pil-numpy-homography-sample

A sample code for homography image transformation.

before	after

Environment

Python 2.7

Dependencies

• Numpy
• PIL
  This code doesn't depend on OpenCV but only Numpy and PIL, so it can be used on Google App Engine :)

Set up commands

pip install numpy==1.6.1
pip install Pillow==4.3.0

This code might be compatible with upper versions of Numpy/Pillow, but it's not tested.

Performances

base image size	color scale	execution time (avg - 10times)
768x1024	RGB	14.5ms
768x1024	Gray	6.28ms
2448x3264	RGB	150ms
2448x3264	Gray	60.2ms

(on Macbook Air)

Sample code to use homography.py

#!/usr/bin/python
# -*- coding: utf-8 -*-

import path.to.homography as homography
import base64
from io import BytesIO

def some_method_to_process_uploaded_img(base64_img, corner_list, dest_list):
    base_img = Image.open(BytesIO(base64.b64decode(base64_img)))
    # execute homography transformation
    transformed_img = homography.transform(
        img=base_img,
        crop_corner_list=corner_list,
        destination_corner_list=dest_list
    )

    # or, extend to the size of base_img == (destination_corner_list=None)
    # transformed_img = homography.transform(
    #     img=base_img,
    #     crop_corner_list=corner_list
    # )

    # If base64 encoded data is needed:
    in_memory_file = BytesIO()
    transformed_img.save(in_memory_file, format="PNG")
    in_memory_file.seek(0)
    img_bytes = in_memory_file.read()
    transformed_base64_image = base64.b64encode(img_bytes).decode('ascii')

Sample test commands

# execute sample image transformation with ./img/sample.png
test.sh

# transform with gray scale option(-L)
python ./main.py -i ./img/sample.png -c "78, 114, 18, 258, 390, 230, 342, 18" -L

# transform to an intended shape
python ./main.py -i ./img/sample.png -c "78, 114, 18, 258, 390, 230, 342, 18" -d "0, 50, 30, 250, 260, 140, 250, 10"

# save the result as a file
python ./main.py -i ./img/sample.png -c "78, 114, 18, 258, 390, 230, 342, 18" -o ./path/to/destination

Ref

PIL/pillowとNumpyで射影変換(ホモグラフィ変換)をしてみた / GAE環境でも実行可能