def __init__(self,u): self.shape = u.shape self.nchannels =u.shape[2] self.uf = np.zeros( (2,) + self.shape ) self.ub = np.zeros( (2,) + self.shape ) for c in range(self.nchannels): self.uf[:,:,:,c] = FD.forward_differences(u[:,:,c]) self.ub[:,:,:,c] = FD.backward_differences(u[:,:,c])
def compute_tv(u): shape = u.shape uf = np.zeros( (2,) + shape ) for c in range(shape[2]): uf[:,:,:,c] = FD.forward_differences(u[:,:,c]) tv=0 for c in range( shape[2] ): tv+=np.sum( uf[1,:,:,c]**2 + uf[0,:,:,c]**2 )**0.5 return tv
def __init__(self, img):
self.shape = img.shape
self.nchannels = img.shape[2]
self.gradX = np.zeros((img.shape[0], img.shape[1], self.nchannels))
self.gradY = np.zeros((img.shape[0], img.shape[1], self.nchannels))
self.grad2X = np.zeros((img.shape[0], img.shape[1], self.nchannels))
self.grad2Y = np.zeros((img.shape[0], img.shape[1], self.nchannels))
for c in range(self.nchannels):
fd = FD.forward_differences(img[:, :, c])
self.gradX[:, :, c] += fd[0] #dx
self.gradY[:, :, c] += fd[1] #dy
fd2 = FD.forward_differences_second(img[:, :, c])
self.grad2X[:, :, c] += fd2[0] #dx
self.grad2Y[:, :, c] += fd2[1] #dy
def compute_gradient(A):
return FD.forward_differences(A)