Efficient fingerprint features for gender recognition.

After a crime scene, accurate gender recognition by fingerprint analysis is vital for detectives because precise gender recognition highly limits the search space. For extracting high quality features from fingerprint images, each image should be preprocessed. The preprocessing stages include segmentation, normalization, filtering, binarization, and thinning. Next, different features from various domains are elicited from each image. The suggested features are ridge count, minutiae points, discrete cosine transform, entropy, local binary pattern and ridge thickness valley thickness ratio features. Each feature and the combination of features for one and five fingers are separately applied to six efficient classifiers for gender recognition. The best result implies 99% accuracy with the ridge count for all five fingers. A combination of features for each finger in the best case provides 91% gender recognition accuracy. The combination of our candidate features for each finger is compared to singular value decomposition (SVD), discrete wavelet transform (DWT), and the combination of SVD and DWT. Our results statistically (p value < 0.05) outperform the compared methods. [ABSTRACT FROM AUTHOR]