import sys import cv2 import numpy as np import math from os.path import exists, basename from common import image_resize, display, normalize_angle from json import load,dumps DEBUG = True #clockwise from top left order = [ 0, 2, 3, 5, 4, 1 ] matchMethods = ['cv2.TM_CCOEFF', 'cv2.TM_CCOEFF_NORMED', 'cv2.TM_CCORR', 'cv2.TM_CCORR_NORMED', 'cv2.TM_SQDIFF', 'cv2.TM_SQDIFF_NORMED'] def read_text (textPath) : holePunches = {} with open(textPath) as json: holePunches = load(json) return holePunches # # CALIBRATE # if len(sys.argv) < 2: print('Please provide path of normalized scan to calibrate to') exit(1) normalImage = sys.argv[-1] if not exists(normalImage) : print('Normalized scan does not exist, please provide one that does') exit(2) normalText = normalImage + '.json' if not exists(normalText) : print('Corresponding normalized scan text does not exist, please generate one') exit(3) print(f'Calibrating to scan {basename(normalImage)}') registrationMark = cv2.imread('./registrationMark.png', 0) w, h = registrationMark.shape[:2] holePunches = read_text(normalText) original = cv2.imread(normalImage) img = original.copy() height, width = img.shape[:2] orientation = height > width marks = [] if not orientation : print(f'Scan is not in portrait mode, exiting...') exit(3) #print(holePunches) def get_distance(ref, point): # print('ref: {} , point: {}'.format(ref, point)) x1, y1 = ref[0], ref[1] x2, y2 = point[0], point[1] return math.hypot(x2 - x1, y2 - y1) def group_points(points): groups = {} groupnum = 0 while len(points) > 1: groupnum += 1 key = str(groupnum) groups[key] = [] ref = points.pop(0) for i, point in enumerate(points): d = get_distance(ref, point) if d < 30: groups[key].append(points[i]) points[i] = None points = list(filter(lambda x: x is not None, points)) return list([[int(np.mean(list([x[0] for x in groups[arr]]))), int(np.mean(list([x[1] for x in groups[arr]])))] for arr in groups]) def find_closest (pt, pts) : return pts[min(range(len(pts)), key = lambda i: get_distance(pts[i], pt))] def find_in_half (half) : halfGray = cv2.cvtColor(half, cv2.COLOR_BGR2GRAY) res = cv2.matchTemplate(halfGray, registrationMark, cv2.TM_CCOEFF_NORMED) threshold = 0.7 loc = np.where( res >= threshold) for pt in zip(*loc[::-1]): cv2.rectangle(half, pt, (pt[0] + w, pt[1] + h), (0,0,255), 2) return list(zip(*loc[::-1])) ttly = holePunches['0']['y']-round(height*0.05) ttlx = holePunches['0']['x'] topHalf = img[ttly:holePunches['1']['y']+round(height*0.1), ttlx:holePunches['2']['x']] topHalfPts = find_in_half(topHalf) thpts = group_points(topHalfPts) for pt in thpts : #print(f'{ttlx + pt[0]},{ttly + pt[1]}') marks.append((ttlx + pt[0] + round(w / 2), ttly + pt[1] + round(h / 2),)) print(f'Found {len(thpts)} points') #display(topHalf) btly = holePunches['4']['y']-round(height*0.1) btlx = holePunches['3']['x'] bottomHalf = img[btly:holePunches['5']['y']+round(height*0.05), btlx:holePunches['4']['x']] bottomHalfPts = find_in_half(bottomHalf) bhpts = group_points(bottomHalfPts) for pt in bhpts : #print(f'{btlx + pt[0]},{btly + pt[1]}') marks.append((btlx + pt[0] + round(w / 2), btly + pt[1] + round(h / 2), )) clean = original.copy() for pt in marks : #print(pt) cv2.circle(clean, pt, 50, (0,0,255), -1) print(f'Found {len(bhpts)} points') if len(marks) != 16 : print(f'{len(marks)} != 16 marks') exit(1) ## # TOP LEFT ## topLeftTR = find_closest((0,0,), marks) #print(topLeftTR) topLeftSorted = sorted(marks, key=lambda e: get_distance(e, topLeftTR)) topLeftSorted.remove(topLeftTR) if abs(topLeftSorted[0][1] - topLeftTR[1]) < abs(topLeftSorted[1][1] - topLeftTR[1]) : topLeftBR = topLeftSorted[0] else : topLeftBR = topLeftSorted[1] topLeftTL = topLeftSorted[2] topLeftBL = topLeftSorted[4] if DEBUG : cv2.line(clean, topLeftTR, topLeftBR, (0,0,255,), 3) cv2.line(clean, topLeftTR, topLeftTL, (0,0,255,), 3) cv2.line(clean, topLeftBR, topLeftBL, (0,0,255,), 3) cv2.line(clean, topLeftTL, topLeftBL, (0,0,255,), 3) ## # TOP RIGHT ## topRightTL = find_closest((width,0,), marks) #print(topRightTL) topRightSorted = sorted(topLeftSorted, key=lambda e: get_distance(e, topRightTL)) topRightSorted.remove(topRightTL) topRightSorted.remove(topLeftBR) topRightSorted.remove(topLeftTL) topRightSorted.remove(topLeftBL) topRightBL = topRightSorted[0] topRightTR = topRightSorted[1] if abs(topRightTL[0] - topRightSorted[2][0]) > abs(topRightTL[0] - topRightSorted[3][0]) : topRightBR = topRightSorted[2] else : topRightBR = topRightSorted[3] if DEBUG : cv2.line(clean, topRightTR, topRightBR, (0,0,255,), 3) cv2.line(clean, topRightTR, topRightTL, (0,0,255,), 3) cv2.line(clean, topRightBR, topRightBL, (0,0,255,), 3) cv2.line(clean, topRightTL, topRightBL, (0,0,255,), 3) ## # BOTTOM LEFT ## bottomLeftTL = find_closest((0,height,), marks) bottomLeftSorted = sorted(topRightSorted, key=lambda e: get_distance(e, bottomLeftTL)) bottomLeftSorted.remove(bottomLeftTL) bottomLeftSorted.remove(topRightBR) bottomLeftSorted.remove(topRightTR) bottomLeftSorted.remove(topRightBL) bottomLeftBL = bottomLeftSorted[0] bottomLeftTR = bottomLeftSorted[2] bottomLeftBR = bottomLeftSorted[3] if DEBUG : cv2.line(clean, bottomLeftTR, bottomLeftBR, (0,0,255,), 3) cv2.line(clean, bottomLeftTR, bottomLeftTL, (0,0,255,), 3) cv2.line(clean, bottomLeftBR, bottomLeftBL, (0,0,255,), 3) cv2.line(clean, bottomLeftTL, bottomLeftBL, (0,0,255,), 3) ## # BOTTOM RIGHT ## bottomRightTR = find_closest((width,height,), marks) bottomRightSorted = sorted(bottomLeftSorted, key=lambda e: get_distance(e, bottomRightTR)) bottomRightSorted.remove(bottomRightTR) bottomRightSorted.remove(bottomLeftBR) bottomRightSorted.remove(bottomLeftTR) bottomRightSorted.remove(bottomLeftBL) bottomRightBR = bottomRightSorted[0] bottomRightTL = bottomRightSorted[1] bottomRightBL = bottomRightSorted[2] if DEBUG : cv2.line(clean, bottomRightTR, bottomRightBR, (0,0,255,), 3) cv2.line(clean, bottomRightTR, bottomRightTL, (0,0,255,), 3) cv2.line(clean, bottomRightBR, bottomRightBL, (0,0,255,), 3) cv2.line(clean, bottomRightTL, bottomRightBL, (0,0,255,), 3) jsonOut = { 'width' : width, 'height' : height, 'holePunches' : holePunches, '0' : { '0' : { 'x' : bottomLeftTL[0], 'y' : bottomLeftTL[1] }, '1' : { 'x' : bottomLeftTR[0], 'y' : bottomLeftTR[1] }, '2' : { 'x' : bottomLeftBR[0], 'y' : bottomLeftBR[1] }, '3' : { 'x' : bottomLeftBL[0], 'y' : bottomLeftBL[1] } }, '1' : { '0' : { 'x' : topLeftTL[0], 'y' : topLeftTL[1] }, '1' : { 'x' : topLeftTR[0], 'y' : topLeftTR[1] }, '2' : { 'x' : topLeftBR[0], 'y' : topLeftBR[1] }, '3' : { 'x' : topLeftBL[0], 'y' : topLeftBL[1] } }, '2' : { '0' : { 'x' : topRightTL[0], 'y' : topRightTL[1] }, '1' : { 'x' : topRightTR[0], 'y' : topRightTR[1] }, '2' : { 'x' : topRightBR[0], 'y' : topRightBR[1] }, '3' : { 'x' : topRightBL[0], 'y' : topRightBL[1] } }, '3' : { '0' : { 'x' : bottomRightTL[0], 'y' : bottomRightTL[1] }, '1' : { 'x' : bottomRightTR[0], 'y' : bottomRightTR[1] }, '2' : { 'x' : bottomRightBR[0], 'y' : bottomRightBR[1] }, '3' : { 'x' : bottomRightBL[0], 'y' : bottomRightBL[1] } } } calibrationFile = f'{normalImage}.calibration.json' with open(calibrationFile, 'w') as output: output.write(dumps(jsonOut, sort_keys = True, indent = 4)) print(f'Wrote calibration file to {calibrationFile}') display(clean)