RECOGNITION ENGINE DESIGN

 

Logical Designs can help your company perform classification tasks with higher accuracy then can be achieved by traditional methods. Whether your project is OCR related or not, Logical designs can built the classifier you need.

Logical Designs has worked on a number of classification projects including loan application scoring and assembly line inspection. We have built OCR recognition engines for hand printed numbers, hand printed uppercase alphas, and multi font machine print character sets.

Tasks involved in a typical recognition engine design project are straightforward. We start by analyzing the requirements for the classifier in terms of operating environment, speed of operation, required performance, input information and availability of training data.

 With this information, a determination is made as to the architecture of the network to be used. Network choices can include Backpropagation, Learning Vector Quantization, MADALINE, Probabilistic Neural Network. Logical Designs also has developed several proprietary networks that can satisfy your special requirements for high speed and integer operation.

To control the size and training requirements of the chosen network, a reduced input representation must be found. Most classification and recognition tasks have a great deal of information available for use by the network. A typical page of text scanned at 300 dpi can contain 8 million pixels. A typical image from a video camera can have 65 thousand pixels, each with 256 levels of gray. Preprocessing of this information to reduce the dimensionality is a typical task in the design of a classifier. Methods used at this stage of development tend to be traditional transform techniques and heuristics.

The generation of sufficient training and test data is of key importance to a network classification application. The chosen network is trained on the available data leaving some portion of the data aside for testing purposes. At the end of training, statistics related to network performance on test data are examined to determine if the network has achieved the desired performance in terms of accuracy of recognition and substitution performance.

Logical Designs can handle all aspects of the design of your recognition engine. We can supply custom programming for preprocessing of data. Logical Designs can build and verify a training set from your data. We have the equipment to train and test the classifier and can also integrate the recognition engine into your application.

Logical Designs has developed neural algorithms not available from any other source. We can supply a custom software and hardware solution to your problem, or work with the tools you have in house. Logical Designs has the ability and experience to make your classification project a success.