Feature Extraction System for Contextual Classification within Security Imaging Applications

摘要

Throughout security imaging applications, there is a persistent need for accurate contextual classification of objects within the scene so proper subsequent decisions can be made. To generate a set of scene attributes necessary for this analysis, this paper presents a novel feature extraction system composed of three divisions: an edge detection system, a segmentation system, and a recognition system. System inputs are considered to be low resolution, low quality images, often collected from inexpensive security imaging cameras. This work concentrates on enhancing the accuracy of the detected boundaries and edge pixel locations within the edge detection system as a pre-processing step for the segmentation and recognition systems. The edge detection described here is based on Boolean derivatives, calculated using partial derivatives of Boolean functions in combination with fusion and binarization steps. This edge detection system allows overall subsequent improvements in the segmentation and recognition systems, producing a stronger overall feature extraction system for processing data within security imaging applications.