Improving small images for data extraction-CodePudding

The

I tried below code with multiple options like thresholding, blur, enchance, however not able to improve.

img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) print(pytesseract.image_to_string(img)) img_medianBlur = cv2.blur(img, (3 , 3)) print(pytesseract.image_to_string(img_medianBlur)) blur = cv2.GaussianBlur(img,(5,5),0) ret3,th3 = cv2.threshold(img,150,255,cv2.THRESH_BINARY cv2.THRESH_OTSU) print(pytesseract.image_to_string(th3))

CodePudding user response：

To improve the images for tesseract, you can try using some of the following techniques:

Increase the contrast of the image by using histogram equalization or stretching the intensity range of the image.

img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) equ = cv2.equalizeHist(img) print(pytesseract.image_to_string(equ))

Remove any noise or artifacts from the image using a median or Gaussian blur filter

img_medianBlur = cv2.medianBlur(img, 3) print(pytesseract.image_to_string(img_medianBlur))

Use adaptive thresholding to segment the text from the background.

th = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2) print(pytesseract.image_to_string(th))

Sharpen the image to make the text more distinct and easier to read.

sharpened = cv2.filter2D(img, -1, np.array([[-1, -1, -1], [-1, 9, -1], [-1, -1, -1]])) print(pytesseract.image_to_string(sharpened))

Use morphological operations such as dilation or erosion to fill in gaps in the text and remove small objects that may be interfering with the text recognition.

kernel = np.ones((3, 3), np.uint8) dilated = cv2.dilate(img, kernel, iterations=1) eroded = cv2.erode(dilated, kernel, iterations=1) print(pytesseract.image_to_string(eroded))

It may also be useful to try combining these techniques and experimenting with different parameter values to find the optimal configuration for your specific image.

CodePudding user response：

It looks like the first two images are the result of motion blur.

If denoising filters, such as Gaussian, don't work, I suggest research on two topics, motion deblur and deconvolution.

You might want to check out this OpenCV tutorial on deblurring filter and motion deblur filter.

Here explains mathematical aspect of the algorithm.

I'll come back to improve my answer if time permits.