Robust specular reflection removal and visibility enhancement of endoscopic images using 3-channel thresholding technique and image inpainting

Abstract

Specular reflections create artifacts in endoscopic images, which may lead to misdiagnosis. In this paper, we propose a method for robust removal of specular reflections by using a thresholding technique in each of the RGB channels to segment the specular reflections from images. We further use dilation to ensure full local segmentation and inpainting to replace the areas of reflections with non-specular regions. Our method also provides a visibility enhancement feature to improve the decreased brightness due to the reflection removal by using the gamma-correction, histogram shift, and histogram equalization. On the Iparkmall Clinic dataset, our method has achieved average Peak-to-Signal-Noise Ratio (PSNR) of 42.62dB with a standard deviation of 5.80 dB and a minimum value of 23.3 dB. The average processing time was 219ms, enabling average 4 5 frames per second (FPS) processing speed on an Intel i7 processor.