Super Resolution and Brightfield Fourier Transforms

Author: chekfung

brightfield_helper.py

@@ -0,0 +1,46 @@
+import sys
+import numpy as np
+import matplotlib.pyplot as plt
+import cv2
+
+def show_hist(main_bf, test_bf, show_histograms):
+    if show_histograms:
+        # Main Brightfield Image 
+        hist,bins = np.histogram(main_bf.flatten(),256,[0,256])
+        cdf_normalized = hist.cumsum() * float(hist.max()) / hist.cumsum().max()
+        fig, ax = plt.subplots(1,2)
+        ax[0].plot(cdf_normalized, color = 'b')
+        ax[0].hist(main_bf.flatten(),256,[0,256], color = 'r')
+        ax[0].legend(('cdf','histogram'), loc = 'upper left')
+        ax[0].set_title("Main BF Histogram")
+
+        # Test Brightfield Image
+        hist,bins = np.histogram(test_bf.flatten(),256,[0,256])
+        cdf_normalized = hist.cumsum() * float(hist.max()) / hist.cumsum().max()
+        ax[1].plot(cdf_normalized, color = 'b')
+        ax[1].hist(test_bf.flatten(),256,[0,256], color = 'r')
+        ax[1].legend(('cdf','histogram'), loc = 'upper left')
+        ax[1].set_title("Test BF Histogram")
+
+        plt.show()
+
+def show_equal(main_bf, equ_main, test_bf, equ_test, show_equalized):
+    if show_equalized:
+        # TODO: Put this in a function and another file so that it is easier to read
+        fig = plt.figure(2)
+        fig.add_subplot(2,2,1)
+        plt.title("Main Brightfield")
+        plt.imshow(main_bf)
+
+        fig.add_subplot(2,2,2)
+        plt.title("Equalized Main Brightfield")
+        plt.imshow(equ_main)
+        
+        fig.add_subplot(2,2,3)
+        plt.title("Test Brightfield")
+        plt.imshow(test_bf)
+
+        fig.add_subplot(2,2,4)
+        plt.title("Equalized Test Brightfield")
+        plt.imshow(equ_test)
+        plt.show()

brightfield_match_attempt_2.py

@@ -2,6 +2,7 @@
 import numpy as np
 import matplotlib.pyplot as plt
 import cv2
+from brightfield_helper import show_hist, show_equal
 
 # MACROS FOR SHOWING IMAGES OF WHAT IS HAPPENING
 show_histograms = False
@@ -11,87 +12,67 @@
 # Import Clean and TestBrightfield Image
 main_bf_c = cv2.imread("data/brightfield_main_minerva.tif")
 test_bf_c = cv2.imread("data/f0113/F0113_10012021_initial_BF.tif")
-#main_bf_c = cv2.imread("data/EGaudi_1.jpg")
-#test_bf_c = cv2.imread("data/EGaudi_2.jpg")
 
 # Convert to grayscale
 main_bf = cv2.cvtColor(main_bf_c, cv2.COLOR_BGR2GRAY)
 test_bf = cv2.cvtColor(test_bf_c, cv2.COLOR_BGR2GRAY)
 
-kernel_size = 5
-main_bf = cv2.GaussianBlur(main_bf, (kernel_size, kernel_size), 0)
-test_bf = cv2.GaussianBlur(test_bf, (kernel_size, kernel_size), 0)
-# plt.imshow(main_bf)
+# f = np.fft.fft2(test_bf)
+# fshift = np.fft.fftshift(f)
+# magnitude_spectrum = 20*np.log(np.abs(fshift))
+# plt.subplot(121),plt.imshow(test_bf, cmap = 'gray')
+# plt.title('Input Image')
+# plt.subplot(122),plt.imshow(magnitude_spectrum, cmap = 'gray')
+# plt.title('Magnitude Spectrum')
 # plt.show()
 
-# dst = cv2.cornerHarris(main_bf, 15, 15, 0.03)
-# dst = cv2.dilate(dst, None) 
-# main_bf_c[dst > 0.01 * dst.max()]=[255,0,0]
-# plt.imshow(main_bf_c)
+# f = np.fft.fft2(main_bf)
+# fshift = np.fft.fftshift(f)
+# magnitude_spectrum = 20*np.log(np.abs(fshift))
+# plt.subplot(121),plt.imshow(main_bf, cmap = 'gray')
+# plt.title('Input Image')
+# plt.subplot(122),plt.imshow(magnitude_spectrum, cmap = 'gray')
+# plt.title('Magnitude Spectrum')
 # plt.show()
 
-# Equalize Histogram to get Same Brightness and Contrast of Both Images
-if show_histograms:
-    # TODO: Put this in a function and another file so that it is easier to read
-    # Main Brightfield Image 
-    hist,bins = np.histogram(main_bf.flatten(),256,[0,256])
-    cdf_normalized = hist.cumsum() * float(hist.max()) / hist.cumsum().max()
-    fig, ax = plt.subplots(1,2)
-    ax[0].plot(cdf_normalized, color = 'b')
-    ax[0].hist(main_bf.flatten(),256,[0,256], color = 'r')
-    ax[0].legend(('cdf','histogram'), loc = 'upper left')
-    ax[0].set_title("Main BF Histogram")
-
-    # Test Brightfield Image
-    hist,bins = np.histogram(test_bf.flatten(),256,[0,256])
-    cdf_normalized = hist.cumsum() * float(hist.max()) / hist.cumsum().max()
-    ax[1].plot(cdf_normalized, color = 'b')
-    ax[1].hist(test_bf.flatten(),256,[0,256], color = 'r')
-    ax[1].legend(('cdf','histogram'), loc = 'upper left')
-    ax[1].set_title("Test BF Histogram")
 
