"""
Created by: Rishav Raj and Qie Shang Pua, University of New South Wales
This program performs image processing to obtain multiple segments of a patient's spine.
Input: Ultrasound scans of the patient's back (in png format)
Output: 1. Processed Spine Images
2. A csv file that will be fed into registration.py
"""
# Import Packages
import os
import cv2
import glob
import math
import pandas as pd
import cv2
import numpy as np
import scipy.ndimage
import statistics
from skimage import data, color
import matplotlib.pyplot as plt
from skimage.color import rgb2gray
from skimage.transform import rescale, resize, downscale_local_mean
from timeit import default_timer as timer
import sys
from skimage.io import imread
# Python Migration of "find_best_path_jumping.m" of MATLAB
# Applies Dynamic Programming to calculate the path of least cost.
def find_best_path_jumping(curr):
x = 151 # these are hardcoded - should be determined instead.
y = 289
# if at end
if curr[1] == y:
curr_cost = 0
cost[int(curr[0]), int(curr[1])] = curr_cost # Casted into int to access array index
nexts[int(curr[0]), int(curr[1])] = 0
return
max_jump = 50
no_cells = max_jump * 2 + 1
min_next = [0, 0]
min_cost = math.inf # inf gives the largest num in float
# print(curr)
# print(int(curr[0]))
# calc costs
for i in range(no_cells):
# cell no
next = [0, 0]
next[0] = curr[0] + i - max_jump - 1
next[1] = curr[1] + 1
# print(next)
# out of bounds
if next[0] < 1 or next[0] > x:
next_cost = math.inf
# already calculated
elif cost[int(next[0]), int(next[1])] != math.inf:
next_cost = cost[int(next[0]), int(next[1])]
# need to recursively calculate
else:
next_cost = find_best_path_jumping(next)
# print(next_cost)
# add penalty if needed
if abs(curr[0] - next[0]) > 2:
if next_cost == None:
next_cost = 0
next_cost = next_cost + 0.2
# calculate running min
if next_cost != None and next_cost < min_cost:
min_cost = next_cost
min_next = next[0]
# set results
# print([int(curr[0]), int(curr[1])])
curr_cost = inv_prob[int(curr[0]), int(curr[1])] + min_cost
cost[int(curr[0]), int(curr[1])] = curr_cost
nexts[int(curr[0]), int(curr[1])] = min_next
return curr_cost
# Python Migration of "get_prob_map.m" of MATLAB
# Produces processed image in black and white
def get_prob_map(grayscale):
# Start prob map as simple intensity
intensity_map = rescale(grayscale, .5, anti_aliasing=False)
intensity_map = np.asarray(intensity_map)
prob_map = intensity_map * 0.5
# print(np.shape(prob_map))
# Create probablity map from intensity after gaussian filtering
gausian = cv2.GaussianBlur(prob_map, (5, 5), 5)
gausian = np.asarray(gausian)
# print(np.shape(gausian))
num = np.multiply(gausian, prob_map)
den = num + np.multiply((1 - gausian), (1 - prob_map))
prob_map = np.divide(num, den)
kernel1 = np.ones((3, 3), np.float32) / 9
kernel1[1][1] = 0.8888889
y = 1
x = 1
shadow = np.ones((y, x)) * 0.2
overall_mean = np.mean(np.mean(intensity_map))
for i in range(1, x):
j = y
while (j > 0 and (
gausian[j, i] < overall_mean * 1.5 or np.mean(gausian[j - 10:j - 1, i]) < overall_mean * 1.5)):
shadow[j, i] = 0.1
j = j - 1
while (j > 0 and (gausian[j, i] > overall_mean * 1.5) or j > 5 and np.mean(
gausian[j - 5:j - 1, i]) > overall_mean * 1.5):
shadow[j, i] = intensity_map(j, i)
j = j - 1
shadow = cv2.GaussianBlur(shadow, (5, 5), 5)
prob_map = (shadow * prob_map) / (shadow * prob_map + (1 - shadow) * (1 - prob_map))
return prob_map
k = cv2.waitKey(0) & 0xFF
if k == 27: # wait for ESC key to exit
cv2.destroyAllWindows()
elif k == ord('s'): # wait for 's' key to save and exit
cv2.imwrite('prob map of a single scan.png', prob_map)
cv2.destroyAllWindows()
# Main File
# Similar to single_line_path.m of MATLAB
def main():
points = []
images = glob.glob('/Users/puaqieshang/Desktop/Taste of Research/MATLAB code/everything/phantom_images/phantom_3/scan_2/*.png')
images.sort()
count = 0
startTime = timer()
for fname in images:
print(f"Image No.{count+1}")
img = cv2.imread(fname)
# im = plt.imread(fname)
# implot = plt.imshow(im)
count = count + 1
us_img = img[80:400, 270:850]
gray = cv2.cvtColor(us_img, cv2.COLOR_BGR2GRAY)
# cv2.imshow('gray_image',gray)
# [x, y] = np.shape(gray)
# print(x)
# print(y)
prob_map = get_prob_map(gray)
highlyLikely = 0.05*np.max(prob_map)
np.savetxt("prob_map.csv", prob_map, delimiter=",")
# print(np.shape(prob_map))
"""
Dynamic Programming Section
global variables are defined below
"""
global inv_prob
inv_prob = 0.5 - prob_map
[a, b] = np.shape(prob_map)
start = [np.floor(a/2), 1]
global cost
cost = np.ones([a, b]) * np.Inf
global nexts
nexts = np.ones([a, b]) * -1
find_best_path_jumping(start)
# get points on this scan
point_scan = np.zeros([1, b])
implot = plt.imshow(prob_map)
# print(start[0])
curr_x = start[0]
for curr_y in range(b):
point_scan[0, curr_y] = curr_x
if prob_map[int(curr_x), int(curr_y)] > highlyLikely: #NEED TO CHANGE TO HIGHLY LIKELY!!!!!
# print("helloooooo")
# cv2.circle(gray, (int(curr_y), int(curr_x)), 1, (0, 0, 255), 1)
# append coloured points
plt.scatter(curr_x, curr_y, c='r', s=20)
coloured_pt = [curr_x, curr_y, count * 0.2]
# points = np.concatenate((points, coloured_pt), axis=1)
points.append(coloured_pt)
else:
# cv2.circle(gray, (int(curr_y), int(curr_x)), 1, (255, 0, 0), 1)
plt.scatter(curr_x, curr_y, c='b', s=20)
curr_x = nexts[int(curr_x), int(curr_y)]
plt.show()
endTime = timer()
# print(str(endTime - startTime) + " seconds")
print(f"The time taken is {endTime - startTime} seconds")
np.savetxt("pls-work.csv", [*zip(*points)], delimiter=",") # Transpose points data and save into csv format
key = cv2.waitKey(0) & 0xFF
if key == ord("q"):
cv2.destroyAllWindows()
main()