Generic Compression Workflow

I’ve been developing compression techniques for 17 years now and have many unpublished items. One great thing about compression is it never gets old. Data never changes and Moores law does not apply to the field of digital compression. Data compression is bound by speed, memory, compression ratio very similar to the business triangle of quality, speed, and price.

My most recent soon to be published experimental work follows a generic workflow for compression which I believe can be useful for future rapid development and testing.

The below workflow allows for a base accumulator or multiple accumulators, multiple predictors using different accumulators (models). Those predictors would be put through a mixer or number of mixers. After the mixers produce a final refined prediction of the best guess to the encoder or decoder (depending on the input being a compressed or uncompressed file) and outputs the code or original information and continues along the process back to the accumulator.

What are the benefits of this type of model ? So far it allows for rapid testing of different mixing strategies, different models for predicting, and different rates of change for accumulating statistics. This workflow also allows for the creation of a neural network mixed with traditional compression techniques.

An example of this beneficial strategy is that adding a predictor has no impact or change on the accumulator, multiple predictors can be swapped out or tested at anytime. Changing the accumulator will however have an impact on the predictor using that accumulator as well as the mixer.

 

This entry was posted in Articles. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *