def normalize_2D_xanes_rescale(img_stack, xanes_eng, pre_edge, post_edge):
x_eng = xanes_eng
pre_s, pre_e = pre_edge
post_s, post_e = post_edge
img = deepcopy(img_stack)
img_median = img_smooth(img, 3)
img_flat = img.flatten()
img_median_flat = img_median.flatten()
img_flat_dif = np.abs(img_flat - img_median_flat)
bad_pix_index = (img_flat_dif > 2)
img_flat[bad_pix_index] = 0
img_norm = img_flat.reshape(img.shape)
xs, xe = find_nearest(x_eng, pre_s), find_nearest(x_eng, pre_e)
img_pre_avg = np.mean(img_median[xs: max(xs+1, xe)], axis=0, keepdims=True)
img_pre_avg = img_smooth(img_pre_avg, 3)
xs, xe = find_nearest(x_eng, post_s), find_nearest(x_eng, post_e)
img_post_avg = np.mean(img_median[xs: max(xs + 1, xe)], axis=0, keepdims=True)
img_post_avg[np.abs(img_post_avg) < 1e-6] = 1e6
img_post_avg = img_smooth(img_post_avg, 3)
img_norm = (img_norm - img_pre_avg) / (img_post_avg - img_pre_avg)
img_norm = rm_abnormal(img_norm)
return img_norm
def normalize_2D_xanes_pre_edge(img_stack, xanes_eng, pre_edge):
pre_s, pre_e = pre_edge
img_norm = deepcopy(img_stack)
s0 = img_norm.shape
x_eng = xanes_eng
xs, xe = find_nearest(x_eng, pre_s), find_nearest(x_eng, pre_e)
if xs == xe:
img_pre = img_norm[xs].reshape(1, s0[1], s0[2])
img_pre = img_smooth(img_pre, 3)
img_norm = img_norm - img_pre
elif xe > xs:
eng_pre = x_eng[xs:xe]
img_pre = img_norm[xs:xe]
img_pre = img_smooth(img_pre, 3)
s = img_pre.shape
x_pre = eng_pre.reshape(len(eng_pre), 1)
x_bar_pre = np.mean(x_pre)
x_dif_pre = x_pre - x_bar_pre
SSx_pre = np.dot(x_dif_pre.T, x_dif_pre)
y_bar_pre = np.mean(img_pre, axis=0)
p = img_pre - y_bar_pre
for i in range(s[0]):
p[i] = p[i] * x_dif_pre[i]
SSxy_pre = np.sum(p, axis=0)
b0_pre = y_bar_pre - SSxy_pre / SSx_pre * x_bar_pre
b1_pre = SSxy_pre / SSx_pre
for i in range(s0[0]):
if not i % 10:
print(f'current image: {i}')
img_norm[i] = img_norm[i] - (b0_pre + b1_pre * x_eng[i])
img_norm = rm_abnormal(img_norm)
return img_norm
def normalize_2D_xanes2(img_stack, xanes_eng, pre_edge, post_edge, pre_edge_only_flag=0):
pre_s, pre_e = pre_edge
post_s, post_e = post_edge
s = img_stack.shape
xs_pre, xe_pre = find_nearest(xanes_eng, pre_s), find_nearest(xanes_eng, pre_e)
xs_post, xe_post = find_nearest(xanes_eng, post_s), find_nearest(xanes_eng, post_e)
xe_pre = max(xs_pre+1, xe_pre)
xe_post = max(xs_post+1, xe_post)
img_pre_mean = np.mean(img_stack[xs_pre:xe_pre], axis=0)
img_post_mean = np.mean(img_stack[xs_post:xe_post], axis=0)
img_post_mean = np.squeeze(img_smooth(img_post_mean, 5))
img_post_flat = np.sort(img_post_mean.flatten())
img_post_flat = img_post_flat[img_post_flat > 0]
n_post = len(img_post_flat)
if not n_post:
img_post_flat = np.sort(img_post_mean.flatten())
img_post_flat = img_post_flat[img_post_flat >= 0]
