How to convert from float 32 to 8 bit without losing information?

Question

I tried to find contours in my image with cv2.findContours. So as it uses CV_8UC1 images I tried to convert my array with dtype=np.uint8, before it was 32 bit. But there I am loosing information. Is there any other way?

The second porblem is the bounding box. The informations are saved in rect but it is not draw in the picture. Does anyone know why?

Here is my picture/array in 32 bit:

And this is my picture when I added dtype=np.uint8 :

img_hr = np.array(b[1],dtype=np.uint8)

img_hr=img_hr*255
plt.imshow(img_hr)

hierachy, img_threshold = cv2.threshold(img_hr, 100, 255, cv2.THRESH_BINARY)

contours,_ = cv2.findContours(img_threshold, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

cv2.drawContours(img_threshold, contours, -1, (0, 255, 0), 2, cv2.LINE_AA)

for cnt in contours:         
      rect = cv2.minAreaRect(cnt)
      box = cv2.boxPoints(rect)
      box = np.int0(box)
      cv2.drawContours(img_threshold,[box],0,(0,0,255),2)
      cv2.circle(img_threshold,(int(rect[0][0]),int(rect[0][1])),5,(255,0,0),-1)

plt.imshow(img_threshold)

I hope you understand my problem. If not please ask. I appreciate your help. Thanks

I don't understand your question very well, but there's nothing stopping you finding contours using an 8-bit image and applying the contours back to your original 32-bit data when you have them. — Mark Setchell, Jul 07 '21 at 13:26
My problem is, that I lose contours as shown in the pictures when I convert from 32 to 8 bit. I dont need to convert back to 32 bit. — lkrss, Jul 07 '21 at 13:30
When your data was 32-bit, what were the brightest and darkest pixels? And what were they after you converted to 8-bit? — Mark Setchell, Jul 07 '21 at 13:39
32 bit: min--> 6.723715e-20; max --> 255; 8 bit: min -->0; max -->255 — lkrss, Jul 07 '21 at 14:07

score 0 · Accepted Answer · answered Jul 07 '21 at 15:04

I found a solution myself:

First I used the 32-bit image to find the contours with cv2.threshold in the 32-bit image. After this I convert the 32-bit array to 8-bit np.array(img_threshold_32bit,dtype=np.uint8) without losing any contours. So now the input for cv2.drawContours is an 8-bit array.

But I still have the problem that the bounding-box is not draw in the plot. Any ideas?

Here is the code:

img_hr = np.array(b[1])

hierachy, img_threshold_32bit = cv2.threshold(img_hr, 100, 255, cv2.THRESH_BINARY)
img_8bit = np.array(img_threshold_32bit,dtype=np.uint8)
contours,_ = cv2.findContours(img_8bit, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cv2.drawContours(img_8bit, contours, -1, (0, 255, 0), 2, cv2.LINE_AA)

for cnt in contours:
      rect = cv2.minAreaRect(cnt)
      box = cv2.boxPoints(rect)
      box = np.int0(box)
      cv2.drawContours(img_8bit,[box],0,(0,0,255),2)
      cv2.circle(img_8bit,(int(rect[0][0]),int(rect[0][1])),5,(255,0,0),-1)

plt.imshow(img_8bit)

score 0 · Answer 2 · answered Jul 08 '21 at 04:12

0

because you draw yourself drawContours only draws contours outlines or filled contours. you must be find a rectangle coords for each contour by using boundingRec or minAreaRect then draw it by using drawing functions of opencv

check this sample code https://github.com/birolkuyumcu/opencvbook_python/blob/master/Ders6/Ders6.py#L114

answered Jul 08 '21 at 04:12

Birol Kuyumcu

1,195
7
17

I think that is what I did. As i said, I found the countours correctly with `minAreaRect` because my variable `rect` has the right values for the bounding box. And then I use `drawContours` but the bounding box is not shown in the image – lkrss Jul 08 '21 at 09:14
drawContours ONLY draws contours outlines or filled contours. read code to draw a rectangle you must be use cv2.rectangle function – Birol Kuyumcu Jul 11 '21 at 04:15

How to convert from float 32 to 8 bit without losing information?

2 Answers2

Linked