American Sign Language

1
American Sign Language
2
(No Transcript)
3
Hausdorff Distance
4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11

• Multivariate normal density function was used for
  skin detection.
• Used HW 1 data for training
• Used median filtering and connected component
  labeling to fill the gaps and remove noise.

12
Analysis

• Used Histogram technique to do matching based on
  number of skin pixels in each row and column of
  the test image and the model image.
• This method worked well in the cases where the
  fingers were not bent or hidden.
• The characters that were recognized successfully
  are A, B, D, F, I, L, R, U, V, W, Y, which gave a
  recognition rate of 40 for all the datasets that
  I tested.

13

• For the edge detection technique
• Recognition Rate27 Characters Correctly
  Identified C, K, L, Q, O, R, V, W, Y
• Combining the two approaches together I was able
  to recognize 14 of the 26 alphabets which make
  the recognition rate of the system to be 53.84
  percent.

14

• Edge detection approach will work well as long as
  the test image is similar to the ensemble of
  images used for training purpose.
• The training set should include multiple images
  of the same hand image with some variations to
  obtain a better performance. Here only one set of
  image was for training and the others for testing.