Content

1. Introduction
2. Video Capture
3. Segmentation
4. Classification

Classification

Concept of Classification

Building up a model

• Training
  • Use to build up the classification model
• Testing
  • Use to evaluate the accuracy of the classification model
• Recognition
  • Input an unknown pattern into the model, an output a known pattern.

Artificial Neural Network

• Also called ANN
• It is a computation model of a human brain.
• Human brain composed of a network of biological neurons
• Similarly, ANN composed of a network of artificial neurons

Artificial VS Biological Neuron

A network of Artificial Neurons to form a network

• A neural network is an interconnected group of nodes, akin to the vast network of neurons in thehuman brain.

Training Neural Network

• An artificial neural network can learn from a set of samples.
• Training is the process of adjusting the connection weights and bias of all neurons in the network, to enable the network to produce expected output for all input sets.
• For training a neural network,
  1. First you provide a set of inputs and outputs.
  2. Then train the network with a number of samples.
  3. Obtain the output and tell the network of the worng output.
  4. After training the network a number of times with a number of samples, it is assumed that the neural network can 'know' what is the output what we want
• Application:
  • Neural Network Configuration
    • Single layer (Input layer + output layer)
    • The number of input neurons map to the number of pixel in the image)
    • The number of output neurons map to the number of training dataset.
  • Training dataset
    • Static hand sign database found in the website of IDIAP Research Institute from Swaziland
    • We use 24 people white background hand signs
    • Converted into 32x32 gray-scale bitmap format
    • 10 letter: A, B, C, D, G, H, I, L, V and Y
    • 8 people hand signs are reserved for testing
    • Separate 4 batches of different size for training: i.e. group of 4, 8, 12, 16
  • Sample of Hand Sign Image:
    • a b c d
• Analysis: Training Data Size VS Accuracy
  • As the number of training data size increase, the accuracy increase.
  • Increasing training data size does not grantee to increase accuracy on all hand sign image.
  • The accuracy seems to increase up at a certain level. (In this case, around 69% - 70%)
  • As the training data size increases, the accuracy distribution becomes more even.
  • 
• Difficulty and possible improvement
  • Exponential training time
    • Optimization needed Increase the rate of learning
  • Accuracy
    • Increase training data size
    • Complex neural network, i.e. try adding more layer
  • Similarity among different hand signs images
    • Possibly eliminated during image segmentation