Observation model based retinal fundus image normalization and enhancement

摘要

Based on an observation model, a method to normalize non-uniformly luminosity and contrast present in retinal funds images is proposed. By sub-sampling and interpolation techniques, the background of the retinal images can be estimated. On this basis, the details of the image will be enhanced by singular value processing and homomorphic filtering techniques. Experiment results show that this method can effectively achieve non-uniform illumination and contrast normalization and image contrast enhancement, which meet the requirements of retinal image preprocessing. Moreover, subjective visual observation and objective evaluation test show that this method is significantly better than many other similar methods.