Image Sequence Abstraction for Python

Video data can appear in different forms: bunch of files, a video file, a network stream etc. We provide an abstraction for ordered image sources that serve as an isolation from a specific source type.

The interface is simple:

import imagesource

images = imagesource.VideoSource('tests/data/MOV02522.MPG')
img100 = images.get_image(100)
img101 = images.get_next_image()
images.rewind()
img000 = images.get_next_image()
images.write_images('out/%03d.png', 100)

images2 = imagesource.FilesSource('tests/data/frames/%03d.jpg')
# same interface as above ...

The basic sources are VideoSource and FilesSource for video files and sequences of image files respectively. The TimedVideoSource extracts frame timestamps from video files. The SynchronizedSource translates frame numbers using a table. This can be used for creating a synchronized set of sources.

For more examples see tests/test.py

Installation

Install OpenCV 3.x with Python bindings and Numpy using a system package manager.

$ pip install imagesource

The TimedVideoSource requires ffprobe command from the ffmpeg suite.

Testing

$ pip install nose
$ nosetests

Writing Extensions

It is simple to write transparent image source wrappers that post-process image data from an underlying image source (e.g. background subtraction, radial distortion removal, ...).

An example background subtracted image source:

class BackgroundSubtractedSource(imagesource.ImageSource):
    def __init__(self, source):
        self.source = source
        self.bgs = cv2.createBackgroundSubtractorMOG2(...)

        def get_image(self, frame):
            img = self.source.get_image(frame)
            return self.bgs.apply(img)

        def get_next_image(self):
            img = self.source.get_next_image()
            return self.bgs.apply(img)

        def rewind(self):
            self.source.rewind()