Abstract:

Previous matrix factorization algorithms can not guarantee the nonnegativity of results. Inspired by the nonnegative character of NMF (Non-negative Matrix Factorization), NMF is utilized to detect shadow areas. The BNMF (Block Nonnegative Matrix Factorization) is introduced for shadow detection. The brightness features of shadow points of training samples are extracted by NMF or BNMF method. The shadow areas of testing sample are recognized by features. The results demonstrate that the approach preserves edges clearer than the method based on singular value decomposition.

Key words: shadow detection, non-negative matrix factorization (NMF), block non-negative matrix factorization (BNMF)