A Set of Criteria for Face Detection Preprocessing

The goal of this paper is to provide a robust set of preprocessing steps to be used with any face detection system. Usually, the purpose of using preprocessing steps in face detection system is to speed up the detection process and reducing false positives. A preprocessing step should reject an acceptable amount of non-face windows. First proposed criterion is based on linear image transform (LIT) which ignores scanning a number of non-face windows. Second criterion utilizes regional minima (RM) to reject non-face windows. The last one uses a modified adaptive thresholding (ADT) technique to convert input image into a binary representation and perform an exclusion process on the latter form. The proposed criteria have been used in conjunction with a version of Viola-Jones face detector. Experimental results show significant advantage against early exclusion criterion or variance classifier in terms of speed and rejection rate. CMU-MIT and BioID datasets have been used in the experiments