在给定图像中查找白色矩形区域以增加 OCR 结果

Question

Basically, I am trying to detect labels on boxes and these labels are white. Every label has own serial number and letter. I am using EasyOCR for detecting numbers and letters but sometimes EasyOCR can not detect texts accurate but If I take images from closer range, It detects very well. So, I need to crop label area with using image processing method. After that I will give it to EasyOCR to better results. I am open to any idea!

My Example Image:

Answer 1

I had solved the issue by cropping image based on image dimension with respect to the rack columns and rack rows. The code is,

import easyocr
import cv2


image = cv2.imread('2.jpg')
h, w, _ = image.shape
rack_rows = 3
rack_cols = 3
total_boxes = []
reader = easyocr.Reader(['en'])
for row in range(rack_rows):
    for col in range(rack_cols):
        # crop individual boxes based on dimension of image
        cropped_image = image[row*(int(h/3)) : ((row+1) * int(h/3)),(col*(int(w/3))) : ((col+1)* int(w/3)),:]
        # resize image based on custom size
        resize_image = cv2.resize(cropped_image,(800,800))
        # write cropped image to feed easyocr
        cv2.imwrite(f'image{row}_{col}.png',resize_image)
        # easyocr execution
        result = reader.readtext(f'image{row}_{col}.png')
        texts = [detection[1] for detection in result if detection[2] > 0.4]
        if len(texts[1])!=4:
            texts.pop(1)
        total_boxes.append(texts)

print(total_boxes)

Here I resized the image based on custom dimension (800,800) to feed easyocr after dimensional cropping. I haven't use any image preprocessing technique rather than resizer here. The output what I gained is,

[['S', '5607'], ['N', '4457'], ['U', '4017'], ['P', '4529'], ['X', '3311'], ['J', '2075'], ['W', '8075'], ['M', '9623'], ['K', '5412']]

Happy coding:)