Neuron Image Synthesizer Via Gaussian Mixture Model and Perlin Noise

Microscopy imaging provides valuable information to neuroscientists. There are many different algorithms to analyze these images but due to an absence of ground truths, it is difficult to quantitatively compare the algorithm performance. Realistic synthetic images with known ground truths offer a means of assessing their performance. This approach has been successfully applied in several areas such as segmentation and classification. In this paper, we propose a synthesizer for neuron nucleus micro-environment using Gaussian mixture model (GMM) and Perlin noise function. Nucleus shapes are generated by spline interpolation of random points on elliptical shapes. The textures of the foreground (nuclei) and the background are generated via Perlin noise, and then assigned intensities generated by applying GMM to the real data. The cell orientations are also implemented via Perlin noise to mimic the behavior of the actual neuron cells.