Implementation of align image

dscnsec/alignment/align_images.py

@@ -1 +1,201 @@
+# Import package
+import cv2
+import numpy as np
+
+
+def nothing(x):# callback function which is executed everytime trackbar value changes.
+    pass
+
+# trackbars
+cv2.namedWindow('trackbars')
+cv2.createTrackbar('minVal','trackbars',0,500,nothing)  # 1.tracbar name
+cv2.createTrackbar('maxVal','trackbars',0,1000,nothing) # 2.window name
+
+# Set trackbar
+cv2.setTrackbarPos('minVal','trackbars',0)
+cv2.setTrackbarPos('maxVal','trackbars',300)
+
+# Read image
+image = cv2.imread('images/1.jpg')
+print("Width : {} and height : {}".format(image.shape[1],image.shape[0]))
+cv2.imshow('original image',image)
+
+# Resize the image
+resized_image = cv2.resize(image,(900,700))
+heightImg, widthImg, channel = resized_image.shape
+
+def automatic_canny(images, sigma=0.33):
+    median = np.median(images)
+
+    ## Based on some statistics
+    lower = int(max(0, (1-sigma)*median))
+    upper = int(min(255, (1+sigma)*median))
+    edge = cv2.Canny(images, lower, upper)
+    return edge
+
+def getBiggestContour(contours):
+    biggest = np.array([])
+    max_area = 0
+    for i in contours:
+        area = cv2.contourArea(i)
+        if area>5000:
+            perimeter = cv2.arcLength(i,True)
+            approx = cv2.approxPolyDP(i, 0.02*perimeter, True)
+            if area > max_area and len(approx)==4:
+                biggest = approx
+                max_area = area
+    return biggest,max_area
+
+def getOrderOfLabel(points):
+
+    points = points.reshape((4,2))
+    new_points = np.zeros((4,1,2),dtype=np.int32)
+    add = points.sum(1)
+
+    new_points[0] = points[np.argmin(add)]
+    new_points[3] = points[np.argmax(add)]
+
+    diff = np.diff(points,axis=1)
+    new_points[1] = points[np.argmin(diff)]
+    new_points[2] = points[np.argmax(diff)]
+
+    return new_points    
+
+def drawRectangle(img,biggest,thickness):
+    cv2.line(img, (biggest[0][0][0], biggest[0][0][1]), (biggest[1][0][0], biggest[1][0][1]), (0, 255, 0), thickness)
+    cv2.line(img, (biggest[0][0][0], biggest[0][0][1]), (biggest[2][0][0], biggest[2][0][1]), (0, 255, 0), thickness)
+    cv2.line(img, (biggest[3][0][0], biggest[3][0][1]), (biggest[2][0][0], biggest[2][0][1]), (0, 255, 0), thickness)
+    cv2.line(img, (biggest[3][0][0], biggest[3][0][1]), (biggest[1][0][0], biggest[1][0][1]), (0, 255, 0), thickness)
+
+    return img
+
+while True:
+
+    # Grayscale
+    gray = cv2.cvtColor(resized_image,cv2.COLOR_BGR2GRAY)
+    cv2.imshow('gray',gray)
+
+    # Kernel size
+    kernel_size = (5,5)
+
+    # Gaussian blur 
+    gaussian_blur = cv2.GaussianBlur(gray,kernel_size,1)
+    cv2.imshow('Gaussian blur',gaussian_blur)
+
+    # Canny edge detection
+    '''
+    canny_edge = automatic_canny(gaussian_blur)
+    cv2.imshow('Canny edge',canny_edge)
+    '''
+
+    # Canny edge detection using trackbar
+    min_val = cv2.getTrackbarPos('minVal','trackbars')
+    max_val = cv2.getTrackbarPos('maxVal','trackbars')
+    canny_edge_trackbar = cv2.Canny(gaussian_blur,min_val,max_val)
+    cv2.imshow('Canny edge trackbar',canny_edge_trackbar)
+
+    # kernel for dilation and erosion
+    kernel = np.ones((5,5))
+
+    # Apply dilation
+    image_dilation = cv2.dilate(canny_edge_trackbar,kernel, iterations=2)        
+    cv2.imshow('Image Dilation',image_dilation)
+    '''
+    # Apply erosion
+    image_erosion = cv2.dilate(image_dilation,kernel, iterations=1)        
+    cv2.imshow('Image erosion',image_erosion)
+
+    # Apply closing
+    closing = cv2.morphologyEx(canny_edge_trackbar, cv2.MORPH_CLOSE, kernel)
+    cv2.imshow('Image closing',closing)
+    '''
