How do I get bounding boxes from OCR results?

To get bounding boxes from OCR results, you usually need to look at the OCR engine’s output structure: most OCR tools return either word-level, line-level, or character-level boxes along with text.

General idea

A bounding box is typically represented as:

• x, y = top-left corner
• width, height = box size

or as:

• left, top, right, bottom

If you’re using common OCR tools

1) Tesseract

Tesseract can return bounding boxes for each recognized word.

Python example

import pytesseract
from PIL import Image

img = Image.open("image.png")
data = pytesseract.image_to_data(img, output_type=pytesseract.Output.DICT)

n = len(data["text"])
for i in range(n):
    if int(data["conf"][i]) > 0:
        x = data["left"][i]
        y = data["top"][i]
        w = data["width"][i]
        h = data["height"][i]
        text = data["text"][i]
        print(text, (x, y, w, h))

Output fields

• left, top, width, height
• text
• conf = confidence score

---

2) Google Cloud Vision OCR

Returns bounding polygons for each detected word/paragraph.

for page in response.full_text_annotation.pages:
    for block in page.blocks:
        for para in block.paragraphs:
            for word in para.words:
                vertices = word.bounding_box.vertices
                print([ (v.x, v.y) for v in vertices ])

---

3) AWS Textract

Textract returns a BoundingBox for each detected block.

for block in response["Blocks"]:
    if block["BlockType"] == "WORD":
        box = block["Geometry"]["BoundingBox"]
        print(block["Text"], box)

BoundingBox usually contains:

• Left
• Top
• Width
• Height

---

4) PaddleOCR

PaddleOCR returns both text and quadrilateral boxes:

from paddleocr import PaddleOCR

ocr = PaddleOCR()
result = ocr.ocr("image.png")

for line in result[0]:
    box = line[0]
    text = line[1][0]
    print(text, box)

Box is usually 4 points:

[[x1, y1], [x2, y2], [x3, y3], [x4, y4]]

---

If you already have OCR text output

Look for any of these in the response:

• bbox
• bounding_box
• box
• left/top/width/height
• polygon points
• vertices

If the OCR output does not include boxes, then that OCR mode likely isn’t configured to return them.

Important note

Bounding boxes may be for:

• words
• lines
• characters
• regions/blocks

So make sure you know the level you’re extracting.

If you want, I can show you how to get bounding boxes from OCR results in a specific library like Tesseract, PaddleOCR, EasyOCR, Textract, or OpenCV.