Symbol recognition via statistical integration of pixel-level constraint histograms: a new descriptor
Yang Yong Su · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2005
A new descriptor for symbol recognition is proposed. 1) A histogram is constructed for every pixel to figure out the distribution of the constraints among the other pixels. 2) All the histograms are statistically integrated to form a feature vector with fixed dimension. The robustness and invariance were experimentally confirmed.