Retinal OCT Image Denoising Based on Adaptive Bessel K-form Modeling

In this study, an adaptive approach is addressed to reduce the noise of retinal Optical Coherence Tomography (OCT) images. Since the layered structure of retinal OCT images creates a dependency between adjacent pixels at particular distances, the presented method is based on the adaptive selection of variable neighborhood windows for each pixel of OCT images. Indeed, by defining this spatial adaptivity, we extend our earlier work in which a pixel-wise fixed window was considered. Here, the variance is calculated in an optimal window for each pixel; so that the ultimate distribution of the variance image follows a gamma model. Besides, the asymmetry observed in the distribution of retinal layers led to suggest Asymmetric Bessel K-form (ABKF). This model is easily transformed into a Gaussian distribution through dividing the image into the root of the variance image. Then, it can be used with Gaussian-based algorithms for OCT denoising application. The results show the impressive performance of the proposed adaptive local BKF model in noise reduction and increasing image contrast as visual and quantitative criteria.