-    plt.show()
 
+kernel_size = 5
+main_bf = cv2.GaussianBlur(main_bf, (kernel_size, kernel_size), 0)
+#main_bf = main_bf - main_bf_blur
+test_bf = cv2.GaussianBlur(test_bf, (kernel_size, kernel_size), 0)
+#test_bf = test_bf - test_bf_blur
+
+print("MEOW")
+# Equalize Histogram to get Same Brightness and Contrast of Both Images
+show_hist(main_bf, test_bf, show_histograms)
 equ_main = cv2.equalizeHist(main_bf)
 equ_test = cv2.equalizeHist(test_bf)
+show_equal(main_bf, equ_main, test_bf, equ_test, show_equalized)
 
-if show_equalized:
-    # TODO: Put this in a function and another file so that it is easier to read
-    fig = plt.figure(2)
-    fig.add_subplot(2,2,1)
-    plt.title("Main Brightfield")
-    plt.imshow(main_bf)
-
-    fig.add_subplot(2,2,2)
-    plt.title("Equalized Main Brightfield")
-    plt.imshow(equ_main)
-    
-    fig.add_subplot(2,2,3)
-    plt.title("Test Brightfield")
-    plt.imshow(test_bf)
-
-    fig.add_subplot(2,2,4)
-    plt.title("Equalized Test Brightfield")
-    plt.imshow(equ_test)
-    plt.show()
+print("MEOW1")
 
 # Apply Feature Match Algorithm between Brightfield Images
 sift_detect = cv2.SIFT_create()
 kp1,d1 = sift_detect.detectAndCompute(equ_main, None)
 kp2,d2 = sift_detect.detectAndCompute(equ_test, None)
 
+print("MEOW2")
+
 FLAN_INDEX_KDTREE = 0
 index_params = dict (algorithm = FLAN_INDEX_KDTREE, trees=5)
-search_params = dict (checks=50)
+search_params = dict(checks=50)
 
 flann = cv2.FlannBasedMatcher(index_params, search_params)
-
 matches = flann.knnMatch(d1, d2, k=2)
 
+print("MEOW3")
+
 # Determine Uniqueness of Flann Matches using ratio test
+RATIO = 0.6
 unique_matches = []
 for a,b in matches:
-    if a.distance < 0.6 * b.distance:
+    if a.distance < RATIO * b.distance:
         unique_matches.append(a)
 
 if draw_matches:
@@ -110,8 +91,8 @@
 dst_pts = np.float32([kp2[m.trainIdx].pt for m in unique_matches]).reshape(-1,1,2)
 
 M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC,5.0)
-print(M)
 
+print("MEOW4")
 # All this stuff is just to have
 matchesMask = mask.ravel().tolist()
 h,w = main_bf.shape
@@ -127,6 +108,36 @@
 plt.show()
 
 # With Matrix, we can find chip corner, knowing where it is on clean brightfield
+# TLC, TRC, BLC, BRC
+CLEAN_BRIGHTFIELD_CORNERS = [(241, 765), (1827, 802), (224, 1561), (1810, 1596)]
+NEW_CORNERS = [(0,0), (1600, 0), (0, 800), (1600, 800)]
+
+# Identify chip corners on experimental setup
+bf_test_corners = cv2.perspectiveTransform(np.float32(CLEAN_BRIGHTFIELD_CORNERS).reshape(-1,1,2), M)
+bf_test_corners = sum(np.int32(bf_test_corners.round()).tolist(), [])
+
+print(bf_test_corners)
+
+for (x,y) in bf_test_corners:
+    test_bf_c = cv2.circle(test_bf_c, (x,y), radius=3, color=(255, 0, 0), thickness=-1)
+
+plt.imshow(test_bf_c)
+plt.show()
+
+for (x,y) in CLEAN_BRIGHTFIELD_CORNERS:
+    main_bf_c = cv2.circle(main_bf_c, (x,y), radius=0, color=(255, 0, 0), thickness=-1)
+
+plt.imshow(main_bf_c)
+plt.show()
+
+M_clean_to_sense = cv2.getPerspectiveTransform(np.float32(CLEAN_BRIGHTFIELD_CORNERS),np.float32(NEW_CORNERS))
+dst = cv2.warpPerspective(main_bf_c,M_clean_to_sense,(1600,800))
+
+M_test_to_clean = cv2.getPerspectiveTransform(np.float32(bf_test_corners), np.float32(NEW_CORNERS))
+dst2 = cv2.warpPerspective(test_bf_c, M_test_to_clean, (1600,800))
+
+plt.imshow(dst2)
+plt.show()  
 
 # Multiply Affine Matrix with clean brightfield image to get to square pixels