A novel feature extraction method based on local textures of image for real-time vehicle recognition tasks is proposed. In particular, feature vectors are extracted from judicious combinations of partitioning and overlapping image blocks with a view to reducing cardinality while retaining sufficient feature information for effective recognition. Experimental results are presented that show a reduction in computational load of 80% and robustness to illumination and noise can be achieved.
ASJC Scopus subject areas
- Electrical and Electronic Engineering