Gabor filter banks have been recognized as an effective tool for myocardium localization in tagged MR images. The bank of Gabor filters is designed considering different frequencies, directions, and tag spacings around the original frequency, direction, and tag spacing settings of the MRI scanner. All filters in the bank are applied to the input image and the filters whose parameters best match the local tag pattern in the image produce higher magnitude responses. By combining parameters of the filters with the highest magnitude responses and applying the resulting filter to the image, an optimal local response can be achieved.
The Gabor filter bank was generated with orientation angles theta = theta_int + n_1 x Delta_theta where the initial orientation angles are theta_int = pi/4, -pi/4 for 45 degree and -45 degree tag modulations, n_1 = 0, +- 1, +-2, and minimal angular difference of Delta_theta = pi/18. With tag spacings of 5.27 pixels, the central frequencies were defined by u = u_0 + n_2 x u_0 and v = v_0 + n_2 x v_0 where initial central frequencies are u_0 = v_0 = 1/S = 0.1897 and n_2 = 0, +-0.1, +-0.2.