#179: Float precision issues with epsilon ---------------------+------------------------------------------------------ Reporter: lazzara | Owner: Olena Team Type: defect | Status: new Priority: major | Milestone: Olena 1.0 Component: Milena | Version: 1.0 Keywords: | ---------------------+------------------------------------------------------ In level::stretch, we use float variables to do the following: {{{ float epsilon = mln_epsilon(float), M = mln_max(V) + 0.5f - epsilon, m = 0.0f - 0.5f + epsilon; }}} 'm' is precise enough since it is a really low value. However, M is not precise enough since mln_max(V) may be a high value, so we will loose decimal precision if we store that number as float. As a result, M is always equal to 'mln_max(V) + 0.5' which introduce a bug in level::stretch. This issue can be avoided by using double variables instead, but do we really want that? Don't we want to handle that issue in a more generic way? -- Ticket URL: <https://trac.lrde.org/olena/ticket/179> Olena <http://olena.lrde.epita.fr> Olena, a generic and efficient C++ image processing library.