This code aims at rectifying corrupted videos particularly where frames are shuffled and contains outlier images which are not part of the main video.
Let us understand in detail about the project
The goal is to have an installable Python package that could be integrated into the company’s processing pipelines or just be used by anyone in the team on their own computer to correct any corrupted videos they might have.
The images can vary enormously. Thus it is possible that the frames which are not part of the video (outliers) are very close to an image at the start of the sequence but very far from an image at the end of the sequence. However, the distance between each successive image should be close (this is a hypothesis). Under these conditions, it is likely that there is a dense area which is the area of the images of the sequence, and isolated points around, which are the outliers. Therefore, the technique used here is to detect outliers with density-based clustering. Then we implemented SSIM to quantise the difference between the two images and create a matrix . We sort the images based on the result of the values.
Density-based spatial clustering of applications with noise (DBSCAN) is a data clustering algorithm. It is a density-based clustering non-parametric algorithm: given a set of points in some space, it groups together points that are closely packed together (points with many nearby neighbors), marking as outliers points that lie alone in low-density regions (whose nearest neighbors are too far away). DBSCAN is one of the most common clustering algorithms and also most cited in scientific literature.
The Structural Similarity Index (SSIM) is a much newer equation developed in 2004 by Wang et al. SSIM Index quality assessment index is based on the computation of three factors; luminance (l), contrast (c) and structure (s). The SSIM values range between 0 and 1 where 1 means a perfect match between the original image and the copy.
After implementing SSIM the code builds the m_d_ssim matrix. Each point (i, j) of the m_d_ssim matrix contains the
distance between the images i and j of the good_frames list. The distance used is the ssim.
Each row of the previous matrix is sorted according to the row axis. A new matrix m_d_s_idx is created and contains not
the sorted values, but the indices. For example, if the first row of the m_d_s_idx matrix starts with [1, 60, 9 ...],
it means that the image closest to image 1 according to the ssim metric is indeed image 1 , the second closest is
picture 60, the third is picture 9.
The sort_indx function will retrieve the closest frame for each frame and so on. It returns a list which contains the
order of the frames. It takes as a parameter the frame to use for the start of the sequence.
The distance ssim is equal to 1 for identical images and -1 for distant images. This is why the sort is performed
according to -m_d_ssim rather than m_d_ssim.
Then the difference in distance between all the frames of the reordered video, we keep the beginning which gives a
sequence where the sum of the differences in distances between the frames is the smallest. In other words, it is the
sequence where there are the least “jumps” of distances between the frames. The smoothness function is responsible
for calculating this index.
We can use the code by either installing the package on the local or install the docker image
- Clone the code on your local
git clone https://github.com/princys-laboratory/videoprocess.git
cd videoprocess
- Create and activate the conda environment
conda create --name videoprocess python=3.7
conda activate videoprocess
- Install the package
pip install ./Installable/videoprocess-0.0.1.tar.gz
Please note that the docker file was tested and implemented on Docker version 19.03.8,
The Dockerfile is supplied to build images. Make sure docker is installed on your local.
Type the following on your terminal to confirm docker is intsalled.
docker -v
You should be able to see the version of the docker in response similar to this Docker version 19.03.8, build afacb8b
- Clone the code on your local
- Navigate to the folder
git clone https://github.com/princys-laboratory/videoprocess.git
cd videoprocess
-
Update your video name on Docker image by replacing
shuffled_19.mp4with your file name (consider the local file location while updating the data) in the linesCOPY ./shuffled_19.mp4 .andCMD ["./main.py", "--input_file", "shuffled_19.mp4", "--output_file", "out.mp4"]inDockerfile. -
You can change
out.mp4with your desired output file name. -
Build your image . Below
video-processsignifies the Repository name of the image and0.1signifies the Tag
docker build --file Dockerfile -t video-process:0.1 .
Once the image is successfully build you will see the message Successfully tagged video-process:0.1 on your terminal.
On successful installation, you should be able to see the Repository name and Tag when you type the following
docker images
We can run the application either by command line or docker image
Sample Command line arguments:
usage:python3 main.py --input_file shuffled_19.mp4 --output_file out.mp4
argument :
--input_file Path to the input video file to be recitfied
--output_fie Path to the output file where the recified image is to be stored
- To run your docker image:
docker run -t video-process:0.1
Where video-process is the repository name and 0.1 is the Tag
- Docker Compose is available to run the application when multiple containers are available
docker-compose up -d