Adaptive artificial retina model to improve perception quality of retina implant recipients

Abstract

Modeling of the retina in accordance with its physiological structure plays a key role in development of the high-performance image processing methods for sight restoration studies. In the literature, some Gaussian filter based artificial retinal processing models have been presented. In this study, as an alternative method to models which does not consider the characteristics of image content, adaptive DoG filter based artificial retina model that adaptively changes the bandwidths of the DoG filters according to local image data is developed. This model has two layers: the first layer models the interactions between receptors, horizontal and bipolar cells, the second layer models the interactions between bipolar, amacrine and ganglion cells of the retina layer. To evaluate the contribution of the model for image quality, some tests are performed by using video inputs. For this purpose, in these tests, spike counts based reconstructed images obtained by using this model, are analyzed. The retina model developed in this study which includes a 3-Dimensional Two-Stage Adaptive DOG (3D-ADOG) filter and the standard DOG filter based retina model are analyzed by using the statistical parameters of Mean Gray Level, Universal Quality Index and Histogram Similarity Ratio. In addition, the Histogram Similarity Ratio versus time is obtained and shown graphically for each model. From these performance analyses; it is concluded that 3D-ADOG filter-based retina model is closer to the original image in comparison with well-known classical DoG filter-based retina model and retinal implant systems based on this model will provide a better visual perception.