What are nonrepresentational components of an image?

What should be the approach in making a device which identifies components of an image?

• I want to make a simple device which has a camera which captures an image and a processor that reads and identifies content in that image which we can predefine. For example, the device should identify the colour of a traffic signal in the captured image. How should i go about making this device? I am new to DSP/DIP

• Answer:

  Do some research on bio-metrics, you can look at Eigen Faces, it sounds like you would be looking for image that maps to a pre-defined histogram of a traffic signal. You will have to account for all angle variations up to 10% and scaling. You then compare the histogram of your captured pick to that of the library, based on key marker points defined in the histogram, you can determine probability of a match and the detect color in the key area you are sampling from. Hope this helps?