In the context of eye-related diseases such as diabetic retinopathy (DR), retinal image quality assessment is used for evaluating the quality of an image based on its usefulness in detecting a certain condition or disease. Since poor quality retinal images make the detection process more difficult, it is necessary to assess the quality of retinal images before disease detection. Retinal image quality grading evaluates if the quality of an image is sufficient to allow diagnosis procedure to be applied. Automation of this process would help reduce the cost associated with trained graders and remove the issue of inconsistency introduced by manual grading. In this paper, we present a new method for automatic assessment of retinal image quality. The proposed method is based on shearlet transform which is a new multi-scale and time-frequency image analysis method. In addition to multi-resolution and time-frequency localization provided by traditional wavelet transform, the shearlet transform also provides directionality and anisotropy. We use the statistical features of shearlet coefficients to assess the quality of retinal images. Using SVM classifier, the performance of the proposed method was evaluated on two datasets. Experimental results demonstrate an excellent performance in comparison with other methods reported recently.

Keywords