the dynamic image segmentation for sign language training
play

The Dynamic Image Segmentation for Sign Language Training Simulator - PowerPoint PPT Presentation

The Dynamic Image Segmentation for Sign Language Training Simulator Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 1 Gesture (a) Start of the gesture (b) Intermediate frame (c) Intermediate frame (d)


  1. The Dynamic Image Segmentation for Sign Language Training Simulator Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 1

  2. Gesture (a) Start of the gesture (b) Intermediate frame (c) Intermediate frame (d) Final frame Figure 1: Gesture “What for” Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 2

  3. Dactyl (a) Starting dactyl (b) Final dactyl Figure 2: Dactyls describing gesture “What for” Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 3

  4. Dactyl Matching Figure 3: Recognition of Fingertips Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 4

  5. Colour Space in Image Clustering with SOM (a) Original (b) RGB (c) HSL (d) CIELab Figure 4: Clustering of an image represented in different colour spaces Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 5

  6. Training Data Composition   ( L 1 1 , a 1 1 , b 1 1 ) T ( L 1 2 , a 1 2 , b 1 2 ) T ( L 1 3 , a 1 3 , b 1 3 ) T ( L 1 4 , a 1 4 , b 1 4 ) T     ( L 2 1 , a 2 1 , b 2 1 ) T ( L 2 2 , a 2 2 , b 2 2 ) T ( L 2 3 , a 2 3 , b 2 3 ) T ( L 2 4 , a 2 4 , b 2 4 ) T   A =     ( L 3 1 , a 3 1 , b 3 1 ) T ( L 3 2 , a 3 2 , b 3 2 ) T ( L 3 3 , a 3 3 , b 3 3 ) T ( L 3 4 , a 3 4 , b 3 4 ) T       ( L 4 1 , a 4 1 , b 4 1 ) T ( L 4 2 , a 4 2 , b 4 2 ) T ( L 4 3 , a 4 3 , b 4 3 ) T ( L 4 4 , a 4 4 , b 4 4 ) T   ( L 1 1 , a 1 1 , b 1 1 ) T ( L 1 2 , a 1 2 , b 1 2 ) T S 1 =     1 ) T 2 ) T ( L 2 1 , a 2 1 , b 2 ( L 2 2 , a 2 2 , b 2 Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 6

  7. Cluster Interpretation • Image pixels represented by topologically close neurons should be- long to the same cluster and therefore segment. • The colour or marker used for a segment representation is irrelevant as long as each segment is associated with a different one. R j ← x j + y j × λ ; G j ← x j + y j × λ ; B j ← x j + y j × λ ; (1) Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 7

  8. Application Results (a) Frame 25 original (b) Frame 25 segmented (c) Frame 35 original (d) Frame 35 segmented Figure 5: Frames 25 and 35 Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 8

  9. Application Results (a) Frame 45 original (b) Frame 45 segmented (c) Frame 65 original (d) Frame 65 segmented Figure 6: Frames 45 and 65 Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 9

  10. Other Applications (a) Original (b) Segmented (c) Original (d) Segmented Figure 7: Cloth Segmentation for Image Search Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 10

  11. Alternative Approach (a) Sample 1 (b) Sample 2 Figure 8: Cloth Segmentation for Image Search Oles Hodych, Kostiantyn Hushchyn, Iouri Nikolski, Volodymyr Pasichnyk, Yuri Shcherbyna 11

Recommend


More recommend