Feature extraction using vague semantics approach to pattern recognition

Feature extraction is essential to pattern recognition. From low-level image processing, one has to link up workable pixels into clusters of interest as pattern's features. Nowadays, renowned recognition designs also require additional processes to transform pixel clusters further into image matrices or histograms. Features' similarity between detected patterns and pre-defined models is then surveyed by methodologies of probability and statistics. For designing a humanoid recognition system, we originally develop a promising feature extraction scheme called semantic-based vague image representation (SVIR) for pattern recognition, where feature classification using a series of semantics substitutes for pixel clusters. Refined algorithms with low computing load are set as the guideline of designs. In this paper, we provide specific bipolar encoding for pattern sampling and propose various feature operations for 2D binary image recognition but with unsophisticated computation. The methodology presented in this paper serves as a promising tool for answering the prospect of ambiguous classification in computer vision.