Denoising the document image<\/h2>\n\n\n\n
Removing noise from scanned paper is a necessary task before applying machine learning algorithms. There exist several supervised\/unsupervised denoising methods. In this post, we use non-local means method to eliminate noises from image. It simply replaces the color of a pixel with an average of the colors of similar pixels. But the most similar pixels to a given pixel have no reason to be close at all (paper link<\/span><\/a>). OpenCV fastNlMeansDenoising()<\/strong><\/span> function removes noises using non-local means denoising algorithm with some computational optimizations.<\/p>\n\n\n\n In the fastNlMeanDenoising <\/strong>function we need to specify following parameters :<\/p>\n\n\n\n templateWindowSize <\/strong>: It is size of the template patch which is used to compute weights. Should be odd. Recommended value for better denoising performance is 7<\/p>\n\n\n\n searchWindowSize <\/strong>: It is size of the window that is used to compute weighted average for h<\/strong> : This parameter regulates filter strength. A big h value perfectly removes noise but also has side effects on image details, whereas smaller h value preserves details but also preserves some noise.<\/p>\n\n\n\n Output of above code on a noisy image :<\/p>\n\n\n\n Binarization is a crucial task that should be done before feature extraction, it converts an image into black and white image in which white pixels are represented by 255 and black pixels by 0. We do the binarization using a threshold. If in the given image a pixel value exceeds the threshold, we set it as a white pixel with value of 255, otherwise we set the pixel as black with value of zero. If we choose a good threshold, it can also help in noise reduction. So, choosing appropriate threshold<\/em><\/strong> is the most important part of binarizatio. OTSU\u2019s method calculates a threshold for the whole image considering the several characteristics of the entire image. When we use OTSU method, we don’t need to determine threshold explicitly, so the threshold function will ignore the 0 and 255 in the argument.<\/span><\/p>\n\n\n\n Ouput of above code on our scanned document :<\/p>\n\n\n\n Skewed scanned document is a common issue in feature extraction and also image classification tasks. To solve this problem by re-aligning the document image, first we need to find the deviation angle of the content against the horizontal line. Then, we can rotate the image in the opposite direction of deviation to align the document. To find the deviation angle of the content against the horizontal line, we have to extract content’s lines, we do it using Canny <\/strong>edge detection <\/strong>function along with HoughLinesP <\/strong>line detection function. When we have the widest line of the document we can find the angle between it and the horizontal line, we wrote get_angle <\/strong>function to do this task for us. Finally, we rotate the image content to remove deviation, it is done by rotate_image <\/strong>function in the code below :<\/p>\n\n\n\n Output of the align_image <\/strong>function :<\/p>\n\n\n\nimport cv2\nimport numpy as np\n\n#read the noisy image\nnoisyImage= cv2.imread(\"noisy_image.jpg\",cv2.IMREAD_GRAYSCALE)\n#applying fast non-local means denoisong filter\ndenoisedImage= cv2.fastNlMeansDenoising(noisyImage, None, h = 44, templateWindowSize = 7, searchWindowSize = 21)\n#join noisy and denoised images\nnosiy_denoised = np.concatenate((noisyImage, denoisedImage), axis=1)\n#save joined images in file\ncv2.imwrite(\"nosiy_denoised.jpg\",nosiy_denoised)<\/code><\/pre>\n\n\n\n
given pixel. Should be odd. The greater value leads to longer denoising time. Recommended value for best performance is 21.<\/p>\n\n\n\n![\"image](\"https:\/\/m-shaeri.ir\/blog\/wp-content\/uploads\/2022\/06\/nosiy_denoised-1024x660.jpg\")
<\/a>Scanned Document Binarization <\/span><\/h2>\n\n\n\n
import cv2\nimport numpy as np\n\n#read image from file\nimg = cv2.imread(\"test.jpg\",cv2.IMREAD_GRAYSCALE)\n\n# binarization with OTSU threshold finder. 0 and 255 are ignored\nthreshValue, binaryImage = cv2.threshold(img, 0, 255, cv2.THRESH_OTSU)\n\nnormal_binary = np.concatenate((img, binaryImage), axis=1)\n\ncv2.imwrite(\"normal_binary.jpg\",normal_binary)<\/code><\/pre>\n\n\n\n![\"scanned](\"https:\/\/m-shaeri.ir\/blog\/wp-content\/uploads\/2022\/06\/normal_binary-1024x679.jpg\")
<\/a>Aligning Scanned document<\/h2>\n\n\n\n
\nimport cv2\nimport numpy as np\nimport math\n\ndef get_angle(x1, y1, x2, y2) -> float:\n \"\"\"Get the angle of this line with the horizontal axis.\"\"\"\n deltaX = x2 - x1\n deltaY = y2 - y1\n angleInDegrees = np.arctan2(deltaY , deltaX) * 180 \/ math.pi\n \n return angleInDegrees\n\ndef rotate_image(image, angle):\n image_center = tuple(np.array(image.shape[1::-1]) \/ 2)\n rot_mat = cv2.getRotationMatrix2D(image_center, angle, 1.0)\n result = cv2.warpAffine(image, rot_mat, image.shape[1::-1], flags=cv2.INTER_LINEAR, borderValue=(255,255,255) )\n return result\n\n\ndef align_image(img):\n\n # Median blurring to get rid of the noise; invert image\n #img = cv2.medianBlur(img, 3) # use this if the document image is noisy\n\n edges = cv2.Canny(img, 80, 120)\n\n # Detect and draw lines\n lines = cv2.HoughLinesP(edges, 1, np.pi\/180, 10, minLineLength=20, maxLineGap=10)\n # sort lines from widest to shortest\n lines = sorted(lines,key = (lambda l: abs(l[0][0]-l[0][2])) , reverse = True)\n\n # if there exist any line, compare it by horizontal line\n # and rotate the image if the angle difference is more than 0.25\n for line in lines:\n for x1, y1, x2, y2 in line:\n if (abs(x2-x1) \/ edges.shape[1])>0.25 :\n angle = get_angle(x1, y1, x2, y2)\n if abs(angle) > 1.0 :\n img = rotate_image(img,angle)\n print(\"rotated\")\n #exit after comparing widest line\n break\n\n return img\n\n#read the unligned image\nunalignedImage= cv2.imread(\"unaligned_image.jpg\",cv2.IMREAD_GRAYSCALE)\n\n#apply re-aligning function\naligned_image = align_image(unalignedImage)\n\nunaligned_aligned = np.concatenate((unalignedImage, aligned_image), axis=1)\n#save joined images in file\ncv2.imwrite(\"unaligned_aligned.jpg\",unaligned_aligned)\n<\/code><\/pre>\n\n\n\n![\"image](\"https:\/\/m-shaeri.ir\/blog\/wp-content\/uploads\/2022\/06\/unaligned_aligned-1024x647.jpg\")
<\/a>