python胶囊瑕疵检测胶囊瑕疵检测opencv胶囊表面划痕检测完整项目源码

import cv2 import numpy as np import math import copy def produceThresholds(crop_img): crop = copy.deepcopy(crop_img) thresholds = [] threshold0 = Wolf(crop, 3, 18, 18, 0.05 + (0 * 0.35), 128) threshold0 = cv2.bitwise_not(threshold0) thresholds.append(threshold0) threshold1 = Wolf(crop, 3, 22, 22, 0.05 + (1 * 0.35), 128) threshold1 = cv2.bitwise_not(threshold1) thresholds.append(threshold1) threshold2 = Wolf(crop, 2, 12, 12, 0.18, 128) threshold2 = cv2.bitwise_not(threshold2) thresholds.append(threshold2) return thresholds def calcLocalStats(im, map_m, map_s, winx, winy): rows, cols = im.shape im_sum, im_sum_sq = cv2.integral2(im, sqdepth=cv2.CV_64F) wxh = winx / 2 wyh = winy / 2 x_firstth = wxh y_lastth = rows - wyh - 1 y_firstth = wyh winarea = float(winx * winy) max_s = 0 j = y_firstth for j in range(y_firstth, y_lastth + 1): sum = 0.0 sum_sq = 0.0 sum = im_sum.item(j - wyh + winy, winx) - im_sum.item(j - wyh, winx) - \ im_sum.item(j - wyh + winy, 0) + im_sum.item(j - wyh, 0) sum_sq = im_sum_sq.item(j - wyh + winy, winx) - im_sum_sq.item( j - wyh, winx) - im_sum_sq.item(j - wyh + winy, 0) + im_sum_sq.item(j - wyh, 0) m = sum / winarea s = math.sqrt((sum_sq - m * sum) / winarea) if s > max_s: max_s = s map_m.itemset((j, x_firstth), m) map_s.itemset((j, x_firstth), s) maxrange = cols - winx + 1 for i in range(1, maxrange): sum -= im_sum.item(j - wyh + winy, i) - im_sum.item(j - wyh, i) - im_sum.item(j - wyh + winy, i - 1) \ + im_sum.item(j - wyh, i - 1) sum += im_sum.item(j - wyh + winy, i + winx) - im_sum.item(j - wyh, i + winx) - \ im_sum.item(j - wyh + winy, i + winx - 1) + \ im_sum.item(j - wyh, i + winx - 1) sum_sq -= im_sum_sq.item(j - wyh + winy, i) - im_sum_sq.item( j - wyh, i) - im_sum_sq.item(j - wyh + winy, i - 1) + im_sum_sq.item(j - wyh, i - 1) sum_sq += im_sum_sq.item(j - wyh + winy, i + winx) - im_sum_sq.item(j - wyh, i + winx) - \ im_sum_sq.item(j - wyh + winy, i + winx - 1) + \ im_sum_sq.item(j - wyh, i + winx - 1) m = sum / winarea s = math.sqrt(abs(sum_sq - m * sum) / winarea) if s > max_s: max_s = s map_m.itemset((j, i + wxh), m) map_s.itemset((j, i + wxh), s) return max_s, map_m, map_s, im def Wolf(image, version, winx, winy, k, dR): img = copy.deepcopy(image) m = 0.0 max_s = 0.0 th = 0 wxh = winx / 2 wyh = winy / 2 x_firstth = wxh x_lastth = img.shape[1] - wxh - 1 y_lastth = img.shape[0] - wyh - 1 y_firstth = wyh rows, cols = img.shape[:2] #output = np.zeros((rows, cols), np.uint8) output = np.uint8(img) # print img.dtype # print rows, cols map_m = np.zeros((rows, cols), dtype=float) map_s = np.zeros((rows, cols), dtype=float) max_s, map_m, map_s, img = calcLocalStats(img, map_m, map_s, winx, winy) # Finds the global minimum and maximum in an array min_I, max_I, _, _ = cv2.minMaxLoc(img) # # for i in range(0, map_s.shape[0]): # for j in range(0, map_s.shape[1]): # print map_s.item(i, j) # map_s = np.uint8(map_s) # cv2.imshow('12', map_s) thsurf = np.zeros((rows, cols), dtype=float) for j in range(y_firstth, y_lastth + 1): for i in range(0, cols - winx + 1): m = float(map_m.item(j, i + wxh)) s = float(map_s.item(j, i + wxh)) if version == 1: th = m + k * s elif version == 2: th = m * (1 + k * (s / dR - 1)) else: th = m + k * (s / max_s - 1) * (m - min_I) thsurf.itemset((j, i + wxh), th) if i == 0: for l in range(0, x_firstth + 1): thsurf.itemset((j, l), th) if j == y_firstth: for u in range(0, y_firstth): for t in range(0, x_firstth + 1): thsurf.itemset((u, t), th) if j == y_lastth: for u in range(y_lastth + 1, rows): for l in range(0, x_firstth + 1): thsurf.itemset((u, l), th) if j == y_firstth: for u in range(0, y_firstth): thsurf.itemset((u, i + wxh), th) if j == y_lastth: for u in range(y_lastth + 1, rows): thsurf.itemset((u, i + wxh), th) # right border for i in range(x_lastth, cols): thsurf.itemset((j, i), th) # right-upper corner if j == y_firstth: for u in range(0, y_firstth): for l in range(x_lastth, cols): thsurf.itemset((u, l), th) # right-lower corner if j == y_lastth: for u in range(y_lastth + 1, rows): for l in range(x_lastth, cols): thsurf.itemset((u, l), th) for y in range(0, rows): for x in range(0, cols): if img.item(y, x) >= thsurf.item(y, x): output.itemset((y, x), 255) else: output.itemset((y, x), 0) return output def otsu(img): blur = cv2.GaussianBlur(img, (5, 5), 0) ret3, th3 = cv2.threshold( blur, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU) return th3