1 / 16

A PREPROCESSING METHOD AND ROTATION INVARIANT 2D OBJECT RECOGNITION USING BPG NEURAL NETWORKS

A PREPROCESSING METHOD AND ROTATION INVARIANT 2D OBJECT RECOGNITION USING BPG NEURAL NETWORKS. Irina Topalova . Preprocessing. Backpropagation NN. Class. Image. Introduction to NN processing. Quality. Complex Simple. Simple Complex. Accuracy. The Problem.

garvey
Télécharger la présentation

A PREPROCESSING METHOD AND ROTATION INVARIANT 2D OBJECT RECOGNITION USING BPG NEURAL NETWORKS

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A PREPROCESSING METHOD AND ROTATION INVARIANT 2D OBJECT RECOGNITION USING BPG NEURAL NETWORKS Irina Topalova

  2. Preprocessing Backpropagation NN Class Image Introduction to NN processing Quality Complex Simple Simple Complex • Accuracy

  3. The Problem • Image – Low quality web camera • Preprocessing - ? • Backpropagation NN - ? • Class – High accuracy Oblong Objects Class 1 - Hammer Class 2 - Spanner

  4. The Preprocessing Step 1: Color to grey-level conversion: For each image pixel calculate: . Hammer - color Hammer – grey-level

  5. Hammer – Sobel The Preprocessing Step 2: Sobel contour: • Utilization of the first gradient of the image function • Small amount of noise • Thick edges Hammer – grey-level

  6. The Preprocessing Step 2: Sobel contour: Image function V Sobel mask MxSobel mask My

  7. Hammer – Sobel Hammer – vectorized The Preprocessing Step 3: Contour vectorization: • Outer contour tracing • Weighted chain-code with backtracking • Edge points ordering – ordered list of coordinates

  8. Hammer – rotated Hammer – vectorized The Preprocessing Step 4: Contour rotation: • NN facilitation – especially effective for oblong objects • One large, loose cloud  several small, tight clouds in the parametrical space

  9. The Preprocessing Step 4: Contour rotation: For each calculate: for all n contour points and form the following metric: . Find and rotate the image contour by the angle φ.

  10. Hammer – rotated The Preprocessing Step 5: Radial profile function: • Numerical function passed to the BPG NN • Contour resampling – only N of n edge points • Further enhancement of the rotation invariance Hammer – radial profiles

  11. The Preprocessing Step 5: Radial profile function: Calculate the contour gravity center : . Form the radial profile function: and pass it to the NN.

  12. The BPG Neural Network • Good accuracy after training • Easy supervision of the training process The NeuFrame BPG NN

  13. The BPG Neural Network • 2x24 training images; 2x10 query images • 30 input and 2 output sigmoid neurons The NN Topology

  14. Results • Training error: 0,005 successfully reached • Well-formed error graph • Query accuracy: 20/20 - 100% The NN error graph

  15. Conclusions • The preprocessing stage delivers consistent input data to the NN thus facilitating its training and making the identification of the input descriptors of overlapping classes much easier. • The preprocessing stage is fast enough to be implemented in real time working systems. • Further research on noisy 2D objects could be carried out .

More Related