imagorvideo

imagorvideo is a new initiative that brings video thumbnail capability through ffmpeg, built on the foundations of imagor - a fast, Docker-ready image processing server written in Go with libvips.

imagorvideo uses ffmpeg C bindings that extracts image thumbnail from video, by attempting to select the best frame. It then forwards to libvips to perform all the image cropping, resizing and filters supported by imagor.

imagorvideo integrates reader and seeker interfaces of ffmpeg C bindings with imagor loader, storage and result storage, which supports HTTP(s), S3, Google Cloud Storage out of box. It uses read stream for mkv and webm video types. For other video types that requires seeking from a non seek-able source such as HTTP, it simulates seek using memory or temp file as buffer.

This also aims to be a reference project demonstrating imagor extension.

Quick Start

docker run -p 8000:8000 shumc/imagorvideo -imagor-unsafe

Original:

http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4

Result:

http://localhost:8000/unsafe/300x0/7x7/filters:label(imagorvideo,-10,-7,20,yellow):fill(yellow)/http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4

imagorvideo works by streaming out a limited number of frame data, looping through and calculating the histogram of each frame. It then choose the best frame for imaging, based on root-mean-square error (RMSE). This allow skipping the black frames that usually occur at the beginning of videos.

imagorvideo then converts the selected frame to RGB image data, forwards to the imagor libvips processor, which has always been best at image processing with tons of features. Check out imagor documentations for all the image options supported.

Filters

imagorvideo supports the following filters, which can be used in conjunction with imagor filters:

• frame(n) specifying the frame index n for imaging. This allows skipping the default automatic selection, which involves computing root-mean-square error (RMSE) of histogram frames data.
• max_frames(n) restrict the maximum number of frames allocated for image selection. The smaller the number, the faster the processing time.

Example:

http://localhost:8000/unsafe/300x0/7x7/filters:max_frames(70):fill(yellow)/http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4

Metadata

imagorvideo provides metadata endpoint that extracts video metadata, including dimension, duration and FPS data. The endpoint only processes the header without extracting the frame data.

To use the metadata endpoint, add /meta right after the URL signature hash before the image operations:

http://localhost:8000/unsafe/meta/https://test-videos.co.uk/vids/bigbuckbunny/mp4/h264/1080/Big_Buck_Bunny_1080_10s_30MB.mp4

{
  "format": "mp4",
  "content_type": "video/mp4",
  "orientation": 1,
  "duration": 10000,
  "width": 1920,
  "height": 1080,
  "title": "Big Buck Bunny, Sunflower version",
  "artist": "Blender Foundation 2008, Janus Bager Kristensen 2013",
  "fps": 30,
  "has_video": true,
  "has_audio": false
}

Configuration

Configuration options specific to imagorvideo. Please see imagor configuration for all existing options available.

  -ffmpeg-fallback-image string
        FFmpeg fallback image on processing error. Supports image path enabled by loaders or storages