Reducing spectral reflections through image inpainting

Abstract

The goal of this thesis is to produce an image processing tool that could, with limited user interaction, significantly reduce spectral reflections. The resulting process can then be used by biological researchers in the process of cataloging and recognizing individual organisms via collected images. The process involves the generation of a mask, refinements to that mask such as background segmentation and morphological dilation, and finally the inpainting of the specular regions. This method improves on existing single pixel methods by incorporating both color and texture infilling of the specular reflections. It also works on images with no special setup or preparation.