giacomo-ascari / dichotomic-compression

Image compression algorithm, developed for research and educational purposes

Dichotomic Compression

Image compression algorithm, developed for research and educational purposes.

Overview

The dichotomic compression aims to minimize the size of a raster image by dinamically reducing the resolution. It is done by aggregating regions of homogeneous pixels into a single sector. The similarity of the region is regulated by the threshold value (in a range between 0-255).

What emerged from early experiments, is that we can see that the algorithm is best when compressing drawings, sketches, signs or car plates. It prioritizes high contrast details, such as lines, dots and texts.

These examples show the compressed file size compared to the original. Custom thresholds have been used.

Workflow

This workflow doesn't have to implemented strictly: depending on the implementation, some steps can be (or must be) integrated with one another.

Sectorization

It's the core of the whole process. It's implemented in a recursive function that operates on the pixel matrix. The sectors, when found, are added to a proper list. The "delta" is a 3-dimensional array, representing the absolute difference between channels of each pixel inside a sector. This mean that "details" (anything that will be preserved as a near original resolution) are usually high contrast regions.

.DCI format

DCI (dichotmic compression image) is the format chosen to store the compressed image.

• Metadata (10 bytes)
  • version (1 byte): specifies the version used
  • threshold (1 byte): specifies the chosen threshold
  • width (4 bytes): width of the original image
  • height (4 bytes): height of the original image
• Sectors (variable length depending on sectors count and version)
  • size/position: any data that can give information about the position of the sector
  • color channels (RGB)

.DCI versions

• Version 1
  • Size/position: number of sectorizations done before getting to the sector itself (1 byte)
  • RGB stored as 3 1-byte channels
  • Final sector "weight": 4 bytes (32 bits)
• Version 2
  • Applies the sectorization on the three separate color channels as if they were 3 different images.
  • Size/position: number of sectorizations done before getting to the sector itself (1 byte)
  • Color stored as 8-bit channel.
  • Final sector "weight": 2 bytes (16 bits)
  • Approximately 3 times more sectors

Usage

The repository only includes the source and the compilation script. In order to use it, you must first compile the source code with ./compile.sh.

Website

In order to improve testing and usability of the dichotomic compression, a custom web application can been deployed on a self-hosted server. You can find the code in the ./web folder. The website allows the user to:

• compress any image, select version and threshold, take a look at the program output and download the .dci file
• decompress any .dci file, take a look at the program output and download the decompressed image

Command line arguments

Required

• -c or -d: specifies compression or decompression
• -f FILENAME: specifies which file will be processed

Not required

• -t THRESHOLD: specifies compression threshold
• -v VERSION: specifies used version (default 2)
• -V: verbose

Examples

• ./dc_core -c -f ../examples/lines.png -t 30 -V Compression (verbose, with custom threshold)
• ./dc_core -d -f ../examples/lines.png.dci -V Decompression (verbose)