Image sharpness measure using eigenvalues

This paper proposes a novel statistical approach to formulate image sharpness metric using eigenvalues. Statistical information of image content is represented effectively using a set of eigenvalues which is computed using singular value decomposition (SVD). The approach is started by normalizing the test image with its energy to minimize the effects of image contrast. Covariance matrix which is computed from the normalized image is then diagonalized using SVD to obtain its eigenvalues. Sharpness score of the test image is determined by taking the trace of the first six largest eigenvalues. The performance of the proposed approach is gauged by comparing it with orthogonal moments-based sharpness metrics. Experimental results show the advantages of the proposed approach in terms of providing wider working range and more precise prediction consistency in noisy condition.