OMR Error Patterns in the Van Eyck Collection

Aisha Ahmad-Post, 1 Aug. 2007; minor revisions by Donald Byrd, 8 Oct. 2007

MeTAMuSE/IU working document (ErrorPatterns_VanEyck_AAP.doc)

This document lists the most prominent errors made by several OMR programs in our Van Eyck collection of music page images, based on visual comparison of the OMR program output to the original pages. NB that this is all very simple monophonic music! Numbers in parentheses give the numbers of occurrences of each error. The program versions used were: PhotoScore version 4.2.1; SharpEye 2.67; SmartScore 5.3.

Craig Sapp’s very detailed analysis of OMR errors in much more complex music appears at ; however, it’s just one page of music (from a Mozart piano sonata) and one program (SharpEye). Still, it’d be revealing to compare SharpEye’s errors there to our Van Eyck results.

—DAB

PhotoScore errors

  1. PS fails to recognize flats in its interpretation. (14)
  2. Measure numbers are misinterpreted as rests, altering the rhythmic structure of the measure; any displaced notes as a result of this error are removed in the recognition process. (The number “16” is frequently interpreted as an eighth note rest, for instance.) (6)
  3. When the position of the beam switches in the middle of a scale, the notes adjacent at the switch are misinterpreted as a dyad (due to their closer proximity to each other than the other notes). (5)
  4. When an eighth note is beamed to two sixteenth notes, and the beam runs parallel to the staff lines, PS mistakes the eighth note for a sixteenth note. (3)
  5. The measure number “8” is recognized as a “B.” (3)
  6. Whole notes are misinterpreted as whole rests. (3)
  7. Flats are misidentified as sharps. (3)
  8. In dotted eighth-sixteenth note figures, the sixteenth note beam is misidentified as another notehead (usually, the rhythmic structure remains intact). (3)
  9. Unusually long-stemmed notes at the beginning or end of a measure (i.e., right next to the barline) generate errors involving repeat signs, fines, or double bars. (3)
  10. When the stems change direction mid-measure, PS recognizes it as two distinct voices (this is particularly true when this occurs on a measure that is marked with a measure number, for the reasons discussed above in #9). (2)
  11. A single whole note is interpreted as two whole notes, each with the same pitch value. (2)
  12. The time signature 3/2 is recognized as a common time signature and a half note. (2)
  13. Pickup measures are always filled in by rests to account for the “missing” beats. (1)
  14. When an eighth note is beamed to two sixteenth notes, and the beam intersects the staff line at an angle just after the first notehead, PS mistakes the eighth note for a sixteenth note. (1)
  15. In runs of sixteenth notes (ascending, descending, and mixed), notes with shorter stems are interpreted as cue notes (the resulting “missing” beats are filled with rests). (1)
  16. Depending on the placement of the beam relative to the staff lines, PS will occasionally recognize a line on the staff as a tenuto mark. (1)
  17. When the final measure does not have a full number of beats (due to a pickup measure at the beginning of the piece), PS will interpret it as a time signature change. (1)
  18. PS tends not to recognize finale signs. (1)
  19. Flats are misidentified as cue notes. (1)
  20. When a note has an accidental attached to it, the previous note is not identified. (1)
  21. Arpeggiated chords are misinterpreted as a single chord, played vertically. (1)
  22. When a leap of a 10th occurs between two eighth notes, the second eighth note is recognized as a quarter note. (1)
  23. Breves are misinterpreted as whole note rests. (1)
  24. The time signature 3/2 is recognized as a tetrachord. (1)
  25. The presence of a natural sign causes PS to ignore the preceding note. (1)

Three errors without rhyme or reason:

  1. Inclusion of staccato marks (probably from noise on the scan, but in many instances, no noise could be found).
  2. Slurs seem arbitrarily placed.
  3. Accent marks are also arbitrarily placed.

“12. Courant, of Ach treurt myn bedroefde” is a perfect PS example

SharpEye errors

  1. SE misses beams, creating longer lengths of the earlier notes that remove the later ones from the recognition scan. (15)
  2. When given a large number of notes of homogenous rhythmic value, SE misses notes in its recognition (although no clear pattern emerges here to cause this.) (13)
  3. When the position of the beam switches in the middle of a scale, SE recognizes it as two discrete voices. (When this occurs in conjunction with a missed beam, a “second voice” is also generated by the notes that have been pushed into the next measure.) (8)
  4. Breves are mistaken as whole rests. (2)
  5. A key signature with one flat (FM) is interpreted as an accidental (SE changes the key signature for that line to 0 flats, 0 sharps) (see 10 Modo2 for clarification). (2)
  6. A breve with a stem is interpreted as a blank measure, followed by a measure with a whole rest. (2)
  7. When the stem is very short, SE misses the note entirely. (1)
  8. SE misses the first note of a piece when it is in close proximity to the time signature. (1)
  9. When a note is misinterpreted as being of a longer value, the remaining notes in the measure are shifted into the next, creating two voices. (1)
  10. When a note has an unusually short stem, SE does not recognize it as a notes. (1)
  11. An “unclean” ledger line (i.e., one that is slightly enlarged on either side, or otherwise sloppily printed) is misinterpreted as a dot on the preceding note, thus creating a rhythmic error. (1)
  12. When a note is placed too closely to a measure number, a rhythmic error is generated. (1)
  13. The measure number “7” is interpreted as an eighth note rest. (1)

SharpEye had the greatest number of “perfect” recognitions – i.e., no mistakes in the output.

SmartScore errors

  1. Beams are unrecognized, creating rhythmic errors. (48)
  2. Sharps are not recognized. (28)
  3. Whole notes are misintepreted as whole note rests. (15)
  4. The presence of words in the upper left hand corner at the beginning of the piece creates errors in rhythmic detection (usually in the form of some obscure tuplet marking, such as “27” or “9”). (5)
  5. Sharps are misinterpreted as flats. (5)
  6. A stem that extends slightly beyond the beam is misinterpreted as a dyad (and also causes a rhythmic error). (4)
  7. Augmentation dots are lost. (4)
  8. When there one note in a group that has a stem that extends the opposite way from the other notes, that note does not appear in the recognition. (3)
  9. When a note is repeated, SS misses one of the notes. (3)
  10. SS adds augmentation dots. (3)
  11. A note positioned too closely to the beginning of a staff on the original scan is not recognized by SS, or is misinterpreted as a dyad. (2)
  12. When the beams of a sixteenth note lie below the staff (generating an unusually long stem), SS interprets it as a quarter note. (2)
  13. Flats are not recognized. (1)
  14. Notes with unusually short stems are not recognized. (1)
  15. When the second added beam for a sixteenth note appears to “sit” on a line in the staff, SS interprets it as an eighth note. (1)
  16. The presence of a sharp causes SS to not see the preceding note. (1)
  17. Natural signs are not recognized. (1)
  18. The time signature 3/2 causes SS to add an additional note at the beginning of the piece. (1)
  19. Natural signs are misinterpreted as flats. (1)
  20. Natural signs are misinterpreted as sharps. (1)

SS also frequently loses notes, but I can’t find a distinct pattern to this.

1