Look for the outline of the numbers in the picture and crop the numbers.

This digital image from the Mnist library

The general process is: Display the original image-"grayscale (the main role is to change the number of bits of data)-" To do the anti-binary (make it into a white background)-"Look for contour-" Draw outline-"cut

Why grayscale? Originally the picture is the gray-scale ah? It turns out that there are the following reasons:

ImportCv2#Import CV LibraryImportOsimage_array= Cv2.imread ('D:/number.png')#Reading PicturesCv2.imshow ("src", Image_array)#Show original picture#print (Image_array.shape)#print (image_array.size)#print (Image_array.dtype)Image_array=cv2.cvtcolor (Image_array,cv2. Color_bgr2gray)#Convert images to grayscale images from cv_32 to cv_8uKernel= Cv2.getstructuringelement (cv2. Morph_rect, (3, 3))#define the size of the corrosive nucleidilated = Cv2.dilate (Image_array,kernel)#for CorrosionCv2.imshow ("eroded Image", dilated);#display of corroded imagesret, binary= Cv2.threshold (dilated, 127, 255, Cv2. THRESH_BINARY_INV)#to reverse the binary value ofImage,contours,hierarchy= cv2.findcontours (binary, Cv2. Retr_tree, Cv2. Chain_approx_simple)#looking for outlinesCv2.imshow ('Imageshow', image)#* * Display return value image, in fact, with the input parameters of the Thresh original no difference * *#Print (np.size (contours)) ##Print (contours[0]) # Prints the coordinates of all points of the first outline, changing the 0 here to 0--(total contour number-1) to print the coordinates of all points of the corresponding contour#print (hierarchy) #** prints out the relationship between the corresponding outlines * * forIinchRange (0,len (contours)):#Draw a contourX, Y, W, h =Cv2.boundingrect (Contours[i]) cv2.rectangle (image, (x, y)+W,Y+H), (253,255,0), 5) Cv2.imshow ("finally", image)#Show FinalNewImage= image[y + 2:y + h-2, x + 2:x + w-2]#use Y to determine the height, and then use X to determine the width #裁剪轮廓Nrootdir = ("e:/cut_image/")if  notOs.path.isdir (Nrootdir): Os.makedirs (Nrootdir) cv2.imwrite (Nrootdir+ STR (i) +". jpg", NewImage)Print(i) cv2.waitkey (0)

Display effect:

　　

Look for the outline of the numbers in the picture and crop the numbers.