Phase congruency eigendecomposition for multi-scale neuronal enhancement

In this paper, we present an algorithm for enhancing neuronal structure from 3D Confocal Microscopy Images. Our algorithm computes a multi-scale phase congruency value at every pixel from a 3D image, which assigns values that indicate the presence of image features such as edges and lines. The phase congruency of a 3D image is calculated by carefully combining the convolutions of the image with a quadrature filter bank, so we leverage this information to supplement phase features. We analyse the outputs of the quadrature filter bank to enhance neuronal structure. We compare our method with the Hessian based enhancement of neuronal structure to demonstrate the advantages/efficacy of our algorithm.