Adding image-splitter that uses Canny-Edge detection.

spines/README.md

@@ -0,0 +1,18 @@
+## Splitting images into 2 pages using Canny edge detector and pseudo-image-metadata
+
+### How it Works
+
+* A non machine-learning approach that does not require ground truth
+* First, it converts the given image into black and white
+* Then, it considers only 10% of the image, around the centre (on both sides, thus 20% of the image)
+* On this copped image, Canny-edge detector is run
+* Then, the sum of pixels for all possible vertical lines in this range is tried, and the one corresponding to the most black region (which is present in between any two pages) is used.
+* Emperical analysis shows that this technique performs quite well for the given images
+
+
+### Running it
+
+* Simply run `python splitImagesIntoPages.py <images_Dir> <output_Images_Dir>`. For each image in the input directory, two images (with suffix '_left' and '_right') will be dumped into the output directory.
+* For example, create a directory here named 'OutputImages', and then run `python splitImagesIntoPages.py images/freecen/1841/ OutputImages/`
+
+

spines/splitImagesIntoPages.py

@@ -0,0 +1,57 @@
+import numpy as np 
+from PIL import Image
+import os
+import cv2
+from tqdm import tqdm
+
+# Detect Cany-edges using the given sigma, for the provided input image
+def auto_canny(image, sigma=0.33):
+	v = np.median(image)
+	lower = int(max(0, (1.0 - sigma) * v))
+	upper = int(min(255, (1.0 + sigma) * v))
+	edged = cv2.Canny(image, lower, upper)
+	return edged
+
+# Extract central area fo the given image (as the centre of two pages is likely to be in this region)
+def getRelevantArea(image, sideRatio=0.1):
+	side = image.shape[1]
+	leftSide = int((0.5 - sideRatio) * side)
+	rightSide = int((0.5 + sideRatio) * side)
+	return image[:, leftSide:rightSide], int((0.5 - sideRatio)*side)
+
+# Calculate the optimal straight line that splits the image into 2 pages
+def getOptimalStraightLine(image):
+	scores = np.sum(image, axis=0)
+	return np.argmin(scores)
+
+# Split images in a given directory into 2 pages per image, and dump them in the output directory
+def splitImagesIntoPages(directory, outputDirectory):
+	for file in tqdm(os.listdir(directory)):
+		imagePath = os.path.join(directory, file)
+		# Extract filename for later use
+		fileName = file.split('.')[0]
+		# Open image in original mode
+		rawImage = Image.open(imagePath)
+		# Convert image to B/W
+		readImage = np.asarray(rawImage.convert('L'))
+		rawImage = np.asarray(rawImage)
+		# Extract central area from B/W image
+		trimmedImage, leftSide = getRelevantArea(readImage)
+		# Get Canny-edge image for the above
+		edgeImage = auto_canny(trimmedImage, 0.5)
+		# Split given edge-image into 2 vertically
+		splitLine = getOptimalStraightLine(edgeImage)
+		# Split image into 2 pages using above calculations
+		leftImage = rawImage[:, :splitLine + leftSide]
+		rightImage = rawImage[:, splitLine + leftSide:]
+		# Save these pages as two different images
+		leftPage = Image.fromarray(leftImage)
+		leftPage.save(os.path.join(outputDirectory, fileName + '_left.jpeg'))
+		rightPage = Image.fromarray(rightImage)
+		rightPage.save(os.path.join(outputDirectory, fileName + '_right.jpeg'))
+
+
+if __name__ == "__main__":
+	import sys
+	splitImagesIntoPages(sys.argv[1], sys.argv[2])
+