Applications

• Handwritten character recognition

• Object recognition

• Prototype formation

• Morphing

Prototype formation

• Average curve computed by averaging corresponding segments

• Not a trivial problem

• Useful in many applications like computational atlases, comparing anatomical structures, recognition

 
 
 

Morphing

• Can generate morph sequences for shape that are not very dissimilar

animation

animation

  animation

Handwritten character recognition

• Compute the prototype for each category

• Match each query to all prototypes and find the nearest neighbor

• For 327 characters we get 99% recognintion rate

Object recognition

• Curve matching framework works well in the presence of affine transformations, and visual transformations

• We get acceptable classification rates for a database of 36 shapes (6 categories, 6 shapes each)

• Five nearest neighbors are from "correct" category (36, 35, 35, 33, 27)

 
• We get acceptable classification rates for a database of 99 shapes (9 categories, 11 shapes each)

• Ten nearest neighbors are from "correct" category (99, 97, 97, 98, 94, 94, 90, 84, 74, 50)