STEM_SEEDS step

A stem is a rather long vertical segment meant to join note heads, perhaps with beams. And in a perfect world, any stem could be easily recognized on its own.

But in reality, there may be missing black pixels in the input image, perhaps breaking a stem in smaller segments, or damaging the connection between a stem and some of its partnering heads and beams.

To cope with this, the engine uses several approaches to gradually detect:

  1. (portions of) stem candidates – this STEM_SEEDS step!
  2. beam candidates – the BEAMS step
  3. ledger candidates – the LEDGERS step
  4. heads candidates – the HEADS step

and finally to:

  1. determine all full stem candidates joining heads and beams – the STEMS step
  2. filter all the candidates – the REDUCTION step
  1. Input
  2. Output
  3. Prolog: stem thickness
  4. System processing
    1. Vertical filaments
    2. Checking seeds

Input

  • The systems and staves
  • The no-staff image
  • The vertical LAG
  • The horizontal LAG (very partially)

Output

  • The typical stem thickness for the sheet
  • The stem seeds as glyphs in the VERTICAL_SEED group

Prolog: stem thickness

The user can directly specify a stem thickness value for the sheet at hand via the Sheet → Set scaling data pull-down menu.

In the normal case – that is with no thickness value provided – the engine must, as a step prolog, measure the typical stem thickness. It does so by crafting an input image suitable for a global stem analysis.

To avoid any pollution brought by lyrics, title or other external Main concepts, the engine focuses only on the core staff areas, which is the staff height augmented vertically slightly less than one interline above and one interline below.

Within each staff area, the staff lines have been removed, as well as the detected barlines and the header components.
Note that some stems may have been mistaken for barlines candidates during the GRID step; this should have a negligible impact on the stem thickness computation.

The result is an input image like this one (still for the chula example):

chula stems buffer

On this input, it is easy to build a histogram of horizontal black runs lengths, as shown below.

We can display this histogram via the Sheet → Display stem plot pull-down menu – provided we have activated the PLOTS topic in the menu.

chula stems histo

The retrieved peak gives the stem thickness: in our example, the most frequent value is 2 pixels, and the maximum value is 3 pixels.

These values are stored in the sheet scaling data. Then the rest of step processing is performed system per system.

System processing

Vertical filaments

Based on the retrieved (or specified) stem thickness value, the engine can now build vertical straight filaments from the vertical sections that are slim enough and long enough.

For now, the engine is only looking for stem seeds. Hence, it makes no attempt to extend these filaments vertically – this will be done later in the STEM step.

The filaments can just be thickened with compatible adjacent vertical sections and tiny horizontal sections as long as the resulting thickness stays within the limit on stem thickness value.

Checking seeds

Each retrieved filament is further checked to be kept as a stem seed.

The checks concern:

  • slope
  • straightness
  • length
  • cleanliness – not too many adjacent pixels
  • high black part
  • low white part

The checks are evaluated according to the declared quality of the sheet input (synthetic, standard, poor).

The successful candidates are finally registered in the specific VERTICAL_SEED group of glyphs, ready to be picked up by the following steps.