Inspired by the layout analysis from the phone screen, we built a screenshot detection tool based on the detection of horizontal edges.
Both accuracy and recall of this method are approximate 95% in test data (group photo album).
The detection steps are as follows:
The original image.
Load and convert to gray image.
Detect horizontal edge
Count the number of horizontal edges to decide if it's a screenshot or not. In general, we choose the number of edges greater than or equal to 1.
In the current version, the tool only works with python3.
Build and install from source:
$ git clone https://github.com/dukn/screenshot_detector.git
$ cd screenshot_detector
$ pip install . # pip3
To use as a python library:
from screenshot_detector import detect
The tool reads image from the disk path or link and detects horizontal edges.
The parameters of detect function:
-
list_image: list of tuple (index, image_path, type) - index is index of the image. - image_path is path or url to the image we need to process. - type is type of image_path. The type must be "link" or "path". Example:
list_image = [(1, 'path_to_image.jpg', 'path'), (2, 'https://website.com/url_to_image.jpg', 'link')]. -
nprocess: int, default = 40 Number of the processors.
-
output: str, default = "output.tsv" Path to output file, defaul is output.tsv. The output file with 3 columns and seperate by a tab ('\t'). The first column is index. The second column is image_path. The third column is number of horizontal line in this image.
Detect:
from screenshot_detector import detect
input = [(1, 'images/test/img1.jpg', 'path'), (2, 'images/test/img2.png', 'path')]
detect(list_image=input, nprocess=4, output="output_path.tsv")