5 and repeat. = 10 ( ⋅ ,(1−