We present a novel approach to localization of objects in clutter images with the use of linear adaptive filters in a two-object classifier: target object versus clutter object. An automatic opti-mized feature extraction processing is suggested to generate two pair of models: “target” and “clutter” models from training image databases, and “clut-ter-like-target” and “target-like-clutter” models from positive and negative detection errors exam-ples respectively.
Experimental results obtained on testing database of known acquisition system containing “face” and “non-face” objects show that the proposed ap-proach outperforms other literature reported re-sults both in term of detection rate and false alarm rate.
