| 摘要 |
Projections onto the horizontal axis of four spaces surrounded by staff positions L(5) of from L0 to L4 determined for each music sign, from the staff start position (xs) to the staff end position (xe) are determined. When the width of a section from (x1) to (x2) in which the projection value exceeds a given threshold value thr1 is greater than a given threshold value thr2, a rectangle surrounded by x1, x2, L0 and L4 is used as a label for extracting the feature of the music sign. The above-mentioned four spaces in the label that do not overlap with the staff image are used as extraction sections for extracting peripheral features in the transversal direction, and peripheral features from the right and left ends in the extraction sections are determined up to the third order, respectively. Each extraction section is so divided into meshes that they overlap with each other in the longitudinal direction and an average value of the peripheral features for each order in the meshes is determined and used as a feature amount of each mesh. The feature amounts found from both the right and left ends for each mesh are used as the features of the label, and music signs having fixed positions on the staff or having the staff in a grid in the score image is recognized by the pattern matching with the features stored in a dictionary. Thus, music signs are recognized from the score image, and data are generated for playing and for score displaying. |
