Bengali Sign Digit recognition is a competitive task as there are no robust dataset. Though, here i have used Ishara_Lipi Dataset for my works. It has consists of total 1000 images and almost 100 images per class(10).

To successfully classify the multi_class sign digit classfication problem, i have used transfer learning technique to train my CNN model and found satisfied training and testing accuracy.

Some sample of images of Ishara_Lipi Sign digit dataset-

0(Zero_Digit)

1(One_digit)

2(Two_digit)

3(Three_digit)

4(Four_digit)

5(Five_digit)

6(Six_digit)

7(Seven_digit)

8(Eight_digit)

9(Nine_digit)