simple-chinese-ocr-with-opencv

Opencv is an important library for digital image processing. This repository implemented a simple chinese ocr: recognizing book name, with opencv.

Main pipelines

First of all, we do the perspective transform to get the bird view image. And get the ROI. def perspective_image(img, src, dst= np.float32([ [0,0], [600,0], [0,1200], [600,1200]]),): img, Minv = warp(img, src, dst) return img, Minv

	def warp(img, src, dst):
	M = cv2.getPerspectiveTransform(src, dst)
	Minv = cv2.getPerspectiveTransform(dst, src)
	warped = cv2.warpPerspective(img, M, img.shape[0:2][::-1], flags=cv2.INTER_LINEAR)
	return warped, Minv

Then we need to transfer BGR to HSV in order to get the black color regions. hsv = cv2.cvtColor(perspective, cv2.COLOR_BGR2HSV) lower_black = np.array([0,0,0]) upper_black = np.array([200,255,54]) mask = cv2.inRange(hsv, lower_black, upper_black) res = cv2.bitwise_and(perspective,perspective, mask= mask)

Next, we do erode and dilate to the image. def preprocessimage(img,n): kernel=np.uint8(np.zeros((n,n))) for x in range(n): kernel[x,2]=1 kernel[2,x]=1 eroded=cv2.erode(img,kernel) dilated = cv2.dilate(img, kernel) result = cv2.absdiff(dilated, eroded) return result

Right now, we can find contours in this image and the largest contour is the name of book.

Change it to the binary image and then calculate the pixel value in the vertical axis you can do segmentation.

Finally, train a classifier we can finish this simple ocr.