Tham khảo tài liệu 'recent advances in signal processing 2011 part 3', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | Methods for Nonlinear Intersubject Registration in Neuroscience 57 image is unknown. Their similarity is computed as a linear combination of similarities of intensity pairs corresponding to the points in the neighbour-hood of the examined point. Fig. 3. The scheme of the block matching algorithm proposed for coarse spatial normalization. 58 Recent Advances in Signal Processing M xw Xw uw Xw-Uw N xw Fig. 4. Illustration of regional symmetric matching. The similarity is measured in the forward the blue line as well as in the reverse the green line direction of registration. In the forward direction the displacement field computed so far is applied on the floating image voxels. In the reverse direction the inverse displacement field is applied on the reference image voxels. The extent of the neighbourhood depends on the chosen kernel function. Here the first-order the second-order and the third-order B-spline functions with 8 27 and 64 grid points in neighbourhood for 3-D tasks or 4 9 and 16 points in neighbourhood for 2-D tasks are used. The particular choice of the kernel function affects the smoothness of the behaviour of the regional similarity measure see Fig. 5. The number of local optima is the lowest in the case of the third-order B-spline. As the evaluation of the B-splines increases the computational load their values are computed only once and stored in a lookup table with increments equal to . Fig. 5. Comparison of the regional similarity measure computed with the use of GPV and the first-order B-spline solid line the second-order B-spline dashed line and the third-order B-spline dotted line . A region of the size a 10x10 mm b 20x20 mm was translated by yx 10mm in the x direction. Local translations which maximize a matching criterion are searched in optimization procedures. Here the symmetric regional similarity measure is used as the matching criterion which has to be maximized Methods for Nonlinear Intersubject Registration in Neuroscience 59 S fw
