Automatic Detection and Counting of Defects from Cathodoluminescence Maps of GaN Layers

We propose and implement a dedicated procedure for automatic detection and counting of defects from panchromatic cathodoluminescence maps of GaN layers. The procedure consists of low-pass 2D FFT filtering in frequency domain for noise removal, application of the CLAHE algorithm to enhance the contrast of less pronounced defects, global thresholding that implements combined mean and Otsu's method, image segmentation by cyclic application of dilation and the watershed algorithm followed by final marking of the position of defects and their counting. The developed procedure can be used to process a batch of multiple CL maps, dramatically reducing the amount of time required to assess the quality of investigated GaN layers.