Fingerprint Image Denoising in Spatial Domain: An Implementation based on a Combined Median and Average Filtering Approach

Abstract

In this article, image denoising approach using a combined median and average filtering scheme has been proposed. The proposed scheme filters the noisy image in the spatial domain without subjecting it to any transformations. Fingerprint images are more often characterized by ridges and valleys. These ridges and valleys need to be considered as linear and relatively brighter for classification and recognition in many biometric identification systems. Noise accumulates with the input image during the acquisition process, results with the relative change in the ridge structure. This intern affects the classification and recognition processes of an authentication system. Denoising process is more likely suitable only with the proper attainment of ridges and valleys characteristics in the resultant image. About eighty standard fingerprint test images are considered from FVC2004 dataset. Four different standard noises are considered in the intrusion process. Five different schemes are considered to analyse the performance of the proposed scheme along with conventional spatial and frequency domain techniques. An average 96% correlation between host and reconstructed images is observed along with a maximum of 8% average difference between the respective structural contents under different noises. Very minimum absolute errors during structural reconstruction process is also observed, compared to other spatial and frequency (Median & Wiener) domain approaches.