+
+    # Find contours
+
+    # Copy image for display purpose
+    image_copy = resized_image.copy()
+
+    # Find all contours
+    contours, hierarchy = cv2.findContours(image_dilation, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # cv2.RETR_EXTERNAL helps to find the outer edges
+
+    # Draw all detected contours
+    cv2.drawContours(image_copy, contours, -1, (0, 255, 0), 10) # DRAW ALL DETECTED CONTOURS
+
+    # Result of contour
+    cv2.imshow('Contours',image_copy)
+
+    # Get only the biggest contour cordinates and area
+    biggest,maxArea = getBiggestContour(contours)
+
+    # Another sample image for display
+    image_copy_2 = resized_image.copy()
+    image_copy_3 = resized_image.copy()
+
+
+    if biggest.size !=0:
+
+        # Get the order in which we are detecting points of documents(page)
+        biggest_in_order = getOrderOfLabel(biggest)
+        cv2.circle(image_copy_2,(biggest_in_order[0][0][0],biggest_in_order[0][0][1]),5,(0,0,255),cv2.FILLED)
+        cv2.circle(image_copy_2,(biggest_in_order[1][0][0],biggest_in_order[1][0][1]),5,(0,0,255),cv2.FILLED)
+        cv2.circle(image_copy_2,(biggest_in_order[2][0][0],biggest_in_order[2][0][1]),5,(0,0,255),cv2.FILLED)
+        cv2.circle(image_copy_2,(biggest_in_order[3][0][0],biggest_in_order[3][0][1]),5,(0,0,255),cv2.FILLED)
+
+        # Filled circle at 4 corners points
+        cv2.drawContours(image_copy_3, biggest_in_order, -1, (0, 255, 0), 20)
+
+        # Make the lines from each point and draw it
+        image_copy_3 = drawRectangle(image_copy_3,biggest_in_order,2) 
+
+        # Prepare point for wrap
+        pts1 = np.float32(biggest_in_order)
+        pts2 = np.float32([[0,0],[widthImg,0],[0,heightImg],[widthImg,heightImg]])
+
+        # Map the points based on the position of points (i.e first corner = [0,0], second corner = [width,0]
+        # third corner = [0,height], forth corner = [width,height])
+        matrix = cv2.getPerspectiveTransform(pts1,pts2)
+        imageWrapColored = cv2.warpPerspective(resized_image,matrix,(widthImg,heightImg))
+
+        # Remove false positive ( 20 pixel from each side )
+        imageWrapColored = imageWrapColored[20:imageWrapColored.shape[0]-20,20:imageWrapColored.shape[1]-20]
+
+        # Convert BGR to Grayscale
+        ImageWrapGray = cv2.cvtColor(imageWrapColored,cv2.COLOR_BGR2GRAY)
+
+        # Apply Smoothing (Remove some noise in image)
+        ImageWrapGauss = cv2.GaussianBlur(ImageWrapGray,kernel_size,1)
+
+        # Adaptive thresolding
+        ImageWrapThresold = cv2.adaptiveThreshold(ImageWrapGauss, 255, 1,1, 7, 2)
+
+        '''        
+        # Otsu thresolding
+        ret, ImageWrapThresold = cv2.threshold(ImageWrapGauss,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)        
+        '''
+        ImageWrapBitwise = cv2.bitwise_not(ImageWrapThresold)
+        '''
+        # Remove salt and pepper noise
+        ImageWrap = cv2.medianBlur(ImageWrapBitwise,5)
+        '''        
+
+    cv2.imshow('Keypoints detected',image_copy_2)
+    cv2.imshow('Contour Drawn',image_copy_3)
+    cv2.imshow('Get Perspective image',imageWrapColored)
+    cv2.imshow("Otsu thresolded image",ImageWrapBitwise)
+
+    if cv2.waitKey(0) & 0xFF==27:
+        break
+
+# Close all window
+cv2.destroyAllWindows()
+
+print(biggest,maxArea, biggest.size)
+print(biggest_in_order)
+print(biggest_in_order.shape)
+print(biggest_in_order[0][0][0],biggest_in_order[0][0][1])
+print(biggest_in_order[1][0][0],biggest_in_order[1][0][1])
+print(biggest_in_order[2][0][0],biggest_in_order[2][0][1])
+print(biggest_in_order[3][0][0],biggest_in_order[3][0][1])
+print(imageWrapColored.shape)
+