A Blob And Rectangular Form Filter Based On Mathematical Morphology

Both blob form and rectangular form are normal forms which are widely used in computer vision filed,such as detection and tracking of pedestrian in video surveillance,human eyes detection in face recognition and extraction of lesions in medical images.How to efficiently detect these two forms in gray-scale image is an interesting and useftul subject.In this paper,we developed four filters based on mathematical morphology.Three of them are used to detect blob form,the other one is used to detect rectangular form:(1)BFF_ord(Blob Form Filter based on order filtering)is based on order filtering and effectively improve the anti-noise ability,thereby enhance the filter's robustness.(2)BFF_ave(Blob Form Filter based on the average order-value filtering)is based on the average order-value filtering which is more robust than BFF_ord,but its detection accuracy is worse than BFF ord.(3)BFF_max(Blob Form Filter based on the maximum value of the order filtering results)is a variation of BFF_ord.It not only maintains the robustness of BFF_ord but also greatly improves the detection accuracy.(4)RFF_max(Rectangular Form Filter based on the maximum value of the order filtering results)is not only robust to noise but also good at detection accuracy.As a specific application of blob form filter,we proposed a new eye localization algorithm based on a bank of blob filters.We designed a set of experiments to prove that the algorithm can efficiently and accurately locate the eyes in face images.The experiment results show that it gains the high localization accuracy in CAS-PEAL-R1 gray-scale face database(8559),Jaffe gray-scale face database(213),Caltech color face database(450)and BioID gray-scale face database(1521).