Optimization of an image processing library.

From LRDE

Revision as of 17:24, 6 June 2018 by Bot (talk | contribs)

Abstract

With the diversification of the acquisition devices (satellite imagery, medical imagery, HDR photography), the types of images to process has been steadily increasing (2d, 3d, etc...) thus the need for algorithms able to treat all these types. Pylene, an image processing library, aims to bring a simple and generic solution to this problem thanks to high level interfaces. However, such an abstraction often has a cost and we have to do a compromise between to a too gerenic solution which would be inefficient and an efficient solution which would be too specific. Moreover these interfaces are not compliant with the recent versions of the standard which limits his usage simplicity. We propose here a solution which would not has to sacrifice performance at the expense of genericity. This solution relies on the "range" concept, introduced by Eric Niebler, adapted to the image processing needs and to which we introduce the new concept of segmented ranges. It follows a new design which reconciles our interfaces with the new standard to take advantage of the new syntactic simplicity which features a zero cost overhead.