Feature extraction using optical multiple-valued logic
Akihito Nagatsu, Hideo Suzuki · Optical Society of America Annual Meeting · 1992
A new method of optical feature extraction is proposed, in which inter-pixel features of input images are spatially encoded and processed by optical multiple-valued logic. In this method, a conventional input image including only tone-level of each pixel is mapped onto the local feature plane. That is, each pixel is position coded according to the directions of the connections among eight nearest-neighbor pixels. These directional local features can be obtained by optical multichannel pattern-matching system. Next, several types of feature points, such as solitary points, end-points, and crossing points of three or four lines, are recognized. Those feature points will become especially important for recognizing the patterns macroscopically. Feature points extraction can be performed by multi-level thresholding. Also, types of connections between the feature points (horizontal, vertical, or inclined lines in some directions) are important and can be extracted by using optical directional pattern logic. These structured features can be used for upper-level image processing and recognition.