Visual Coding Framework

A programming environment to develop and test image and video compression algorithms.

Install and configuration

Supposing that a Python interpreter and Git are available:

  python -m venv ~/envs/VCF
  git clone git@github.com:Sistemas-Multimedia/VCF.git
  cd VCF
  source ~/envs/VCF/bin/activate
  pip install -r requirements

Usage

Image coding (example)

  cd src
  python PNG.py encode
  display /tmp/encoded.png
  python PNG.py decode
  display /tmp/decoded.png

Video coding (example)

  cd src
  python III.py encode
  ffplay /tmp/encoded_%04d.tif
  python III.py decode
  ffplay /tmp/decoded_%04d.png

Codecs organization

+---------------------+        +----+
| temporal transforms |    III |-T,N|, [IPP] (9), [IBP] (10), [MCTF] (10).
+---------------------+--+     +---++-------+
| spatial transforms  |-T| 2D-DCT* |-B,p,L,x|, 2D-DWT, [LBT] (10), no_spatial_transform.
+---------------------+--+         +--------+
|  color transforms   |-t| YCoCg*, YCrCb, color-DCT, no_color_transform.
+---------------------+--+           +--+           +------+     +----+           +--+
|     quantizers      |-a| deadzone* |-q|, LloydMax |-q,m,n|, VQ |-q,b|, color-VQ |-q|.
+---------------------+--+           +--+           ++--+--+     +----+           +--+
|  decoding filters   |-f| no_filter*, gaussian_blur |-s|, [NLM] (1), [BM3D] (3)
+---------------------+--+                           +--+
|   entropy codecs    |-c| TIFF*, PNG, Huffman, PNM, [adaptive_Huffman] (4), [arith] (4), [adaptive_arith] (5).
+---------------------+--+

...* = default option
[...] = to be implemented
(.) = points for the evaluation of the subject

Decoding Filters

Decoding filters are applied during the decompression phase to improve image quality by removing noise or artifacts.

Non-Local Means (NLM)

The NLM filter reduces noise by averaging pixel values based on the similarity of their surrounding neighborhoods, rather than just their local proximity.

Usage:

python NLM.py decode -i <input_encoded_file> -o <output_image> --h 10 --template_window_size 7 --search_window_size 21

• --h: Filter strength (higher values remove more noise but may blur details).
• --template_window_size: Size of the patch to compare.
• --search_window_size: Size of the area to search for similar patches.

Block-Matching and 3D Filtering (BM3D)

An advanced denoising algorithm that groups similar 2D image patches into 3D stacks and applies filtering in the transform domain. It supports grayscale, color, and multichannel images, as well as deblurring.

Usage:

python BM3D.py decode -i <input_encoded_file> -o <output_image> --sigma_bm3d 25.0 --profile_bm3d np

• --sigma_bm3d: Noise standard deviation (fuerza del filtro).
• --profile_bm3d: Complexity profile (lc, np, high, vn).
• --psd_bm3d: (Optional) Path to a .npy file with the noise Power Spectral Density.
• --psf_bm3d: (Optional) Path to a .npy file with the Point Spread Function for deblurring.