1.4 Attrasoft Image Recognition Basics
The Attrasoft Image Recognition Approach has two basic phases:
-
Signature Matching (matching whole images)
-
Neural Matching (matching image segment)
In matching whole images, a unique image signature is computed for each image. Different images will have different signatures. The Signature Matching is based on these image signatures. This matching can easily reach the speed of 100,000 to 1,000,000 whole images per second and is very fast.
In the Image Segment matching, a neural net learns what the image looks like in a very similar way as the human brain does, i.e. adjusting the internal synaptic connections to remember the image. A typical Feature Space Recognition will use about 100 points and a typical Input Space Recognition will use about 10,000 points; therefore, the Signature Recognition is much faster than the Image Segment Recognition.
Image Matching is further divided into filters. It is these filters that will perform the image matching tasks.
The ability to learn is a fundamental trait of intelligence. Neural Nets are deployed in both Signature Matching and Segment Matching. Neural Nets can learn from examples. There are two basic phases in image recognition:
-
Training; and
-
Recognition.
In the training phase, data is imposed upon a neural network to force the network to remember the pattern of training data. A neural network can remember the training data pattern by adjusting its internal synaptic connections.
In the recognition phase, the neural network, based on its internal synaptic connections, will determine whether the newly captured image matches the sample image.
Copyright (c) 2006 - 2007 Attrasoft. All rights reserved.