n_post = len(img_post_flat)
thresh_post = img_post_flat[int(n_post * 0.8)]
index_zero = img_post_mean < thresh_post
num_non_zero = np.sum(np.sum(1 - np.array(index_zero, dtype=int), axis=0), axis=0)
index_zero = np.repeat(index_zero.reshape([1, s[1], s[2]]), s[0], axis=0)
img = deepcopy(img_stack)
img[index_zero] = 0
x = xanes_eng
y = np.sum(np.sum(img, axis=1), axis=1)/num_non_zero
x1 = x[xs_pre: xe_pre]
y1 = y[xs_pre: xe_pre]
coef1 = polyfit(x1, y1, 1)
x1_mean = np.mean(x1)
x2 = x[xs_post: xe_post]
y2 = y[xs_post: xe_post]
coef2 = polyfit(x2, y2, 1)
coef2 = coef2 - coef1
x2_mean = np.mean(x2)
img_norm = deepcopy(img_stack)
for i in range(s[0]):
img_norm[i] = img_norm[i] - (coef1[1] * (x[i]-x1_mean) + img_pre_mean)
img_pre_mean = np.mean(img_norm[xs_pre:xe_pre], axis=0)
img_post_mean = np.mean(img_norm[xs_post:xe_post], axis=0)
img_post_mean = np.squeeze(img_smooth(img_post_mean, 5))
if not pre_edge_only_flag: # normalizing pre-edge only
for i in range(s[0]):
tmp = coef2[1] * (x[i]-x2_mean) + img_post_mean - img_pre_mean
tmp[tmp <= 0] = 1e6
img_norm[i] = img_norm[i] / tmp
img_thickness = img_post_mean - img_pre_mean
img_thickness[img_thickness<0] = 0
img_norm = rm_abnormal(img_norm)
return img_norm, img_thickness
def normalize_2D_xanes_rm_abornmal(img_stack):
img = deepcopy(img_stack)
img_median = img_smooth(img, 3)
img_flat = img.flatten()
img_median_flat = img_median.flatten()
img_flat_dif = np.abs(img_flat - img_median_flat)
bad_pix_index = (img_flat_dif > 2)
img_flat[bad_pix_index] = 0
img_norm = img_flat.reshape(img.shape)
return img_norm
def normalize_2D_xanes_post_edge(img_stack, xanes_eng, post_edge):
post_s, post_e = post_edge
img_norm = deepcopy(img_stack)
s0 = img_norm.shape
x_eng = xanes_eng
xs, xe = find_nearest(x_eng, post_s), find_nearest(x_eng, post_e)
if xs == xe:
img_post = img_norm[xs].reshape(1, s0[1], s0[2])
img_post = img_smooth(img_post, 3)
img_norm = img_norm / img_post
img_norm[np.isnan(img_norm)] = 0
img_norm[np.isinf(img_norm)] = 0
elif xe > xs:
eng_post = x_eng[xs:xe]
img_post = img_norm[xs:xe]
img_post = img_smooth(img_post, 3)
s = img_post.shape
x_post = eng_post.reshape(len(eng_post), 1)
x_bar_post = np.mean(x_post)
x_dif_post = x_post - x_bar_post
SSx_post = np.dot(x_dif_post.T, x_dif_post)
y_bar_post = np.mean(img_post, axis=0)
p = img_post - y_bar_post
for i in range(s[0]):
p[i] = p[i] * x_dif_post[i]
SSxy_post = np.sum(p, axis=0)
b0_post = y_bar_post - SSxy_post / SSx_post * x_bar_post
b1_post = SSxy_post / SSx_post
for i in range(s0[0]):
tmp = np.abs(b0_post + b1_post * x_eng[i])
tmp[tmp<1e-6] = 1e6
img_norm[i] = img_norm[i] / tmp
else:
print('check pre-edge/post-edge energy')
img_norm = rm_abnormal(img_norm)
return img_norm