[ python ] OpenCV

Installation

$ apt-get install opencv-python

Non-free version

$ apt-get install opencv-contrib-python

1.Threshold

ref : http://docs.opencv.org/3.2.0/d7/d4d/tutorial_py_thresholding.html

.threshold(THRESH_BINARY(around 192)+THRESH_OTSU(around 177))

.adaptiveThreshold(ADAPTIVE_THRESH_MEAN_C|ADAPTIVE_THRESH_GAUSSIAN_C)

2.Morphology(Erosion/Dilation, Opening/Closing)

.morphologyEx(cv2.MORPH_OPEN) : to remove noise

kernel = np.ones((5,5), np.uint8)
erosion = cv2.erode(img, kernel,iterations = 1)

.dilation

3.Contour ( to find the border )

TIPs : We can rotate the picture 90 degrees, then execute findContour again.

ref : http://docs.opencv.org/3.2.0/d4/d73/tutorial_py_contours_begin.html

.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)

imgray = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)

.drawContours()

4. Morphological transform ( to find the border )

ref : http://docs.opencv.org/3.2.0/d9/d61/tutorial_py_morphological_ops.html

.getStructuringElement(cv2.MORPH_RECT, long array)

get max ( dilate ) : return 111111… if there is a matched long array

get min ( erosion ) : return 000000… if there is a matched long array

5. Template matching ( to find a specific charactor, i.e. bullet, to detect whether it is a first page )

ref : http://docs.opencv.org/3.2.0/d4/dc6/tutorial_py_template_matching.html

result = .matchTemplate(img, template, TM_CCORR_NORMED)

_, max_val, _, max_loc = cv2.minMaxLoc(result)

// now we got max_loc

6. reduce ( to extract text line; example : we can detect the block of text line and seperate it with the blank space )

.reduce(src, dst, 1, REDUCE_AVG) >= THRESHOLD

7. OCR

tesseract / tensor flow
OCRopus / ocropy ( https://github.com/tmbdev/ocropy : python document analysis tool )
scantailor ( http://scantailor.org/ : GUI to clean scan docs )

8. Vision Course

CE316 / CE866 ( http://orb.essex.ac.uk/ce/ce316/ ) University of Essex, UK

dsin