Wavelet weighted distortion measure for retinal images

Abstract—Abstract TThe diagnostic features of retinal images undergo
changes in the course of processing such as for storage,retrieval and transmission. The conventional mean square error and peak signal to noise ratio have limitations in quantifying these local distortions. In this work, a novel wavelet weighted distortion measure (WWDM) is proposed for accurate quantification of diagnostic information loss. The wavelet analysis of retinal image shows that the significant information of a retinal feature is captured by a fewsubbands.The new approach is based on assigning a weight to each of the subbands depending on its diagnostic significance. The proposed distortion measure is defined as the sum of wavelet weighted root of the normalized mean square error of subbands expressed in percentage. The experimental results show that WWDM performs better in capturing the distortion in retinal features, whereas for distortion in clinically nonsignificant regions, it gives a low value. The qualitative evaluation using Pearson linear correlation coefficient and Spearman rank order correlation coefficient is performed for different artifacts. The investigation shows better correlation values between WWDM and the subjective scores.