| Summary: | A large image usually consists of several smaller objects. People can recognize the objects automatically. The objects can be differented because they have different patterns. The aim of this research is for computer to recognize an object in image.
The objects which will be recognized are transformed to the frequency domain, so spectrum frequencies are obtained for patterns, and these spectra are used as input. The objects are sampled by 4x4, 8x8, and 16x16 pixels. The object recognition uses an Artificial Neural Network method with step function.
From this research, it is found that pattern recognition by spectrum frequency inputs is resistant to changes in position, such as rotation, translation, and reflection. Thus the same object is still recognized well, even though it has different position, or different place. Our experiment results show that pattern recognition in the frequency domain is more resistant to changes in position changing than pattern recognition in the spatial domain.
|
|---|