A no reference objective color image sharpness metric

In this work, we propose a no reference color image quality assessment metric. The proposed metric makes use of a wavelet-based multiscale structure tensor [1] as an extension of the single-scale structure tensor proposed by Di Zenzo [15]. The multiscale structure tensor allows for accumulating multiscale gradient information of local regions of the color image. Thus, averaging properties are maintained while preserving edge structure. This structure tensor is capable of identifying edges in spite of the presence of noise. Once edges are identified, we define a sharpness metric based on the eigenvalues of the multiscale structure tensor. Particularly, we show that the difference of the eigenvalues of the multiscale structure tensor can be used to measure the sharpness of color edges. Based on this fact we formulate our no reference sharpness metric for color images. Experiments performed on LIVE database indicate that the objective scores obtained by the proposed metric agree well with the subjective assessment scores.